Financial Markets Theory-Barucci

Springer Finance
Textbooks
Emilio Barucci
Claudio Fontana
Financial
Markets Theory
Equilibrium, Efficiency and Information
Second Edition
Springer Finance
Textbooks
Editorial Board
Marco Avellaneda
Giovanni Barone-Adesi
Mark Broadie
Mark Davis
Emanuel Derman
Claudia KlRuppelberg
Walter Schachermayer
Springer Finance Textbooks
Springer Finance is a programme of books addressing students, academics and
practitioners working on increasingly technical approaches to the analysis of
financial markets. It aims to cover a variety of topics, not only mathematical
finance but foreign exchanges, term structure, risk management, portfolio theory,
equity derivatives, and financial economics.
This subseries of Springer Finance consists of graduate textbooks.
More information about this series at http://www.springer.com/series/3674

Emilio Barucci • Claudio Fontana
Financial Markets Theory

Equilibrium, Efficiency and Information
Second Edition
123
Emilio Barucci Claudio Fontana
Dipartimento di Matematica Laboratoire de Probabilités et Modèles
Politecnico di Milano Aléatoires
Milano, Italy Université Paris Diderot (Paris 7)
Paris, France
ISSN 1616-0533 ISSN 2195-0687 (electronic)

Springer Finance
ISBN 978-1-4471-7321-2 ISBN 978-1-4471-7322-9 (eBook)
DOI 10.1007/978-1-4471-7322-9
Library of Congress Control Number: 2017940989
Mathematics Subject Classification (2010): 91B06, 91B08, 91B16, 91B24, 91B25, 91B30, 91B50,
91G10, 91G20
© Springer-Verlag London Ltd. 2003, 2017

The author(s) has/have asserted their right(s) to be identified as the author(s) of this work in accordance
with the Copyright, Designs and Patents Act 1988.
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer-Verlag London Ltd.
The registered company address is: 236 Gray’s Inn Road, London WC1X 8HB, United Kingdom
To Piero, Maria, Elena, Teresa and Maria
E.B.
To Anna, Angelo and Liliana
C.F.
Preface to the Second Edition (2017)
Already at the time of the first edition of this book, financial markets theory
represented, and still represents nowadays, an extremely rich field of research,
characterized by a huge and growing literature. Since the publication of the first
edition, the literature has grown even more significantly, to the extent that it is
simply impossible to give a truly comprehensive account of financial markets
theory. In this second edition, we aim at providing a broad overview of the
fundamental aspects of the theory, balancing economic intuition and mathematical
detail. Moreover, following the philosophy of the first edition, the presentation of
theoretical results is always accompanied by a discussion of the empirical evidence
reported in the financial literature.
In comparison with the first edition, the book has been thoroughly revised and
significantly expanded. While the overall structure of the book is similar to that
of the first edition, most of the chapters have been entirely rewritten. In particular,
besides presenting a large amount of additional material, we aim at providing a
self-contained derivation of almost all the results and models discussed in the book,
with detailed proofs, thus making the book accessible for self-study. Furthermore,
this second edition contains more than 200 exercises. The solutions to most of the
exercises are given at the end of the book, while the solutions to the remaining
exercises can be downloaded from the publisher’s website. Some exercises contain
proofs of theoretical results which are not proven in the text, and other exercises
develop interesting side results or explore some models not discussed in the text,
while other exercises introduce simple applications of the results presented in the
text, with the aim of helping the reader to develop a better understanding of the
theory. Our presentation is partly influenced by several excellent books which have
appeared after the publication of the first edition. In particular, without any pretence
to be exhaustive, these include the books by Back [98], Cvitanić & Zapatero
[511], De Jong & Rindi [540], Demange & Laroque [551], Eeckhoudt et al. [631],
Munk [1363], Hens & Rieger [938], Lengwiler [1182], Pennacchi [1411] and Vives
[1630].
The book is organized as follows. The first seven chapters contain a presentation
of classical asset pricing theory, first in a single period setting (Chaps. 1–5) and
vii
viii Preface to the Second Edition (2017)
then in a dynamic setting (Chaps. 6–7), together with a discussion of the empirical
evidence on the implications of the theory. The last three chapters deal with more
advanced topics and present a number of extensions of classical asset pricing theory
as well as models going beyond classical asset pricing theory, in order to address
several asset pricing puzzles. Each chapter ends with a section dedicated to notes
and further readings and a collection of exercises. We now give a more detailed
outline of the contents of each chapter.
Chapter 1 presents a quick overview of some fundamental results of decision
theory and equilibrium theory under certainty (classical microeconomic theory).
The chapter starts with a discussion of preference relations and utility functions
and then introduces the concepts of general equilibrium, Pareto optimality and
representative agent. The aim of this chapter is to make the book as self-contained
as possible as far as classical economic analysis is concerned.
Chapter 2 lays the foundations of decision theory in the presence of risk. The
chapter starts with the presentation of expected utility theory and then analyses
the concepts of risk aversion, certainty equivalent, stochastic dominance and mean-
variance preferences, also explaining the diversification and the insurance principle.
Chapter 3, in a simple two-period setting, analyses the optimal portfolio problem,
first in the case of a single risky asset and then in the more general case of multiple
risky assets. The chapter contains several comparative statics results as well as
closed-form solutions under suitable assumptions on the utility function and on the
distribution of asset returns. The second part of the chapter contains a presentation
of the classical Markowitz’s mean-variance theory, exploring the properties of the
mean-variance portfolio frontier. We also establish several mutual fund separation
results. The last part of the chapter deals with optimal insurance problems and
optimal consumption-saving problems in the context of a two-period economy.
Chapter 4 deals with general equilibrium theory in a risky environment, where
agents interact in a financial market. The chapter starts by presenting the notion
of Pareto optimality and its implications in terms of risk sharing. The concept of
rational expectations equilibrium is introduced and characterized in the context
of a two-period economy. In this chapter, different financial market structures are
considered, with a particular attention to the important case of complete markets.
The last part of the chapter is devoted to the fundamental theorem of asset pricing,
which links the absence of arbitrage opportunities to the existence of a strictly
positive linear pricing functional. The relation of this important result to the
existence of an equilibrium of an economy and its implications for the valuation
of financial assets are also discussed.
Chapter 5, on the basis of the general equilibrium theory developed in Chap. 4,
presents some of the most important asset pricing models, including the consump-
tion capital asset pricing model (CCAPM), the capital asset pricing model (CAPM)
and the arbitrage pricing theory (APT). The relations of these asset pricing models
with the absence of arbitrage opportunities are also discussed. In this chapter, the
theoretical presentation of the models and of their implications is accompanied by
an overview of the empirical evidence reported in the literature. In particular, several
asset pricing anomalies and puzzles are described and discussed.
Preface to the Second Edition (2017) ix
Chapter 6 extends the analysis developed in the previous chapters to the case
of dynamic multiperiod economies. The chapter starts by studying the optimal
investment-consumption problem of an individual agent in a multiperiod setting,
by relying on the dynamic programming approach. Under suitable assumptions on
the utility function, closed-form solutions are derived. The chapter then proceeds
by extending the general equilibrium theory established in Chap. 4 to a dynamic
setting, introducing the notion of dynamic market completeness and analysing the
aggregation property of the economy. The fundamental theorem of asset pricing is
then established in a multiperiod setting, and its relation to the equilibrium of an
economy is discussed. Later in the chapter, the asset pricing relations presented in
Chap. 5 are extended to a dynamic context and specialized for several classes of
utility functions. The chapter ends by considering multiperiod economies with an
infinite time horizon and the possibility of asset price bubbles.
Chapter 7 is devoted to an extensive overview of the empirical evidence on
classical asset pricing theory. In particular, the attention is focused on the empirical
properties of the observed prices and returns and on several anomalies reported in
the literature, including the excess volatility phenomenon, the predictability of asset
returns, the equity premium puzzle, the risk free rate puzzle and other related asset
pricing puzzles.
Chapter 8 concerns the role of information in financial markets. The chapter starts
by studying the value of information from the point of view of an individual agent
and then proceeds by analysing the impact of information in the economy. This leads
to the introduction of the notion of Green-Lucas equilibrium and, in particular, to
the question of whether equilibrium prices transmit, aggregate and reveal the private
information of the agents. This concept is intrinsically linked to the informational
efficiency of financial markets and provides a microfoundation to the efficient
markets theory. In the chapter, the possibility or the impossibility of informationally
efficient markets is discussed and established in the context of several models. The
chapter closes by briefly examining the case where agents exhibit heterogeneous
opinions and surveying the empirical evidence on the informational efficiency of
financial markets.
Chapter 9 presents a critical analysis of several fundamental hypotheses underly-
ing the theory developed in the previous chapters of the book. More specifically,
the attention is focused on the notions of risk and uncertainty, on the classical
assumptions of expected utility theory, on the agents’ rationality and on the
imperfections of financial markets. By relaxing the assumptions made so far in
the book, several alternative paradigms are proposed and their implications are
discussed, also on the basis of the empirical evidence reported in the literature. Some
basic aspects of behavioural finance are explored and compared to the implications
of classical asset pricing theory.
Finally, Chap. 10 deals with financial markets microstructure. Relaxing the
assumption of perfect competition adopted so far in the book, this chapter explores
the real functioning of financial markets and the role of different categories of
market participants. In particular, the central themes are represented by the role
of information under imperfect competition and market liquidity. The chapter
x Preface to the Second Edition (2017)
closes with an overview of insider trading, market manipulation and the different
institutional features of financial markets.
In our presentation, we do not aim at presenting results in their most general
formulations with full mathematical details, but rather we aim at explaining the
central themes of financial markets theory in a transparent setting. For this reason,
we have chosen to keep the mathematical technicalities at a minimum level,
only requiring a basic knowledge of algebra, differential calculus and elementary
probability theory. While we present the theoretical results with complete and
rigorous proofs, we always try to emphasize the economic intuition over the
technical aspects. Even though some familiarity with basic microeconomics is
helpful, the book does not require any previous knowledge of economic theory,
since all the relevant results are reviewed in Chap. 1.
This book can be used in several ways. The first five chapters can form the basis
for a first course on the economics of uncertainty, focused on expected utility theory,
portfolio selection, equilibrium and no-arbitrage. The remaining chapters contain
more advanced material and can be used for more specialized courses or as a starting
point for the self-study of these topics. The extensive bibliography contained in
the book should be useful to every researcher in the field as a guide to the wide
literature on financial markets theory. An updated pdf file containing the errata of
the book will be available for download on the webpage http://sites.google.com/site/
fontanaclaud/fmt.
We are thankful to Gaetano La Bua who has checked the exercises of the book
and made several useful comments. We are also thankful to Gaetano for his help in
the preparation of the figures appearing in the text. We are grateful to Rémi Lodh
and Catriona Byrne from Springer for their precious assistance in the several stages
of the preparation of this second edition.
Milano, Italy Emilio Barucci

Paris, France Claudio Fontana
February 2017
From the Preface to the First Edition (2003)
This is just another book on financial markets theory. Why another book?
Most of the time an author will answer that there was no book covering the same
topics with the same approach. This is also my answer. Organizing my lectures for
an advanced financial markets theory course, I tried to make my students understand
how financial markets theory was a body in continuous transformation animated
both by a rich theoretical debate and by a strict interaction with real financial
markets. I did not find a book with such an approach and so I decided to write
this one. I hope my students were and will be fascinated by the picture.
The book is driven by two perspectives: on the one side the theoretical debate
about financial markets and agent’s behaviour under risk and on the other the
comparison of theoretical results with the empirical evidence. I intend to highlight
how financial markets theory has developed during the last 50 years along these
two perspectives. The first one has been a driving force for a long time; my
personal reconstruction of the development of the theory includes Bachelier (1900),
Arrow (1953), Modigliani and Miller (1958), Debreu (1959), Sharpe (1964), Lintner
(1965), Black and Scholes (1973), LeRoy (1973), Merton (1973), Jensen and
Meckling (1976), Rubinstein (1976), Ross (1977), Lucas (1978), Harrison and
Kreps (1979) and Grossman (1981). As a matter of fact, the classical asset pricing
theory was almost fully developed by the mid-1980s; the debate then turned to
the empirical evidence and hence the second perspective came into play. Results
are mixed with some puzzles still alive after 20 years. These puzzles generated a
debate with developments inside the classical asset pricing theory as well as outside
with an attempt to build alternative paradigms (e.g. behavioural finance). Classical
asset pricing theory tries to explain asset pricing anomalies by changing agent’s
preferences, probability distributions for fundamentals, economic environment
maintaining its traditional pillars, i.e. agents’ rationality, equilibrium or no-arbitrage
arguments, and rational expectations. According to the classical asset pricing theory,
asset prices are explained through risk factors related to asset fundamentals. On the
other hand, behavioural finance tries to build an alternative paradigm by relaxing
some hypotheses on agents’ rationality and introducing market frictions. The debate
on asset pricing puzzles is one of the main topics of the book. On this point
xi
xii From the Preface to the First Edition (2003)
we follow the Kuhn (1970) perspective: a paradigm has never been rejected by
falsification through direct comparison with the real world; a new paradigm should
be accepted.
The book is an adventure in modern financial markets theory with two plots:
theoretical developments and real financial markets. There are two actors: classical
asset pricing theory and heretics. Conclusions on the status of the theory are left to
the reader. My personal belief is that classical asset pricing theory is a strong and
flexible paradigm. Many anomalies can be reinterpreted inside the paradigm, but
some of them are hard to kill. On the other hand, behavioural finance and alternative
approaches provide useful insights to understand real financial markets day by day,
but they are not an alternative paradigm. Agent’s rationality is a simplification of
human behaviour and therefore some anomalies are expected.
The book can be used in several ways. It is an advanced financial markets theory
textbook, and it provides a handbook on recent developments of the literature. The
book covers a wide spectrum of topics. We are not going to deal with mathematical
finance topics (option pricing, term structure, interest rate derivatives) because in
our view they are mainly an application of the fundamental asset pricing theorem;
there are many interesting and hard to solve problems in this field, but they are
mainly technical problems. We handle the topics in the simplest setting (finite states-
discrete time). Some parts of the book have been heavily inspired by Barucci (2000).
The book owes a debt to Huang and Litzenberger (1988), which introduced me to
the economic analysis of financial markets. I strongly believe that the advancement
of the theory will be driven in the future by economic analysis, empirical evidence
and the incorporation of the institutional setting in the picture. This book testifies
this belief. I hope the reader will find interesting hints in reading the book.
It took 5 years to complete this book. During these years, the book has been a
living companion for me with frustrations, worries and many other emotions; now it
is only printed paper. I have to thank many for their encouragement and help. First
of all I would like to thank all my coauthors for having contributed a lot to this book.
I owe enormous intellectual debt to them. Special thanks to Maria Elvira Mancino
for her encouragement. I would like to thank my colleagues at the University of
Pisa, Bruce Marshall and Claudia Neri, for carefully reading the final manuscript.
The book is first of all dedicated to my students in the past and in the future, to their
enthusiasm which is a strong motivation for a teacher. The book is dedicated to all
who sympathized, sympathize or will sympathize with me, in particular my parents,
Teresa and those who are not with me anymore.
Firenze/Pisa, Italy Emilio Barucci

September 2002
Contents
1 Prerequisites.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 Choices Under Certainty .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 2
1.2 General Equilibrium Theory .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 7
1.3 Pareto Optimality and Aggregation . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.4 Notes and Further Readings . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 12
2 Choices Under Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
2.1 Expected Utility Theory . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
2.2 Risk Aversion .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.3 Stochastic Dominance.. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 34
2.4 Mean-Variance Analysis . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 38
2.6 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
3 Portfolio, Insurance and Saving Decisions . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55
3.1 Portfolio Theory .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55
3.2 Mean-Variance Portfolio Selection .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 76
3.3 Insurance Demand.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 102
3.4 Optimal Saving and Consumption .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
3.6 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 116
4 General Equilibrium Theory and No-Arbitrage .. . .. . . . . . . . . . . . . . . . . . . . 123
4.1 Pareto Optimality and Risk Sharing .. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
4.2 Asset Markets and Equilibria . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 138
4.3 Equilibrium with Intertemporal Consumption ... . . . . . . . . . . . . . . . . . . . 149
4.4 The Fundamental Theorem of Asset Pricing .. . .. . . . . . . . . . . . . . . . . . . . 164
4.6 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 192
5 Factor Asset Pricing Models: CAPM and APT . . . . .. . . . . . . . . . . . . . . . . . . . 201
5.1 The Consumption Capital Asset Pricing Model (CCAPM) . . . . . . . . 202
5.2 The Capital Asset Pricing Model (CAPM). . . . . .. . . . . . . . . . . . . . . . . . . . 213
xiii
xiv Contents
5.3 Empirical Tests of the CAPM . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 220

5.4 The Arbitrage Pricing Theory (APT).. . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 230
5.5 Empirical Tests of the APT . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 241
5.7 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 250
6 Multi-Period Models: Portfolio Choice, Equilibrium
and No-Arbitrage .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 255
6.1 Optimal Investment and Consumption Problems .. . . . . . . . . . . . . . . . . . 259
6.2 Equilibrium and Pareto Optimality .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 280
6.3 The Fundamental Theorem of Asset Pricing .. . .. . . . . . . . . . . . . . . . . . . . 292
6.4 Asset Pricing Relations . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 310
6.5 Infinite Horizon Economies and Rational Bubbles . . . . . . . . . . . . . . . . . 324
6.7 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 337
7 Multi-Period Models: Empirical Tests . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 347
7.1 Tests on the Price-Dividend Process: Bubbles and Excess
Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349
7.2 Tests on the Price-Dividend Process: Return Predictability .. . . . . . . 362
7.3 Tests on Intertemporal Equilibrium Models . . . .. . . . . . . . . . . . . . . . . . . . 378
7.5 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 392
8 Information and Financial Markets . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 397
8.1 The Role of Information in Financial Markets. .. . . . . . . . . . . . . . . . . . . . 403
8.2 On the Possibility of Informationally Efficient Markets.. . . . . . . . . . . 415
8.3 On the Impossibility of Informationally Efficient Markets.. . . . . . . . 427
8.4 Information in Dynamic Market Models .. . . . . . .. . . . . . . . . . . . . . . . . . . . 445
8.5 Difference of Opinions .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 451
8.6 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 458
8.8 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 469
9 Uncertainty, Rationality and Heterogeneity .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 479
9.1 Probability, Risk and Uncertainty . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 480
9.2 On Expected Utility Theory . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 488
9.3 Beyond Substantial Rationality .. . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 520
9.4 Bounded Rationality and Distorted Beliefs . . . . .. . . . . . . . . . . . . . . . . . . . 541
9.5 Incomplete and Imperfect Markets . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 551
9.7 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 574
10 Financial Markets Microstructure . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 583
10.1 The Role of Information Under Imperfect Competition .. . . . . . . . . . . 585
10.2 Order-Driven Markets .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 594
10.3 Quote-Driven Markets . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 607
Contents xv
10.4 Multi-Period Models of Market Microstructure . . . . . . . . . . . . . . . . . . . . 619

10.5 Market Abuse: Insider Trading and Market Manipulation . . . . . . . . . 629
10.6 Market Design .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 634
10.8 Exercises .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 650
11 Solutions of Selected Exercises . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 661
11.1 Exercises of Chap. 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 661
11.9 Exercises of Chap. 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 751
References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 765
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 829
Chapter 1
Prerequisites
All theory depends on assumptions which are not quite true.

That is what makes it theory. The art of successful theorizing is
to make the inevitable simplifying assumptions in such a way
that the final results are not very sensitive. A “crucial”
assumption is one on which the conclusions do depend
sensitively, and it is important that crucial assumptions be
reasonably realistic. When the results of a theory seem to flow
specifically from a special crucial assumption, then if the
assumption is dubious, the results are suspect.
Solow (1956)
Once it has achieved the status of paradigm, a scientific theory

is declared invalid only if an alternative candidate is available
to take its place. No process yet disclosed by the historical study
of scientific development at all resembles the methodological
stereotype of falsification by direct comparison with nature.
[: : :] The act of judgement that leads scientists to reject a
previously accepted theory is always based upon more than a
comparison of that theory with the world. The decision to reject
one paradigm is always simultaneously the decision to accept
another, and the judgement leading to that decision involves the
comparison of both paradigms with nature and with each other.
Kuhn (1970)
In this chapter, we provide a brief overview of the fundamental results on pure

exchange economies under certainty. In Sect. 1.1, we present the classical axiomatic
framework to decision making under certainty, discussing the axioms of preference
relations and introducing the notion of utility function. In Sects. 1.2 and 1.3,
we provide the main results on general equilibrium theory and we introduce the
notions of Pareto optimality and representative agent together with the two Welfare
Theorems.
© Springer-Verlag London Ltd. 2017 1

E. Barucci, C. Fontana, Financial Markets Theory,
Springer Finance, DOI 10.1007/978-1-4471-7322-9_1
2 1 Prerequisites
1.1 Choices Under Certainty
Let us consider an agent who makes choices under certainty. In an environment

which is not affected by any source of randomness, choices can be analysed by
means of two ingredients:
• a set of choices X;
• a weak preference relation R representing the agent’s preferences over couples
of elements of X.
In the classical consumption problem, choices concern bundles of goods and,
therefore, X will be a subset of RL (e.g., X D RLC ), where L 2 N is the number
of goods.1 A vector x 2 X identifies a bundle of goods.
A (weak) preference relation R is defined on X. Given x; y 2 X; if x R y we
say that the bundle of goods x is at least as preferred as the bundle of goods y. The
preference relation R allows us to deduce two relations:
• the strong preference relation P defined as
x P y ” x R y but not y R x;
meaning that the bundle of goods x is strictly preferred to the bundle of goods y;
• the indifference relation I defined as
x I y ” x R y and y R x;
meaning that the bundle of goods x and the bundle of goods y are equally
preferred.
The couple .X; R/ characterizes completely the agent’s choice problem: the
choice set X and the preference relation R. To yield a meaningful analysis, some
hypotheses on the preference relation are needed. We will assume that the preference
relation R satisfies the following rationality axiom.
Assumption 1.1 (Rationality) The preference relation R on the set of choices X
is said to be rational if it satisfies the following properties:
• Reflexivity: for every x 2 X, it holds that x R x ;
• Completeness: for every x; y 2 X, it holds that x R y or y R x ;
• Transitivity: for every x; y; z 2 X, if x R y and y R z, it then holds that x R z .
1
The set RLC denotes the set of non-negative vectors and RLCC the set of strictly positive vectors
with L elements.
1.1 Choices Under Certainty 3
According to the above assumption, an agent is rational if he is able to rank any

couple of alternatives and if he is consistent over different choices. The transitivity
of the preference relation represents the hypothesis that the agent is able to “think
big”, not only with respect to a single couple of alternatives.
In order to mathematically describe decision making problems, it is useful to
introduce a utility function. To this end, we require the preference relation R to
satisfy an additional technical hypothesis called the continuity hypothesis.
Assumption 1.2 (Continuity) The preference relation R on the set of choices X is
said to be continuous if, for every x 2 X, the sets fy 2 X W y R xg and fy 2 X W x R yg
are closed.
We say that a utility function u W X ! R represents the preference relation R if
x R y ” u.x/ u.y/; for all x; y 2 X:
The following theorem holds (see Mas-Colell et al. [1310, Proposition 3.C.1] for a
proof).
Theorem 1.1 Let R be a preference relation on the set of choices X satisfying
Assumptions 1.1 and 1.2. Then R can be represented through a continuous utility
function u W X ! R.
If Assumptions 1.1 and 1.2 hold, then the decision making problem of an agent
characterized by the preference relation R on the set of choices X will be completely
described by the couple .X; u/. Note that Theorem 1.1 does not assert the uniqueness
of the utility function u representing the preference relation R. Indeed, if the
utility function u represents the preference relation R, then any strictly increasing
transformation of u will also represent the same preference relation R. For a given
utility function u W X ! R, the indifference sets (or indifference curves) of u are
defined as the level sets of the function u, i.e., the sets fx 2 X W u.x/ D cg, for c 2 R.
Some important properties of a preference relation R are the following2:
• R is strictly monotone if, for every x; y 2 X such that y > x, then y P x ;
• R is strictly convex if, for every x; y; z 2 X with x ¤ y such that x R z and y R z ,
then ˛ x C .1 ˛/y P z, for every ˛ 2 .0; 1/.
If the preference relation R satisfies the two above properties, then the indiffer-
ence sets of the utility function u are ordered in an increasing sense (equivalently,
u is strictly increasing) and u.˛x C .1 ˛/y/ > min.u.x/; u.y//, for all bundles of
goods x; y 2 X with x ¤ y and ˛ 2 .0; 1/. In the following, we shall generally
assume utility functions to be twice continuously differentiable. In that case, if
the above properties hold, then the first partial derivative of a utility function
2
Given two vectors x; y 2 RL ; the inequality x y means that xn yn .8 n D 1; : : : ; L/, while
x > y means that xn yn .8 n D 1; : : : ; L/ with x ¤ y, and x >> y means that xn > yn .8 n D
1; : : : ; L/.
4 1 Prerequisites
(marginal utility) with respect to a consumption good is positive and the second
partial derivative is negative. In the following, Assumptions 1.1 and 1.2 will be
always supposed to hold unless otherwise specified.
We dip the agent in a perfectly competitive market. Following the classical
approach, the interaction of multiple agents in the market can be addressed in two
subsequent steps. The first part of the problem is addressed in this section, the
second part in the next one. In a perfectly competitive market, agents do not affect
market prices (i.e., agents are price takers) and, therefore, we can parametrize their
behavior with respect to market prices. In the following, we identify market prices
by a vector p 2 RLC .
We assume that every agent is characterized by substantial rationality, i.e., an
agent pursues his goals in the most appropriate way under the constraints imposed
by the environment. In a perfectly competitive market, the constraints are given by
the budget constraint (determined by market prices and wealth): given the wealth
w 2 RC and the price vector p 2 RLC , the set on which an agent makes his choices
is
B WD fx 2 X W p> x wg;
for some closed set X RLC and with the superscript > denoting transposition.
According to the substantial rationality paradigm, the most appropriate bundle of
goods x 2 X solving the agent’s optimal choice problem will be the one resulting
from the maximization of the utility function representing the agent’s preferences
subject to the budget constraint. Hence, a vector x 2 X is obtained as the solution
of the following optimization problem:
max u.x/: (MP0)

x2B
Note that, if the utility function u is continuous (as follows from Theorem 1.1),
then there always exists a solution to Problem (MP0) if p 2 RLCC . This is a direct
consequence of the fact that any continuous function admits a maximum over a
compact set (see Mas-Colell et al. [1310], Proposition 3.D.1). Moreover, by the strict
monotonicity of the utility function u, it is never optimal for an agent not to use all
his wealth w, so that the budget constraint can be written as an equality constraint
and the optimum problem can be handled via the Lagrange multipliers method. For
simplicity of presentation, we shall suppose from now on that X D RLC . Given the
Lagrangian
L.x; / D u.x/ C .w p> x/;

1.1 Choices Under Certainty 5
where represents the Lagrange multiplier, the first order necessary conditions for
a strictly positive optimal consumption vector x 2 RLCC become3
uxk .x / D pk ; for all k D 1; : : : ; LI

(1.1)
w D p > x ;
where uxk denotes the first partial derivative of the utility function u with respect
to its k-th argument. The quantities xk and pk denote the amount of good k and its
price, respectively. Condition (1.1) shows that the gradient of the utility function
u in correspondence of the optimal choice x must be equal to the price vector p
multiplied by the Lagrange multiplier .
Associated with the utility maximization problem (MP0), we define the indirect
utility function v W RC RLCC ! R by v.w; p/ WD u.x /, where x 2 RLCC is
the solution to problem (MP0) for wealth w and market prices p. As can be checked
from (1.1), the Lagrange multiplier represents the derivative of the indirect utility
function with respect to wealth.
A strictly positive optimal solution x 2 RLCC is obtained if the marginal utility
for every good goes to C1 when the amount of the good tends to zero and if it goes
to zero when the amount of the good increases to C1 (Inada conditions), i.e.,
lim u0 .x/ D 0 and lim u0 .x/ D C1:

x!C1 x&0
Moreover, when the utility function is strictly concave, the optimal solution x is
unique. Due to condition (1.1), the optimality conditions for a vector x 2 RLCC can
be synthesized in the following classical requirement (provided that p 2 RLCC ):
uxk .x / pk

D ; for all k; ` D 1; : : : ; L: (1.2)
ux` .x / p`
We call marginal rate of substitution between good k and good ` the ratio of the
partial derivatives of the utility function u with respect to xk and x` , respectively.
Condition (1.2) means that, in correspondence of the optimal solution x , marginal
rates of substitution are equal to price ratios between different goods. The optimal
consumption vector x associated with the price vector p will be denoted by the
function x. p/. The demand of agent i, endowed with the vector of goods ei 2 RLC
(which corresponds to the wealth endowment w D p> ei ), is given by the function
zi W RLC ! RL defined as zi . p/ WD xi . p/ ei . The inequality zik . p/ > 0 means that
agent i has a positive demand of good k, while zik . p/ < 0 means that agent i has
a positive supply (negative demand) of good k, in correspondence of the vector of
prices p.
3
Given a function f W RN ! R of N real variables, the partial derivative with respect to the variable
xi will be denoted by fxi , while the partial derivative evaluated in x 2 RN will be denoted by fxi .x/.
6 1 Prerequisites
The maximization of the utility function under the budget constraint in a perfectly
competitive market represents one of the pillars of economic analysis. In what
follows, we will refer to it as the internal consistency requirement.
In the present context, we can also analyse saving and optimal consumption
decisions in an intertemporal setting under certainty. For simplicity, let us consider
a one good economy (L D 1), with the representative good being generically
interpreted as wealth, and two distinct time periods t D 0 (today) and t D 1
(tomorrow). Let e0 denote the wealth at time t D 0 and e1 the wealth (income)
at time t D 1. We suppose that at t D 0 the agent can choose to save an amount s
of his initial wealth e0 and invest that amount in a risk free asset (i.e., bank account)
yielding rf > 0 at date t D 1 per unit of wealth invested at date t D 0. If the agent
decides to save the amount s, then the consumption c0 and c1 at t D 0 and t D 1,
respectively, will be given by
c0 D e0 s and c1 D e1 C s rf ;
thus leading to the budget constraint

c1 e1
c0 C D e0 C : (1.3)
rf rf
Observe that rf can be interpreted as the ratio between the price of one unit of wealth
obtained tomorrow (at t D 1) and the price of one unit of wealth today (at t D 0). We
assume that the agent’s preferences can be represented by a utility function additive
across time, i.e., there exists a utility function u W RC ! R such that the agent’s
preferences can be represented by
u.c0 / C ı u.c1 /; (1.4)
where ı 2 .0; 1/ represents a (subjective) discount factor, accounting for the

natural fact that wealth obtained in the future is less attractive than wealth obtained
today. Assuming that u is strictly increasing, the first order condition for the
maximization of the utility function (1.4) with respect to c0 and c1 subject to the
budget constraint (1.3) yields
u0 .c0 /
D rf ı: (1.5)
u0 .c1 /
In particular, assuming that the function u is strictly concave, this optimality

condition implies that
c0 < c1 ” rf ı > 1;
meaning that the agent consumes more in t D 1 than in t D 0 if and only if the
return rf on the risk free asset is greater than the reciprocal of the discount factor
1.2 General Equilibrium Theory 7
ı. Furthermore, it can be easily shown from (1.5) that the optimal saving rate s is
decreasing with respect to the income at t D 1, while it is increasing with respect to
the wealth at t D 0. As far as the risk free rate is concerned, the sensitivity analysis
depends on the substitution and income effect (compare with the analysis in Sect. 3.4
under risk). Note that if rf ı D 1, the optimality condition yields c0 D c1 , showing
a preference for consumption smoothing over time.
1.2 General Equilibrium Theory
In the previous section we have studied the decision making problem of a single
agent. Let us now consider an economy with I > 1 agents and L > 1 goods. The
economy is fully described by I couples .R i ; ei /, for i D 1; : : : ; I. The vector ei 2
RLC represents the endowment of agent i in terms of the L goods and R i denotes his
preference relation.
We assume that every agent i has solved his optimal consumption problem (MP0)
in correspondence of the wealth wi D p> ei , so that his choices are represented by
the demand function zi . p/. By the function z W RLC ! RL defined as z. p/ WD
PI
iD1 z . p/ we denote the aggregate excess demand in the economy.
i
In a perfectly competitive economy, the market represents an anonymous

fictitious place of centralized trades where agents communicate exclusively through
prices. Economic theory has addressed market interaction favoring the equilibrium
analysis, i.e., the analysis of the economy in correspondence of a price vector p
such that the aggregate demand is equal to the aggregate supply:
z. p / D 0: (1.6)
Equation (1.6) actually consists of L equations in L unknowns (prices), one for

each good. A price vector p solving system (1.6) is called an equilibrium price
vector. A couple . p ; x / 2 RLC RCLI
, where x D .x1 . p /; : : : ; xI . p // is the
optimal demand of the I agents in the economy, is called competitive equilibrium
and the matrix x 2 RLI C is called equilibrium allocation. We will refer to (1.6) as
the external consistency requirement. A competitive equilibrium is thus represented
by a couple of price vector and allocation of goods satisfying the internal as well
as the external consistency requirements. In other words, in a competitive market
equilibrium, every agent chooses an optimal consumption bundle, coherently with
his budget constraint (internal consistency), and all goods are fully allocated among
the agents (external consistency). Condition (1.6) is required with strictly increasing
preferences: under weaker assumptions (local non-satiation), it can be replaced by
z. p / 0.
The question of the existence of a vector of prices p solving system (1.6) is
addressed in the following result (see Mas-Colell et al. [1310, Propositions 17.B.2
and 17.C.1]).
8 1 Prerequisites
Theorem 1.2 Let the preference relation R i be rational, continuous, P strictly

monotone and strictly convex, for all i D 1; : : : ; I, and suppose that IiD1 ei 2 RLCC .
Then there exists an equilibrium price vector p 2 RLCC solving (1.6).
This result does not represent the end of the story. Indeed, it might well happen
that the equilibrium is not unique, in which case coordination problems arise:
on which equilibrium will agents coordinate? There are coordination failure and
selection of equilibria problems. Provided that agents have the capacity to carry out
the above steps, which equilibrium price will they choose among those compatible
with the economy? We just mention that, in order to establish the uniqueness of the
equilibrium, additional properties on the excess demand function are required (see
also the notes at the end of this chapter).
In correspondence of a competitive equilibrium . p ; x / 2 RLCC RLI CC with a
strictly positive allocation of goods, it holds that
uixk .xi / pk

D ; for all k; ` D 1; : : : ; L and i D 1; : : : ; I; (1.7)
uix` .xi / p`
where xi denotes the equilibrium consumption bundle of agent i, for i D 1; : : : ; I.

The above condition means that the marginal rate of substitution of every agent
i for any couple of goods .k; `/ is equal to the equilibrium price ratio pk =p` .
As a consequence, marginal rates of substitution are equal among all agents.
Condition (1.7) can also be expressed in the following equivalent way: given an
equilibrium . p ; x / 2 RLCC RCC LI
, there exists a vector of positive real numbers
.1 ; : : : ; I / such that rui .xi / D i p , for all i D 1; : : : ; I. The term i is the
Lagrange multiplier of the optimal consumption problem (MP0) of agent i in
correspondence of the equilibrium prices p . Therefore, in correspondence of the
equilibrium prices p , the gradients of the utility functions of all the agents are
proportional to the equilibrium price vector.
1.3 Pareto Optimality and Aggregation
Equilibrium theory provides us with a tool for the analysis of the market. As
economics is a social science, we need tools for evaluating the outcome of the
market’s coordination. The privileged tool is represented by the Pareto optimality
criterion.
Given an economy described by I couples .R i ; ei /, with ei 2 RLC , for i D
1; : : : ; I, we denote an allocation of goods by the tuple x D .x1 ; :P : : ; xI /, with
xi 2 RLC for all i D 1; : : : ; I. We say that the allocation x is feasible if IiD1 xi e,
PI
where e WD i
iD1 e is the aggregate endowment. We can now introduce the
following Pareto optimality criterion to evaluate the efficiency of an allocation:
given two feasible allocations x; y, we say that y Pareto dominates x if yi R i xi , for all
1.3 Pareto Optimality and Aggregation 9
i D 1; : : : ; I, and yj P j xj for at least one agent j 2 f1; : : : ; Ig. We can then give the
following definition.
Definition 1.3 A feasible allocation x is an efficient allocation (or Pareto optimal
allocation) if there exists no other feasible allocation y that Pareto dominates x.
Using the representation of the preference relations .R 1 ; : : : ; R I / via utility
functions .u1 ; : : : ; uI / (see Theorem 1.1), an allocation x 2 RC
LI
is Pareto optimal
if and only if there does not exist an allocation x 2 RC such that
LI
X
I X
I
xik eik ; for all k D 1; : : : ; L;
iD1 iD1
and
ui .xi / ui .xi /; for all i D 1; : : : ; I;
with strict inequality holding for some j 2 f1; : : : ; Ig.

The set of Pareto optimal allocations constitutes the contract curve which can be
described by means of the Edgeworth box in a simple economy with two agents and
two goods (see Sect. 4.1).
The following two Welfare Theorems establish a connection between equilibrium
and Pareto optimality (see Mas-Colell et al. [1310, Chapter 16] for more details).
Theorem 1.4 (First Welfare Theorem) If . p ; x / is a competitive equilibrium
and the preference relations R i are strictly monotone, for all i D 1; : : : ; I, then the
equilibrium allocation x is a Pareto optimal allocation.
Theorem 1.5 (Second Welfare Theorem) Let x be a Pareto optimal allocation,
with xi 2 RLCC for all i D 1; : : : ; I. If R i is continuous, strictly monotone and
convex, for all i D 1; : : : ; I, then x is a competitive equilibrium allocation once the
initial endowment ei D xi , i D 1; : : : ; I, is allocated to the agents.
The first theorem provides a formal proof of the presence of an “invisible
hand” in the market: the agents, maximizing their utility in a perfectly competitive
market, reach an allocation which is socially optimal. The two theorems can be
read in opposite perspectives. The first Welfare Theorem is the crowning of the
dreams of the supporters of the free market: a perfectly competitive market with
agents maximizing their utilities leads to an outcome which is optimal for the
society (according to the Pareto optimality criterion). On the other side, there
are many optimal allocations and the Pareto criterion does not induce a complete
order among feasible allocations. Therefore, there is room for wealth redistribution,
i.e., for choosing among Pareto optimal allocations. The second Welfare Theorem
establishes that we can reach in equilibrium any Pareto optimal allocation if the
resources are redistributed ex-ante in an appropriate way (lump sum transfers). In
particular, if the initial allocation is already a Pareto optimal allocation then it also
represents an equilibrium allocation and the agents have no incentive to trade.
10 1 Prerequisites
The identification of Pareto optimal allocations as equilibrium allocations pro-

vides an interesting tool for the analysis of the economy at the aggregate level and,
in particular, of the equilibrium price vector p . Indeed, under some conditions,
the equilibrium price vector of the economy can be determined by the choices of a
single representative agent or consumer.
Let us consider an economy described by f.ui ; ei /I i D 1; : : : ; Ig and a
corresponding competitive equilibrium . p ; x /. We want to characterize the repre-
sentative agent/consumer through a utility function such that the optimal choice for
that
PI utility function in correspondence of p will be the aggregate endowment e D
i
iD1 e . This means that, if the representative agent is endowed with the aggregate
endowment e, then his excess demand will be equal to zero in correspondence of
the equilibrium prices p of the economy with I agents. Note that, in an economy
with a single agent, the external consistency requirement amounts to a no-trade
condition and, therefore, the couple . p ; e/ is an equilibrium for the economy with
one representative agent ( no-trade equilibrium).
We endow our representative agent with a utility function u W RLC ! R defined
on the set RLC of aggregate consumption vectors x. For any x 2 RLC , the function u.x/
is obtained by maximizing a weighted sum of the utilities ui of the I agents among
all allocations .x1 ; : : : ; xI / which are feasible with respect to a given aggregate
endowment x. More precisely, the function u.x/ is called social welfare function
and is defined as follows, for any x 2 RLC :
X
I
u.x/ WD max ai ui .xi /; (1.8)
xi 2 RLC ; iD1;:::;I
iD1
where the maximization is subject to the constraint
X
I
xik xk ; for all k D 1; : : : ; L: (1.9)
iD1
The coefficients ai 0, for i D 1; : : : ; I, represent the weights assigned to the utility

functions of the individual agents. For a fixed x 2 RLC , an allocation .x1 ; : : : ; xI / 2
RCLI
satisfying the feasibility constraint (1.9) is said to define the function u.x/ if it
P
holds that u.x/ D IiD1 ai ui .xi /. The following proposition describes the relation
between a Pareto optimal allocation and the social welfare function u introduced
in (1.8)–(1.9) (see Varian [1610] and Mas-Colell et al. [1310, Proposition 16.E.2]
for a proof).
Proposition 1.6 Consider an economy with aggregate endowment e 2 RLCC and
continuous, strictly increasing and concave utility functions ui , for all i D 1; : : : ; I.
Let u.e/ be defined as in (1.8)–(1.9), with x D e. Then the following hold:
(i) the allocation .x1 ; : : : ; xI / 2 RC
LI
which defines u.e/ with ai > 0, for all
i D 1; : : : ; I, is Pareto optimal;
1.3 Pareto Optimality and Aggregation 11
(ii) let .x1 ; : : : ; xI / be a Pareto optimal allocation such that xi 2 RLCC for all
i D 1; : : : ; I. Then there exists a set of weights fai 0I i D 1; : : : ; Ig such
that .x1 ; : : : ; xI / defines u.e/ and not all weights are null;
(iii) let fai > 0I i D 1; : : : ; Ig be the set of weights associated to a Pareto optimal
allocation .x1 ; : : : ; xI / 2 RCCLI
. Then, for every i D 1; : : : ; I, it holds that
i i
a D 1= , i.e., the weight ai is equal to the reciprocal of the Lagrange
multiplier i of the optimum problem (MP0) of agent i in correspondence of
the vector of equilibrium prices p and the allocation .x1 ; : : : ; xI /.
Thanks to the above proposition, one can choose a set of strictly positive weights
ai , i D 1; : : : ; I. Then, with such a choice of weights, the problem
X
I
max ai ui .xi /;
xi 2 RLC ;iD1;:::;I
iD1
P
under the constraint IiD1 xik ek , for all k D 1; : : : ; L, yields as solution an
allocation .x1 ; : : : ; xI / which is Pareto optimal and, therefore, of equilibrium for
an economy with I agents. Furthermore, the Lagrange multiplier i of the optimal
consumption problem of agent i in correspondence of the equilibrium price vector
p associated to the allocation .x1 ; : : : ; xI / is given by i D 1=ai .
The above result allows us to establish that the representative agent endowed with
the aggregate resources e 2 RLC of the economy and with a utility function given
by the social welfare function will not trade in correspondence of the equilibrium
prices of the original economy with I agents. Let p be an equilibrium price
vector supported by the equilibrium allocation x in the economy with I agents.
By Theorem 1.4, the allocation x is Pareto optimal and, in view of Proposition 1.6,
it defines the social welfare function u with respect to the aggregate endowment e,
see (1.8)–(1.9). Consider then the problem of maximizing the function u:
max u.x/; (1.10)

x 2 RLC
under the constraint . p /> .x e/ 0. Note that, for all k D 1; : : : ; L,
X
I
dxi X 1 I i
i > dx
uxk .e/ D ai rui .xi /> .e/ D i
p .e/ D pk ; (1.11)
iD1
dxk iD1
dx k
where rui .xi / denotes the gradient of the utility function ui in correspondence of
the optimal consumption vector xi of agent i. In (1.11), the second equality follows
from the fact that xi represents the optimal consumption vector for agent i, for
all i D 1; : : : ; I, in correspondence of the price vector p , together with the first
order condition (1.1) and part (iii) of Proposition 1.6. The last equality in (1.11) is
due to the fact that the allocation .x1 ; : : : ; xI / is feasible, together with the strict
monotonicity of the utility functions ui , for i D 1; : : : ; I. Condition (1.11) implies
12 1 Prerequisites
that the representative agent will not trade in correspondence of the prices p (i.e.,
x D e), thereby defining a competitive equilibrium . p ; e/ for an economy with a
unique agent (no-trade equilibrium). Hence, the equilibrium prices of the economy
with I agents are obtained as equilibrium prices of an economy composed by a
single agent with a specific utility function (social welfare function) and endowed
with the aggregate resources of the entire economy.
The function u.x/ defined in (1.8)–(1.9) depends on the weights ai , i D 1; : : : ; I,
and, therefore, on the Pareto optimal allocation under consideration. Indeed, the
utility function of the representative agent associates a Pareto optimal allocation
.x1 ; : : : ; xI / with a vector of weights .a1 ; : : : ; aI / and, through the no-trade
equilibrium condition, with the corresponding vector of equilibrium prices p . The
vector p coincides with the equilibrium prices of the economy with I agents where
each agent i is endowed with xi , for i D 1; : : : ; I. As the weights change, the Pareto
optimal allocation and the social welfare function u change: what happens to the
equilibrium prices?
In general, equilibrium prices also change as the weights change and, therefore,
the prices of the economy with I agents depend on the wealth distribution. We
will say that an economy enjoys the aggregation property (or that there exists a
representative agent in a strong sense) if equilibrium prices only depend on the
resources of the whole economy and not on the weights or on the initial allocation.
A sufficient condition for an economy to satisfy this property is that the preferences
of all the agents are quasi-homothetic, i.e., characterized by indirect utility functions
of the Gorman form vi .wi ; p/ D ˛i . p/ C ˇ. p/ wi , for all i D 1; : : : ; I, with the
coefficient ˇ. p/ being the same for all the agents (see e.g. Mas-Colell et al. [1310,
Chapters 4 and 16]). In that case, the aggregate demand is independent of the initial
allocation and the equilibrium prices p defined through the no-trade equilibrium
condition depend neither on the weights .ai ; : : : ; aI / nor on the allocation of the
goods, but only on the resources of the economy as a whole. Three classes of
utility functions satisfy these requirements: logarithmic utility, power utility and
exponential utility. If the aggregation property holds, then the equilibrium prices
of the economy with I agents can be identified through the optimum problem of
the representative agent endowed with the resources of the economy and the no-
trade condition, regardless of the wealth distribution. In this sense, the economy
with I agents and the economy with the representative agent are observationally
equivalent.
1.4 Notes and Further Readings
In economic theory, the focus on equilibrium analysis is due to several reasons.

In what follows we only propose some remarks, referring to Arrow [77], Hahn
[874], Ingrao & Israel [1001] for a more comprehensive analysis. In the first place,
equilibrium analysis has a methodological-historical motivation: in comparison with
other sciences, economic theory is a young science. As a consequence, economics
1.4 Notes and Further Readings 13
has borrowed methodologies and instruments from more mature sciences (e.g.,
physics) where equilibrium analysis represents a classical tool.
The definition of an equilibrium state implies the identification of dynamics in the
phenomenon under consideration, dynamics which seem to be completely outside
the picture described so far. Actually, we have not proposed a dynamic system, but
we can identify a dynamic process in our context by referring to the internal and to
the external consistency requirements. If p ¤ p , then one of the two consistency
requirements is violated and one of the following situations is verified: (a) given
the prices, not all agents maximize their utility; (b) there are no goods to satisfy
the demand of all agents or some goods do not have a buyer. In both cases, it seems
plausible that some forces inside the market will move the prices. The main problem
is how to describe these forces. On the other hand, if the economy stands exactly in
p , then there are no endogenous forces moving prices: if agents trade in a perfectly
competitive market and communicate exclusively through prices, then only p can
represent a stationary state of the economy.
This interpretation appears rather weak, since nothing is said about the ability of
the market to converge to p starting from a different price vector p ¤ p . Stability
analysis of the equilibrium requires introducing the time dimension in the above
picture and to formalize the dynamics outside the equilibrium. Pursuing this goal,
general equilibrium theory shows all its limits: it is difficult in fact to represent
internal and external consistency requirements through a dynamical system. The
two consistency requirements, which represent a strong point to prove the existence
of the equilibrium and to characterize it analytically, represent a serious obstacle
to introducing dynamic elements in the context of general equilibrium theory. It is
difficult to model disequilibrium dynamics with consumption-exchange of goods.
These problems have been avoided in the literature by extending the time for trade
and admitting consumption only at the final time when an equilibrium is reached. In
this way, the internal consistency requirement is maintained, every agent during the
bargaining declares his theoretical demand, while at that time there is no trade and
no consumption. Trade and consumption only occur in equilibrium. Prices change
over time when demand and supply differ. The dynamics is modelled by assuming
a tâtonnement process: the price dynamics is described by a differential-difference
equation defined through a function which preserves the sign of the excess demand
and assumes value zero if the excess demand is equal to zero. If the excess demand
function satisfies some additional hypothesis (typically, goods are gross substitutes:
two goods i and j are said to be gross substitutes if for any price vector p it holds
that @zi =@pj > 0 with i ¤ j), then the competitive equilibrium will be stable with
respect to the tâtonnement process, see Mas-Colell et al. [1310, Proposition 17.H.9].
The dynamic process described above formally represents the so-called demand and
supply law.
There is another reason for using equilibrium analysis. The existence of a vector
of equilibrium prices means that the decisions of the agents pursuing their self-
interest are compatible in a perfectly competitive market. This motivation goes back
to the axiomatic approach of general equilibrium theory developed in the ’30s as
an answer to the following question: under what conditions are agents’ decisions
14 1 Prerequisites
compatible? The question circulated for a long time in the environment of social
scientists (since Adam Smith, if not before) and it was motivated by the fact that
the market seemed able to coordinate the demand of all agents. It is obvious that in
this perspective the accent is on equilibrium existence rather than on its dynamics.
To this end, consistency requirements simplify the analysis in a substantial way and
allow to bring the problem of the compatibility of agents’ decisions back to the
solution of a (non-linear) system of L equations in L variables. In this perspective,
a positive answer to the problem of the existence of a vector of equilibrium prices
appears a complete success for general equilibrium theory.
In this perspective, the existence of a general economic equilibrium leads to
interesting philosophical-moral-theoretical implications, but little implications on
the real world. On this side, an implication is the following: if agents have a thorough
knowledge of the economy and know all the couples .R i ; ei /, for i D 1; : : : ; I, then
they can cover the road described in this chapter to identify the vector of equilibrium
prices p and determine their demand consistently. This interpretation of general
equilibrium theory requires very strong hypotheses: the two consistency conditions
are satisfied, agents pursue their self-interest, they know the economic model and
they are endowed with a remarkable computational capacity.
Concerning the uniqueness of the equilibrium vector p 2 RLCC in Theorem 1.2,
the fact that all goods are gross substitutes ensures that the equilibrium price vector
p 2 RLCC is unique, see Mas-Colell et al. [1310, Proposition 17.F.3]. However, this
assumption implies strong hypotheses on the initial allocation of goods as well as
on the preference relations of the agents.
Chapter 2
Choices Under Risk
Behaviour is substantively rational when it is appropriate to the

achievement of given goals within the limits imposed by given
conditions and constraints. [: : :] behaviour is procedurally
rational when it is the outcome of appropriate deliberation. is
the outcome of appropriate deliberation.
Simon (1976)
The optimal consumption problem introduced in the previous chapter is formulated

in a very simple context: agents act in a single time period and choices concern
bundles of goods. The characteristics of the goods are perfectly known ex-ante
and, hence, the optimal consumption problem is a purely deterministic optimization
problem. The introduction of risk requires at least two time periods t 2 f0; 1g. This
is due to the necessity of describing the agents’ imperfect knowledge: at the initial
date t D 0 agents do not have a full knowledge of the world, they only have some
beliefs about it, while at t D 1 the uncertainty is fully resolved and revealed. The
agents’ ignorance at the initial date t D 0 with respect to the state of the world at
t D 1 may concern many different features (economic conditions, meteorological
conditions, quality of the goods, dividends, losses, etc.) and several examples will
be discussed in the following.
If an agent only makes his choices at t D 1, once the state of the world has been
revealed, then we relapse into the analysis developed in the previous chapter. In this
chapter, we aim at analysing the choices of an agent at the initial date t D 0, when he
does not have a complete knowledge of the state of the world and, therefore, of the
future consequences of the choices made at t D 0. We assume that the true state of
the world can be only observed at t D 1. In this setting, we first have to describe the
status of ignorance of the agent and then to analyse his choices through something
similar to the couples .X; R/ or, equivalently, .X; u/ used in the previous chapter
under certainty. To this end, we introduce two crucial elements: a probability space
and a utility function u.
The beliefs of the agents at t D 0 about the state of the world at the future
date t D 1 are described by means of a probability space .˝; ; P/. The set ˝ is
the set of all possible states of the world at time t D 1: a state of the world (or
state of nature) is represented by an element ! 2 ˝ and provides a complete and
exhaustive description of the world at t D 1 (an element ! 2 ˝ is also called

16 2 Choices Under Risk
elementary event). The states of the world are mutually exclusive, i.e., only one
elementary event is realized at t D 1. The collection is a -algebra on ˝.1 If ˝
is a finite set, then a natural choice for is given by the set of all the subsets of
˝. An event is an element E 2 and, therefore, a subset of ˝. The application
P W ! Œ0; 1 is a probability measure, with P.E/ denoting the probability of
the event E 2 . Intuitively, P.E/ represents the confidence an agent places on the
realization of the event E. The probability of an event can have an objective or a
subjective interpretation. The distinction is however subtle and it also depends on
the situation faced by the agent (see Chap. 9). In what follows, we will typically
assume that the probability has an objective interpretation.2
At the initial date t D 0, the agent is assumed to know the probability space
and, therefore, the set ˝ of all possible elementary events as well as the probability
measure P. At t D 1, the state of the world is fully revealed and the agent discovers
the realized elementary event ! 2 ˝. When an agent cannot observe at t D 0
the state of the world but has a complete knowledge of the probability space, we
say that he faces a risky situation. This situation differs from the case of an agent
who does not perfectly know the probability of the events: in this case, we will say
that the agent faces an uncertain situation. In our analysis, the word risk refers to a
1
is a collection of subsets of ˝ that includes the empty set and satisfies the following
conditions:
– if E 2 , then its complement
S1 Ec also belongs to ;
– if fEn gn2N then nD1 En 2 .
2
In the case where ˝ is a finite set, the probability measure P W ! Œ0; 1 is assumed to satisfy
the following axioms:
1. for every E1 ; E2 2 such that E1 \ E2 D ; we have P.E1 [ E2 / D P.E1 / C P.E2 /;
2. P.˝/ D 1.
These axioms lead to some fundamental properties of the probability measure P:
– for every E 2 , it holds that P.E/ 1 and P.E/ C P.Ec / D 1;
– for every E1 ; E2 2 with E1 E2 , it holds that P.E1 / P.E2 / and P.E2 n E1 / D P.E2 /
P.E1 /;
– for every E1 ; E2 ; : : : ; En 2 , n 2 N, it holds that
P.E1 [ E2 [ : : : [ En / P.E1 / C P.E2 / C : : : C P.En /;
with equality holding for disjoint events ( finite additivity).

In the more general case where ˝ is not a finite set, then finite additivity has to be replaced by
-additivity: for any countable collection of disjoint events fEk gk2N it holds that
[
1 X1
P Ek D P.Ek /:
kD1 kD1
We refer to Chung [448] for a classical account of probability theory.

2 Choices Under Risk 17
setting with known probabilities, while the word uncertainty refers to a setting with
unknown probabilities. We will return to this distinction in Sect. 9.1.
The probability space .˝; ; P/ describes an agent’s knowledge/beliefs about the
world: it remains to describe his choices in a risky environment. Following Savage
[1499], we identify choices as real-valued random variables (or acts). An act is an
application xQ W ˝ ! R, where R (the set of real numbers) represents the space
of consequences. From the point of view of an agent, this application describes the
“consequence” xQ .!/ associated by the act xQ to the state of the world ! 2 ˝. In the
case of financial choices we shall generally think of the space of consequences R
(or RC ) as representing wealth (or returns).
We have to understand how an act introduces risk in the space of consequences of
an agent. Let B.R/ denote the Borel -algebra of R (i.e., the -algebra generated
by all the intervals in R). Given that an agent has chosen the act xQ , what is the
probability of its consequences belonging to a Borel set B 2 B.R/? Intuitively, this
corresponds to asking what is the probability of a set B of consequences associated
to a given act xQ of the agent. The natural answer is to evaluate the probability of B as
the probability of the event in representing the pre-image xQ 1 .B/ of B according
to the act xQ . As a consequence, the probability measure induced by xQ on .R; B.R//
is defined as .B/ WD P.Qx1 .B//. Note that this construction is possible only if for
every B 2 B.R/ it holds that xQ 1 .B/ 2 , i.e., only if the application xQ W ˝ ! R is
.; B.R//-measurable. In that case, xQ is said to be a random variable. Hence, given
the probability space .˝; ; P/, a random variable xQ induces a probability measure
on .R; B.R//. Each act-random variable induces a specific probability measure.
Given a probability space, in our analysis we will directly refer to acts as random
variables (or gambles) and to their probability measures on .R; B.R//.
Representing risk in this way is only apparently innocuous. Indeed, an agent does
not possess a complete knowledge of the future state of the world, but he is assumed
to be able to describe in an exhaustive way his ignorance status through a probability
space (and, therefore, he is assumed to know all possible future states of the world).
Furthermore, he is able to quantify his ignorance through a probability measure.
This means that the agent is able to quantify his confidence on the realization of
any event, where that quantity can be the outcome of objective and/or subjective
elements. According to our representation of choices in a risky environment, the
probability assumes an objective interpretation, in contrast with the subjectivist
approach proposed by De Finetti and Savage amongst others (see Chap. 9 for more
details).
In this chapter, we investigate decision making problems in a risky setting. More
specifically, we shall be concerned with the optimal choice of an agent among
several possible acts (or gambles-random variables). The most natural criterion to
evaluate a gamble-random variable xQ seems to be provided by its expected value
E ŒQx. However, this evaluation rule poses some problems, as illustrated by the
classical Saint Petersburg paradox proposed by Daniel Bernoulli in 1738. Let us
consider a game consisting of repeated tosses of a coin: if at the first toss a head
appears then the player receives the payoff 1 D 20 , otherwise the game continues
with another toss. If at that second toss a head appears then the player receives
the payoff 2 D 21 , otherwise the game continues with another toss, at which the
payoff in the case of a head appearing will be 22 , and so on. At the k-th toss, for
any k 2 N, if no head appeared on the first k 1 tosses, the payoff in the case of a
head appearing will be 2k1 . We suppose that the coin is fair, so that the probability
of a head appearing at any toss is equal to 1=2. Since the probability of no head
appearing in the k first tosses is 1=2k , for any k 2 N, the expected value of this game
is given by
1 k
X 1
1 1X
EŒQx D 2k1 D 1 D C1:
kD1
2 2 kD1
Hence, if an agent evaluates this gamble through its expectation, then the gamble
would appear infinitely attractive. In reality, when confronted with such a game, no
one would be prepared to pay an infinite amount of money in order to take part to the
gamble, the reason being that significant payoffs are only obtained with very small
probabilities. In order to resolve this apparent paradox, Daniel Bernoulli proposed
to evaluate gambles via the expected value of their logarithm, i.e., by computing
EŒlog.Qx/. In that case:
1 k
X 1
1 log 2 X k
EŒlog.Qx/ D log.2k1 / D D log 2 < C1:
kD1
2 2 kD1 2k
Therefore, if an agent evaluates a random variable by taking the expectation of

a logarithmic transformation, then the possibility of playing the game xQ and the
possibility of receiving the sure payoff 2 should be equally attractive. Equivalently,
to participate to the game an agent would be willing to pay up to 2 and not an
arbitrarily large amount of wealth as suggested by the expected value criterion.
In synthesis, Bernoulli proposed that an agent should evaluate a random variable
through the expected value of an increasing non-linear transformation of the random
variable. This hypothesis did not find a theoretical justification for a long time until
the expected utility theory was proposed by Von Neumann and Morgenstern in 1947,
as we are now going to explain.
This chapter is structured as follows. In Sect. 2.1, we prove that a preference
relation can be represented in terms of expected utility if it satisfies a certain set
of axioms. In Sect. 2.2, we introduce the notions of risk aversion and certainty
equivalent, while in Sect. 2.3 we study stochastic dominance criteria. Section 2.4
deals with mean-variance preferences and introduces the diversification and insur-
ance principles. At the end of the chapter, we provide a guide to further readings as
well as a series of exercises.
2.1 Expected Utility Theory 19
2.1 Expected Utility Theory
For simplicity, we consider the case of random variables with a finite number S 2 N
of values. Specifying a priori a finite set of values fx1 ; : : : ; xS g, with xi 2 R for all
i D 1; : : : ; S and xi < xiC1 for all i D 1; : : : ; S 1, we have a set M of random
variables (or gambles) xQ valued on the finite support fx1 ; : : : ; xS g. A random variable
xQ 2 M will be identified by a probability distribution f1 ; : : : ; S g associated with
the S possible realizations:
X
S
xQ 2 M ” xQ fx1 ; : : : ; xS I 1 ; : : : ; S g with s 0; 8 s D 1; : : : ; S; and s D 1:
sD1
(2.1)
In the risky environment under consideration, the set M becomes the new set of
choices instead of the set X considered in the previous chapter under certainty.
Therefore, from now on an agent will be identified by a pair .M ; R/, where the
preference relation R is defined over the gambles in M .
Introducing a probability space and a set of random variables allows to mathe-
matically represent the decision making problem of an agent in a risky environment.
In this setting, we would like to parallel the theory established under certainty, i.e.,
to represent the preference relation R through a function defined on the space of
random variables M . For clarity of presentation, we limit our analysis to a single
good which we generically interpret as wealth (the analysis can be easily extended
to multiple goods).
Due to the presence of risk, the optimal choice problem is more complex than
the one considered in the previous chapter. The approach we present below refers to
the axiomatic Von Neumann-Morgenstern expected utility theory, see Von Neumann
& Morgenstern [1632]. According to this approach, one first identifies a set of
hypotheses on the preference relation R which, if satisfied, allows to represent the
relation R through a function defined over M . Such a function will be a linear
combination of the utilities associated with wealth/consumption in each elementary
state of the world, where the coefficients of the linear combination are given by the
probabilities of realization of the different states.
In order to represent R by means of an expected utility function, we require R
to satisfy the following axioms (compare with Assumption 1.1).
Assumption 2.1 (Rationality) The preference relation R is rational if it satisfies
the following properties:
– (Reflexivity): for every xQ 2 M , it holds that xQ R xQ ;
– (Completeness): for every xQ 1 ; xQ 2 2 M , it holds that xQ 1 R xQ 2 or xQ 2 R xQ 1 ;
– (Transitivity): for every xQ 1 ; xQ 2 ; xQ 3 2 M such that xQ 1 R xQ 2 and xQ 2 R xQ 3 , it holds
that xQ 1 R xQ 3 .
The rationality hypothesis alone is not enough for our purposes. Two further
hypotheses are needed, one analogous to the continuity Assumption 1.2 and a crucial
one called the independence axiom.
Assumption 2.2 (Continuity) The preference relation R is continuous if, for
every xQ 1 ; xQ 2 ; xQ 3 2 M ; such that xQ 1 R xQ 2 and xQ 2 R xQ 3 , there exists a scalar ˛ 2 Œ0; 1
such that ˛ xQ 1 C .1 ˛/Qx3 I xQ 2 .
In Assumption 2.2, the random variable ˛ xQ 1 C .1 ˛/Qx3 represents the gamble
fx1 ; : : : ; xS I ˛ 11 C .1 ˛/13 ; : : : ; ˛ S1 C .1 ˛/S3 g, i.e., the gamble obtained as a
convex linear combination with weights ˛ and 1 ˛ of the probabilities associated
by the two gambles xQ 1 and xQ 2 to the values fx1 ; : : : ; xS g. This hypothesis allows us
to represent the order induced by R over M through a function from M to R. The
function will be of the expected utility form if the preference relation R satisfies the
following additional hypothesis.
Assumption 2.3 (Independence) The preference relation R satisfies the indepen-
dence assumption if, for all xQ 1 ; xQ 2 ; xQ 3 2 M and ˛ 2 Œ0; 1, we have that xQ 1 R xQ 2 if
and only if ˛ xQ 1 C .1 ˛/Qx3 R ˛ xQ 2 C .1 ˛/Qx3 .
As can be readily checked, if the preference relation R satisfies Assumption 2.3,
then also I and P do (this fact will be used in the proof of the next theorem).
Expected utility theory states that the preference relation R over the set of gambles
M can be represented through a function U W M ! R that assigns to a gamble
xQ D fx1 ; : : : ; xS I 1 ; : : : ; S g 2 M the value
X
S
U.Qx/ D s u.xs /; (2.2)
sD1
where u is the utility function that represents R for wealth obtained with certainty. In
other words, if R satisfies Assumptions 2.1–2.3, then there exists a utility function
u such that the preference relation R can be represented as in (2.2). Note that the
function u does not depend on the state s D 1; : : : ; S (state independent utility). The
function U is a linear combination of the utilities u.xs / that the agent obtains with
certainty in correspondence of the events s D 1; : : : ; S and the weights of the linear
combination are given by the probabilities s . The function U is called expected
utility function.
Theorem 2.1 Given the pair .M ; R/, if the preference relation R satisfies
Assumptions 2.1–2.3, then there exist S scalars u.xs / 2 R, s D 1; : : : ; S, such
that, for every xQ 1 ; xQ 2 2 M ,
xQ 1 R xQ 2 ” U.Qx1 / U.Qx2 /; (2.3)
where U./ is defined as in (2.2).

Proof Following Mas-Colell et al. [1310, Proposition 6.B.3], we divide the proof
into seven steps.
Step 1. If xQ R xQ 0 and ˛ 2 .0; 1/, then xQ R ˛Qx C .1 ˛/Qx 0 and ˛Qx C .1 ˛/Qx 0 R xQ 0 .
In view of Assumption 2.3, this claim can be readily verified. Indeed:
xQ D ˛ xQ C .1 ˛/Qx R ˛ xQ C .1 ˛/Qx 0 R ˛ xQ 0 C .1 ˛/Qx 0 D xQ 0 :
Note that, as can be easily checked, the same property also holds for P.
Step 2. There exist two gambles xQ ; xQ 2 M which represent respectively the
best and the worst gamble in M with respect to R, in the sense that xQ R xQ and
xQ R xQ for every xQ 2 M (see also Mas-Colell et al. [1310, Exercise 6.B.3]).
Indeed, without loss of generality, suppose that the preference rela-
tion R for wealth obtained with certainty is monotone, meaning that
xQ xS R xQ xS1 R : : : R xQ x1 , where xQ xs denotes the gamble that takes the value xs
with probability one, for s D 1; : : : ; S. Due to Assumptions 2.1–2.3 and applying
repeatedly the result established in the first step, one can show that xQ D xQ xS and
xQ D xQ x1 . When R is not monotone, xQ and xQ will be two gambles which assign
an amount of money different from x1 and xS with probability one.
Now, if xQ I xQ , then all gambles in M are indifferent for the agent and,
therefore, the claim of the theorem follows in a trivial way (it is enough to choose
a constant utility function). So, let us assume that xQ P xQ . We now have to show
that there exists a function F W M ! R which represents R and is a linear
function with weights defined by the probabilities s of the S events, thus yielding
the expected utility function U.
Step 3. For every ˛; ˇ 2 Œ0; 1 the following holds:
ˇ xQ C .1 ˇ/Qx P ˛ xQ C .1 ˛/Qx ” ˇ > ˛:
Indeed, by step 1 (for the strict preference relation P) we know that
xQ P ˛Qx C .1 ˛/Qx
and that, for any 2 .0; 1/,
xQ C .1 /.˛ xQ C .1 ˛/Qx / P ˛ xQ C .1 ˛/Qx :
To prove the ( implication, it is enough to choose D .ˇ ˛/=.1 ˛/, with

ˇ > ˛. To show the converse implication, suppose that ˇ ˛. In the case ˇ D ˛
we have that ˇ xQ C .1 ˇ/Qx I ˛ xQ C .1 ˛/Qx and, therefore, a contradiction
is obtained. If ˇ < ˛, then a contradiction with the hypothesis is easily reached
by relying on the same arguments used above.
Step 4. For every xQ 2 M , there exists a unique ˛xQ 2 Œ0; 1 such that
˛xQ xQ C .1 ˛xQ /Qx I xQ :

Indeed, the existence comes from Assumption 2.2, while the uniqueness can be
easily deduced from Step 3.
Step 5. The function F W M ! R defined as F.Qx/ D ˛xQ represents the relation
R in the sense of (2.3), where ˛xQ 2 Œ0; 1 is as in Step 4.
Indeed, given two gambles xQ ; xQ 0 2 M , Step 4 implies that:
xQ R xQ 0 ” ˛xQ xQ C .1 ˛xQ /Qx R ˛xQ 0 xQ C .1 ˛xQ 0 /Qx :
The claim then follows from the double implication established in Step 3.
Step 6. The function F is linear: for every xQ ; xQ 0 2 M and ˇ 2 Œ0; 1, it holds that

F ˇQx C .1 ˇ/Qx 0 D ˇF.Qx/ C .1 ˇ/F.Qx 0 /:
Indeed, by the definition of F, we have

xQ I F.Qx/Qx C 1 F.Qx/ xQ and xQ 0 I F.Qx 0 /Qx C 1 F.Qx 0 / xQ :
By Assumption 2.3 (applied to the indifference relation I ), we obtain

ˇQx C .1 ˇ/Qx 0 I ˇ F.Qx/Qx C 1 F.Qx/ xQ C .1 ˇ/Qx 0

I ˇ F.Qx/Qx C 1 F.Qx/ xQ C .1 ˇ/ F.Qx 0 /Qx C 1 F.Qx 0 / xQ :
Rearranging the terms we obtain

ˇQx C .1 ˇ/Qx 0 I ˇF.Qx/ C .1 ˇ/F.Qx 0 / xQ C 1 ˇF.Qx/ .1 ˇ/F.Qx 0 / xQ :
By the definition of the function F in Step 5, we thus get

F ˇQx C .1 ˇ/Qx 0 D ˇF.Qx/ C .1 ˇ/F.Qx 0 /:
Step 7. The function F coincides with the function U defined in (2.2).

Indeed, let us define u.xs / WD F.Qxxs /, for all s D 1; : : : ; S. Every gamble xQ 2 M ,
identified through the vector of probabilities fP 1 ; : : : ; S g associated with the
realizations fx1 ; : : : ; xS g, can be rewritten as xQ D SsD1 s xQ xs . Using the linearity
of the function F established in Step 6, this allows us to write
X
S X
S
F.Qx/ D F s xQ xs D s u.xs /:
sD1 sD1
We have thus shown that the function F coincides with the expected utility
function U introduced in (2.2), thus proving the theorem. t
u
Note that, given a preference relation R satisfying the assumptions introduced

in the present section, the expected utility function U representing R is unique up
to strictly increasing linear transformations (see Exercise 2.1). If a gamble-random
variable xQ takes an infinite number of values in R, letting fxQ be its probability density
function, then the expected utility function can be written as follows, extending
representation (2.2),
Z
U.Qx/ D u.z/fxQ .z/dz D E Œu.Qx/ ;
R
provided that the integral exists. In the following, xQ will denote a gamble-random
variable and U.Qx/ and EŒu.Qx/ will equivalently denote the expected utility function.
In the case of general probability spaces, the preference relation R admits a
representation through an expected utility function if Assumptions 2.1–2.3 hold
together with an additional assumption named the sure thing principle (if a gamble
is concentrated on a subset A of R and each amount of money in A is preferred to
another gamble, then the first gamble is preferred to the second) as well as other
hypotheses of technical nature (see Fishburn [706], Kreps [1136]).
The representation of the preference relation R in a risky environment via an
expected utility function can also be established in the case where the agent is
allowed to consume in correspondence of two dates t 2 f0; 1g. As explained at the
beginning of this chapter, we assume that the state of the world is fully revealed only
at t D 1, thus introducing a risk for an agent who has to take decisions at t D 0.
In the case of a finite probability space, as in (2.1), the expected utility function
U W R M ! R takes the form
X
S
U.x0 ; xQ / D s u.x0 ; xs /; for all x0 2 R and xQ 2 M ;
sD1
where the first argument x0 denotes consumption at the initial date t D 0 and xQ 2 M
is the gamble which will assume one of the realizations fx1 ; : : : ; xS g at the future
date t D 1, representing consumption at t D 1. The expected utility is said to be
additively separable over time if, for each t 2 f0; 1g, the utility at date t depends
only on consumption/wealth at t, i.e.,
X
S
U.x0 ; xQ / D u0 .x0 / C s u1 .xs /; for all x0 2 R and xQ 2 M :
sD1
A particular case is the time invariant utility, where u1 ./ D ıu0 ./:
X
S
U.x0 ; xQ / D u.x0 / C ı s u.xs /; for all x0 2 R and xQ 2 M ;
sD1
for some utility function u W R ! R and 0 < ı 1. The term ı is a discount factor
representing the natural fact that an agent prefers to consume today (at t D 0) rather
than in the future (at t D 1).
The expected utility framework and the probability space represent two funda-
mental ingredients of our analysis. A discussion of their role is postponed to Chap. 9,
after having acquired a complete picture of the results obtained assuming that the
agents’ preferences can be represented by means of expected utility functions.
2.2 Risk Aversion
In the setting considered in this chapter, an agent faces an optimal choice problem
in a risky environment: loosely speaking, this means that an agent will not only
consider the utility associated to given amounts of wealth, but he will also take into
account the probabilities of receiving those amounts of wealth.
An agent’s attitude towards risk is described by his preference relation R on the
space of random variables M . If R can be represented through an expected utility
function U, then the attitude towards risk can be analysed through the properties of
the function u, where u is the utility function representing the preference relation R
with respect to wealth obtained with certainty, as in Theorem 2.1. In the following,
we shall always assume that the preference relation R satisfies Assumptions 2.1–
2.3.
The propensity-aversion of an agent towards risk can be evaluated by considering
actuarially fair gambles-random variables. We say that a gamble-random variable xQ
is actuarially fair if EŒQx D 0. In the case of a finite probability P space, as in (2.1), a
gamble xQ D fx1 ; : : : ; xS I 1 ; : : : ; S g is actuarially fair if SsD1 xs s D 0.
We say that an agent is risk averse if he does not accept (or is at most indifferent)
to any actuarially fair gamble at all wealth levels, while an agent is risk neutral if
he is indifferent to any actuarially fair gamble and is risk lover if he accepts any
actuarially fair gamble, for every level of wealth. We shall also say that an agent
is strictly risk averse if he rejects any actuarially fair gamble for every level of
wealth. Note that these definitions are “global” in the sense that they refer to an
indeterminate level of wealth and to any actuarially fair gamble. In our analysis,
we shall mostly restrict our attention to risk neutral and risk averse agents, due to
the fact that risk loving agents are considered to be a negligible minority whose
behavior does not seem to significantly affect financial markets.
In order to illustrate the definition of risk aversion, suppose that the wealth of
an agent is represented by a random variable xQ with expected value EŒQx. If the
agent is risk averse, then he will prefer to obtain the sure amount EŒQx with certainty
rather than being exposed to the random amount of wealth xQ . Equivalently, a risk
averse agent will be willing to pay a positive price to move from a random variable
xQ describing his wealth to the expected value EŒQx of that random variable obtained
with certainty. Such a price is called the risk premium.
2.2 Risk Aversion 25
Definition 2.2 The risk premium u .Qx/ of a gamble xQ for an agent characterized by
a utility function u is the maximum amount that the agent is willing to pay in order
to receive, instead of the random variable xQ , its expected value EŒQx with certainty,
i.e., u .Qx/ is such that

u EŒQx u .Qx/ D E Œu.Qx/ : (2.4)
According to Definition 2.2, for a risk averse agent characterized by the utility
function u and exposed to the gamble xQ , paying the price u .Qx/ and receiving the
sure amount EŒQx is equally attractive to receiving the random realization xQ of the
gamble. Note that, in this representation, the payoff of the gamble xQ includes the
initial wealth of the agent. The definition of risk premium directly leads to the
concept of certainty equivalent CEu .Qx/ of the gamble xQ for an agent with utility
function u.
Definition 2.3 The certainty equivalent of the gamble xQ for an agent characterized
by the utility function u is the amount of wealth CEu .Qx/ which makes the agent
indifferent to the gamble, i.e., CEu .Qx/ is such that

u CEu .Qx/ D E Œu.Qx/ :
Note that the certainty equivalent and the risk premium of a gamble are unique if
the function u is strictly increasing. In this case, exploiting the two above definitions
we get the equality
CEu .Qx/ D EŒQx u .Qx/:
If the risk premium u .Qx/ can be interpreted as the maximum price that an agent is
willing to pay in order to receive the sure amount EŒQx instead of the gamble xQ , then
the certainty equivalent CEu .Qx/ admits the interpretation of the smallest amount of
wealth required by the agent to give up the gamble xQ and accept a sure amount of
wealth.
Exploiting concavity and Jensen’s inequality,3 we can establish that an agent is
(strictly) risk averse if and only if u is (strictly) concave. More precisely, we have
the following proposition (see Fig. 2.1 for a graphical illustration).
Proposition 2.4 Given an agent characterized by an increasing utility function u,
the following conditions are equivalent:
(i) the agent is risk averse;
(ii) the utility function u is concave;
(iii) CEu .Qx/ EŒQx, for every random variable xQ with EŒjQxj < 1;
(iv) u .Qx/ 0, for every random variable xQ with EŒjQxj < 1.
3
Let xQ be a random variable with finite expected value EŒQx. Jensen’s inequality establishes that,
for any concave function g W R ! R, it holds that E Œg.Qx/ g.EŒQx/.
Fig. 2.1 Expected utility and certainty equivalent, fx ; x C I 1=2; 1=2g
Proof .i/ ) .ii/: Let us consider a risk averse agent characterized by the utility
function u. For any w; w0 2 R and 2 .0; 1/, define wN WD w C .1 /w0 and
consider the gamble Q defined as follows:
(
.1 /.w w0 / with probability I
Q D
.w0 w/ with probability 1 :
Since EŒQ D .1 /.w w0 / C .1 /.w0 w/ D 0, the gamble Q is actuarially

fair. Hence, due to the definition of risk aversion,

u w C .1 /w0 D u.w/
N EŒu.wN C Q / D u.w/ C .1 /u.w0 /:
Since this holds for every w; w0 2 R and 2 .0; 1/, this implies that the utility
function u W R ! R is concave.
.ii/ ) .i/: Suppose that u is concave. Then, for any w 2 R and for any gamble Q
such that EŒQ D 0, Jensen’s inequality implies that

EŒu.w C Q / u w C EŒQ D u.w/;
thus showing that the agent is risk averse, since he does not accept (or is at most
indifferent to) any actuarially fair gamble.
.ii/ , .iii/ , .iv/: Since the function u is assumed to be increasing, these
equivalences follow directly from Definitions 2.2–2.3. t
u
Note that the equivalence between the first two claims of Proposition 2.4 also
holds for a non-increasing utility function u.
As can be seen from Definitions 2.2–2.3, the risk premium and the certainty
equivalent of a gamble xQ are defined with respect to a specific utility function u.
We now intend to put the risk premium u .Qx/ in relation with the random variable
xQ and the utility function u. To this end, we consider two particular cases: a gamble
described by a random variable with additive noise and a gamble described by a
random variable with multiplicative noise. This will lead to the introduction of the
coefficients of absolute and relative risk aversion.
Let xQ D x C Q ; where Q is a random variable with zero mean and variance 2 ,
so that EŒQx D x. We assume that the function u is twice differentiable and that the
noise component Q is “small” with respect to x. By Definition 2.2, we have that

u x u .Qx/ D U.Qx/: (2.5)
For every realization of Q , we can approximate u.x C / by a second order Taylor

expansion of the function u centered in x, thus obtaining
2 00
u.x C / u.x/ C u0 .x/ C u .x/:
2
Using this expression to evaluate the expected utility, we obtain the approximation
2 00
U.Qx/ D EŒu.x C Q / u.x/ C u .x/: (2.6)
2
If the noise component Q is “small” with respect to x, then also the risk premium
u .Qx/ will be “small” with respect to x and, therefore, we can use a Taylor expansion
(centered in x and up to the first order) of the utility function u evaluated at the
certainty equivalent CEu .Qx/ D x u .Qx/, thus obtaining

u x u .Qx/ u.x/ u .Qx/u0 .x/: (2.7)
Using the two approximations (2.6)–(2.7) and assuming that the function u is strictly
increasing, we obtain the following estimate for the risk premium u .Qx/:
1 u00 .x/ 2
u .Qx/ ; (2.8)
2 u0 .x/
which can be rewritten as u .Qx/ 12 rua .x/ 2 , where we define
u00 .x/
rua .x/ WD :
u0 .x/
The quantity rua .x/ is called the coefficient of absolute risk aversion of the utility
function u in correspondence of wealth x. In view of the approximation (2.8), the
risk premium is decomposed into two factors: the variance of the gamble and the
absolute risk aversion coefficient. Note that the risk premium is increasing in both
factors. The reciprocal of the absolute risk aversion coefficient is called the absolute
risk tolerance and is defined as (provided that u00 > 0)
1 u0 .x/
tu .x/ WD D :
rua .x/ u00 .x/
Let us now consider the case where xQ D x.1 C Q /, where Q is a random variable
with zero mean and variance 2 . We assume that u is twice differentiable and strictly
increasing. By relying on a reasoning similar to the one employed above, we can
provide the following estimate of the relative risk premium ur .Qx/ defined by the
relation xur .Qx/ WD u .Qx/:
1 u00 .x/ 2
ur .Qx/ x ;
2 u0 .x/
which can be rewritten as ur .Qx/ 12 rur .x/ 2 ; where
u00 .x/
rur .x/ WD x
u0 .x/
is the coefficient of relative risk aversion of the utility function u in correspondence

of the wealth x. By definition, it holds that rur .x/ D rua .x/x, for every x 2 R.
The following classification of the utility function u can be given according
to the behavior of the coefficient of absolute risk aversion with respect to wealth
changes:
• u is characterized by decreasing absolute risk aversion (DARA) if x 7! rua .x/ is a
decreasing function;
• u is characterized by constant absolute risk aversion (CARA) if x 7! rua .x/ is a
constant function;
• u is characterized by increasing absolute risk aversion (IARA) if x 7! rua .x/ is an
increasing function.
Of course, a similar classification can be given on the basis of the behavior of the
coefficient of relative risk aversion with respect to wealth changes.
Risk Aversion Comparison
Having characterized the risk aversion of an agent in terms of the properties of his
utility function u, we now aim at comparing the behavior of different agents with
respect to risky choices.
Let us consider two agents a and b characterized by utility functions ua and ub ,
respectively, and the same initial wealth. We say that agent a is more risk averse than
agent b if agent b always accepts a gamble if agent a does. Equivalently (assuming
that the utility functions ua and ub are increasing), we can affirm that, for each
gamble xQ , agent a is more risk averse than agent b if and only if the risk premium
ua .Qx/ of agent a is greater or equal than the risk premium ub .Qx/ of agent b. As far
as the certainty equivalent is concerned, the opposite holds, due to the equivalence
between claims .iii/ and .iv/ in Proposition 2.4. Intuitively, the fact that agent a
is more risk averse than agent b amounts to saying that agent a requires a higher
compensation than agent b in order to accept the riskiness of a gamble xQ .
The following result, first established in De Finetti [537], Pratt [1432], Arrow
[78], gives three equivalent conditions for agent a to be more risk averse than agent
b (for simplicity, we limit our attention to twice differentiable utility functions).
Proposition 2.5 Given two strictly increasing and strictly concave utility functions
ua and ub , the following conditions are equivalent:
(i) ruaa .x/ ruab .x/, for every x 2 R;
(ii) there exists an increasing and concave function g W R ! R such that, for every
x 2 R, it holds that ua .x/ D g.ub .x//;
(iii) ua is more risk averse than ub , i.e., ua .x C Q / ub .x C Q /, for every x 2 R
and for every random variable Q such that EŒQ D 0.
Proof .i/ , .ii/: Since ua and ub are both increasing, concave and twice
differentiable, there exists a twice differentiable increasing function g W R ! R
such that ua .x/ D g.ub .x//, for all x 2 R. Indeed, let g WD ua ı .ub /1 , which can be
verified to be increasing and twice differentiable. Differentiating ua once and twice
we obtain
0 0 00 00 0 2
ua .x/ D g0 ub .x/ ub .x/ and ua .x/ D g0 ub .x/ ub .x/ C g00 ub .x/ ub .x/ :
0
Dividing the second expression by ua .x/ and using the relation between ua and ub ,
we obtain, for every x 2 R,

g00 ub .x/ b0
ruaa .x/ D ruab .x/ 0 b u .x/: (2.9)
g u .x/
Therefore, ruaa .x/ ruab .x/ ” g00 .ub .x// 0, for every x 2 R.
.ii/ ) .iii/: Suppose that agents a and b are endowed with the same wealth
x 2 R. For an arbitrary random variable Q with EŒQ D 0 we have that

ua x ua .x C Q / D EŒua .x C Q / D E g ub .x C Q / g U b .x C Q /

D g ub .x ub .x C Q // D ua x ub .x C Q / ;
(2.10)
where the inequality is due to Jensen’s inequality, since g is concave. Since ua is
increasing, we then have ua .x C Q / ub .x C Q /.
.iii/ ) .ii/: As in the first part of the proof, since ua and ub are strictly increasing,
there exists an increasing function g W R ! R such that ua .x/ D g.ub .x//. We shall
now prove that the function g is concave (in the range of ub ). Similarly as in the
proof of Proposition 2.4, let x; x0 2 R and 2 Œ0; 1, define xN WD x C .1 /x0 and
consider the gamble Q defined as follows:
(
.1 /.x x0 / with probability I
Q D
0
.x x/ with probability 1 :
Note that EŒQ D 0. Then, by the definition of risk premium,

ub .x/ C .1 /ub .x0 / D EŒub .Nx C Q / D ub xN ub .Nx C Q / ;
so that, since ua .x/ D g.ub .x//,

g ub .x/ C .1 /ub .x0 / D ua xN ub .Nx C Q / : (2.11)
On the other hand,

g ub .x/ C .1 /g ub .x0 / D ua .x/ C .1 /ua .x0 / D EŒua .Nx C Q /

D ua xN ua .Nx C Q / :
(2.12)
By (2.11)–(2.12), together with the increasingness of ua and property .iii/, we get

g ub .x/ C .1 /ub .x0 / g ub .x/ C .1 /g ub .x0 / :
Since x; x0 2 R and 2 Œ0; 1 are arbitrary, this proves the concavity of g. t

u
The above proposition bridges the gap between the coefficient of absolute risk
aversion, which is a local measure of risk aversion (being computed in relation to a
specific level of wealth), and a global characterization of risk aversion (which refers
to an indeterminate level of wealth, as explained at the beginning of this section).
If the coefficient of absolute risk aversion of an agent is larger than the coefficient
of another agent for every level of wealth, then the first agent is more risk averse
than the second agent. Note that, on the basis of this result, u./ and a C bu./ (with
b > 0) are equivalent as far as risk aversion is concerned, since the coefficient of
absolute risk aversion is invariant under positive linear transformations.
Proposition 2.5 can also be used to analyse the attitude of an agent towards risk
with respect to changes in wealth, as shown in the next result.
Proposition 2.6 If an agent is characterized by an increasing DARA utility function
u and Q is a random variable such that EŒQ D 0, then the risk premium u .xC Q / is a
decreasing function of x, i.e., the agent becomes less risk averse as wealth increases.
Proof We need to show that, for x0 x, we have
u .x C Q / u .x0 C Q /:
Set x0 D x C k, with k WD x0 x 0, and define the utility functions ua and ub by

ua .x/ WD u.x/ and ub .x/ WD u.k C x/, respectively, for x 2 R. Since u is DARA, we
have that
ruaa .x/ ruab .x/:
As a consequence, since x 2 R is arbitrary, using the implication .i/ ) .iii/ of

Proposition 2.5, it holds that
u .x C Q / D ua .x C Q / ub .x C Q / D u .x0 C Q /;
thus proving the claim. t

u
In other words, Proposition 2.6 says that, for a risk averse agent characterized
by an increasing DARA utility function, any desirable gamble cannot become
undesirable as wealth increases. Equivalently, in view of Proposition 2.4, the
certainty equivalent of a gamble is increasing with respect to wealth. In view of
this result, a DARA utility function seems a reasonable assumption to represent an
agent’s behavior with respect to risk.
Classes of Utility Functions
Four classes of utility functions play a particularly important role. Further properties
of these utility functions will be derived in the exercises proposed at the end of the
chapter.
Quadratic Utility Function
b
u.x/ D x x2 ; with b > 0:
2
The function x 7! u.x/ is increasing only in the domain Œ0; 1=b. For x 2 Œ0; 1=b/,
the coefficient of absolute risk aversion rua is positive and is given by
b
rua .x/ D :
1 bx
0 2
Since rua .x/ D .1bx/
b
2 > 0, for x ¤ 1=b, the function x 7! ru .x/ is increasing
a
in wealth, meaning that the quadratic utility function is IARA. The coefficient of
0
relative risk aversion rur .x/ D 1bx
bx
is increasing, i.e., rur .x/ D .1bx/
b
2 > 0.
Exponential Utility Function
1
u.x/ D eax ; with a > 0:
a
The coefficient of absolute risk aversion rua is positive and is given by
rua .x/ D a:
Therefore, the exponential utility function is CARA and the coefficient of relative
risk aversion rur .x/ D ax is increasing.
Power Utility Function
b 1 1
u.x/ D x b; with b > 0:
b1
The coefficient of absolute risk aversion rua is positive and is given by
1
rua .x/ D :
bx
0
Since rua .x/ D bx12 < 0, the map x 7! rua .x/ is decreasing, meaning that the
power utility function is DARA. Moreover, rur .x/ D 1=b and, hence, the coefficient
of relative risk aversion is constant (CRRA utility function).
Logarithmic Utility Function
u.x/ D log.bx/; with b > 0:
The coefficient of absolute risk aversion rua is positive and given by
1
rua .x/ D :
x
Clearly, rua .x/ is decreasing in wealth, so that the logarithmic utility function is
DARA. Moreover, rur .x/ D 1 and, hence, the coefficient of relative risk aversion
is constant (CRRA utility function). Note that, as shown in Exercise 2.2, the
logarithmic utility function can be obtained as the limit of the power utility function
as b ! 1.
In our analysis, we will often consider the large class of utility functions with
hyperbolic absolute risk aversion (HARA), i.e., the class of utility functions such
that the absolute risk tolerance tu is linear with respect to wealth:
tu .x/ D a C bx; for some a; b 2 R:
The HARA class embeds generalized versions of the above utility functions.
Indeed, the following types of utility functions can be obtained (see Fig. 2.2 and
Exercise 2.10):
1. exponential utility function u.x/ D aex=a , for a > 0 and b D 0;
b1
2. generalized power utility function u.x/ D b1b
.a C bx/ b , for a 2 R and b > 0
with b ¤ 1;
3. generalized logarithmic utility function u.x/ D log.xCa/, for x > a and b D 1.
u(x)
b
b>1 : b−1 (a + b x)(b−1)/b
b = 1 : log(x + a)
b
0<b<1 : b−1 (a + b x)(b−1)/b
b = 0 : −a e−x/a
Fig. 2.2 HARA utility functions

2.3 Stochastic Dominance
Stochastic dominance criteria introduce an order among gambles-random variables

which is shared by all agents characterized by utility functions satisfying some
common properties. The interesting feature of stochastic dominance criteria is that
they exclusively refer to some basic features of the probability distribution of the
gamble-random variables, without referring to specific preference relations or utility
functions. In this section, we present three different stochastic dominance criteria:
first order stochastic dominance, second order stochastic dominance and second
order stochastic monotonic dominance. The mean-variance criterion, partly related
to stochastic dominance, will be discussed in Sect. 2.4.
In a nutshell, we aim at answering the following question: is it possible to rank
gambles-random variables through the order induced by expected utility just by
assuming that the utility function satisfies some basic properties?
The first criterion that we present is the first order stochastic dominance.
Definition 2.7 Let xQ 1 and xQ 2 be two random variables. We say that xQ 1 dominates xQ 2
according to the first order stochastic dominance criterion (written as xQ 1
FSD xQ 2 ) if
U.Qx1 / U.Qx2 / holds for every non-decreasing utility function u W R ! R.
For simplicity, we consider (normalized) random variables taking values in the
interval Œ0; 1. In the following proposition, the first order stochastic dominance
criterion is characterized in terms of the properties of the two gambles-random
variables xQ 1 and xQ 2 . We denote by Fi ./ the cumulative distribution function of
the random variable xQ i , for i D 1; 2. Note that Fi .1/ D 1, while Fi .0/ 0
(meaning that the random variable xQ i is allowed to take value zero with strictly
positive probability).
Proposition 2.8 For any two random variables xQ 1 and xQ 2 taking values in Œ0; 1, the
following are equivalent:
(i) xQ 1
FSD xQ 2 ;
(ii) F1 .x/ F2 .x/, for every x 2 Œ0; 1;
(iii) xQ 1 d xQ 02 C Q , where xQ 02 is a random variable such that xQ 2 d xQ 02 and Q is a
positive random variable.4
Proof .i/ ) .ii/: Arguing by contradiction, suppose that there exists x 2 Œ0; 1 such
that F1 .x/ > F2 .x/. Let us consider the non-decreasing function u W Œ0; 1 ! f0; 1g
defined as follows, for all x 2 Œ0; 1:
(
1 for x > xI
u.x/ WD
0 for x x:
4
The notation d stands for equality in law (or in distribution).
2.3 Stochastic Dominance 35
Let U.Qxi / D EŒu.Qxi / be the expected utility associated with the random variable xQ i ,
for i D 1; 2. Evaluating the difference U.Qx1 /U.Qx2 / of the expected utility functions
of the two random variables xQ 1 and xQ 2 , we get
U.Qx1 /U.Qx2 / D EŒu.Qx1 /EŒu.Qx2 / D P.Qx1 > x/P.Qx2 > x/ D F2 .x/F1 .x/ < 0:
This contradicts the assumption that xQ 1

FSD xQ 2 , thus proving that .i/ ) .ii/.
.ii/ ) .iii/: Let uQ be a uniformly distributed random variable on Œ0; 1, indepen-
dent of xQ 1 and xQ 2 (such a random variable always exists, up to an enlargement of the
original probability space), and define the auxiliary random variables xQ 01 and xQ 02 as
xQ 01 WD inffy 2 Œ0; 1 W F1 .y/ uQ g and xQ 02 WD inffy 2 Œ0; 1 W F2 .y/ uQ g:
Since F1 .x/ F2 .x/ for every x 2 Œ0; 1, it holds that P.Qx01 xQ 02 / D 1. Moreover,
denoting by Fi0 the distribution function of xQ 0i , for i D 1; 2, it holds that, for all
x 2 Œ0; 1,

Fi0 .x/ D P.Qx0i x/ D P Fi .x/ uQ D Fi .x/;
thus showing that xQ 0i d xQ i , for i D 1; 2. The claim then follows by letting Q WD

xQ 01 xQ 02 , so that xQ 1 d xQ 01 D xQ 02 C Q and xQ 02 d xQ 2 .
.iii/ ) .i/: If xQ 1 d xQ 02 C Q , where xQ 02 d xQ 2 and Q is a positive random variable,
then for every non-decreasing utility function u we have that
U.Qx1 / D EŒu.Qx02 C Q / EŒu.Qx02 / D U.Qx2 /;
thus proving the implication. t

u
This first order stochastic dominance criterion is rather weak. It only induces a
partial order on gambles-random variables. Furthermore, since the class of utility
functions considered is very large, the order induced among random variables
simply establishes that a gamble-random variable xQ 1 is preferred to a gamble-
random variable xQ 2 if and only if the distribution function of the first random variable
is always smaller than the one of the second random variable. Note that
xQ 1
FSD xQ 2 ) EŒQx1 EŒQx2 :
However, the converse implication does not necessarily hold (a counterexample is

provided in Exercise 2.19).
A different order among random variables can be established considering risk
averse agents, thus leading to the second order stochastic dominance criterion.
Definition 2.9 Let xQ 1 and xQ 2 be two random variables. We say that xQ 1 dominates xQ 2
according to the second order stochastic dominance criterion (written as xQ 1
SSD xQ 2 )
if U.Qx1 / U.Qx2 / holds for every concave utility function u W R ! R.
Similarly to Proposition 2.8, the following result characterizes the second order
stochastic dominance criterion between two random variables xQ 1 and xQ 2 in terms
of the properties of their probability distributions (for simplicity, in the following
proposition we restrict our attention to continuously differentiable utility functions).
Proposition 2.10 For any two random variables xQ 1 and xQ 2 taking values in Œ0; 1,
the following are equivalent:
(i) xQ 1
SSD xQ 2 ; Ry
(ii) EŒQx1 D EŒQx2 and G.y/ WD 0 .F1 .z/ F2 .z//dz 0, for every y 2 Œ0; 1;
(iii) xQ 2 d xQ 1 C Q , where Q is a random variable such that EŒQ jQx1 D 0.
Proof .i/ ) .ii/: The function u W R ! R defined by u.x/ WD x, for x 2 Œ0; 1, is of
course concave. Hence, the assumption xQ 1
SSD xQ 2 implies that
EŒQx1 EŒQx2 D U.Qx1 / U.Qx2 / 0:
Applying the same reasoning to the function x 7! x yields the reverse equality,
thus showing that EŒQx1 D EŒQx2 . Recall that, for any random variable xQ with
distribution function FxQ W Œ0; 1 ! Œ0; 1, the expected value can be computed as
Z 1
EŒQx D 1 FxQ .z/ dz:
0
Together with the equality EŒQx1 D EŒQx2 , this implies that G.0/ D G.1/ D 0. Let us
consider any twice differentiable concave utility function u W R ! R. Then, using
twice the integration by parts formula and the assumption that xQ 1
SSD xQ 2 , we obtain
Z

0 U.Qx1 / U.Qx2 / D u.z/d F1 .z/ F2 .z/
Œ0;1
Z

D u.1/ F1 .1/ F2 .1/ u0 .z/ F1 .z/ F2 .z/ dz
.0;1
Z Z
0 0 00
D u .1/G.1/ C u .0/G.0/ C G.z/ u .z/dz D G.z/ u00 .z/dz:
.0;1 .0;1
(2.13)
Since the function u is arbitrary and concave, meaning that u00 .z/ 0 for every
z 2 Œ0; 1, this implies that G.z/ 0 for all z 2 Œ0; 1.
.ii/ ) .iii/: The proof of this implication is more technical and we refer the
reader to the original paper Rothschild & Stiglitz [1472], where the implication
is proved first for discrete random variables and then extended to general random
variables.
.iii/ ) .i/: For any concave function u, Jensen’s inequality (applied to the
conditional expectation with respect to the random variable xQ 1 ) together with the
2.3 Stochastic Dominance 37
tower property of the conditional expectation implies that

U.Qx2 / D EŒu.Qx1 C Q / D E EŒu.Qx1 C Q /jQx1 E u EŒQx1 C Q jQx1 D EŒu.Qx1 /;
thus proving the implication. t

u
We want to emphasize that Proposition 2.10 makes no assumptions on the
monotonicity of the utility function. Property .iii/ is particularly interesting, since
it implies that if a gamble xQ 1 is preferred to another gamble xQ 2 by every risk averse
agent then the two gambles have the same expected value EŒQx1 D EŒQx2 while the
first gamble has a smaller variance. Indeed, if xQ 1
SSD xQ 2 , then property .iii/ implies
that
2 .Qx2 / D 2 .Qx1 C Q / D 2 .Qx1 / C 2 .Q / C 2 Cov.Qx1 ; Q / 2 .Qx1 /;
where we have used the fact that

Cov.Qx1 ; Q / D EŒQx1 Q EŒQx1 EŒQ D E xQ 1 EŒQ jQx1 D 0:
Summing up, we have shown that:
xQ 1
SSD xQ 2 ) EŒQx1 D EŒQx2 and 2 .Qx1 / 2 .Qx2 /:
However, the converse implication does not hold. A counterexample can be easily
constructed. Indeed, let us consider the two gambles
xQ 1 D f0; 4; 8; 12I 1=4; 1=4; 1=4; 1=4g and xQ 2 D f0:2; 6; 11:8I 1=3; 1=3; 1=3g:
Elementary computations give us that
EŒQx1 D 6; 2 .Qx1 / D 20 and EŒQx2 D 6; 2 .Qx2 / 22:43:
Consider the concave utility function

(
2x for x 2 Œ0; 6I
u.x/ D
6Cx for x > 6:
Then, one can compute
EŒu.Qx1 / D 10 and EŒu.Qx2 / 10:07:
Even though the two gambles xQ 1 and xQ 2 have the same expectation and the variance
of xQ 1 is lower than the variance of xQ 2 , an agent with the above utility function will
prefer gamble xQ 2 to gamble xQ 1 . Note that the order induced by the criterion of second
order stochastic dominance is obviously partial, since it only allows to compare

gambles-random variables with the same expected value.
As already remarked, the second order stochastic dominance criterion is defined
without any hypothesis on the monotonicity of the utility functions. The second
order stochastic monotonic dominance criterion is obtained by restricting the class
of utility functions to non-decreasing concave functions.
Definition 2.11 Let xQ 1 and xQ 2 be two random variables. We say that xQ 1 dominates
xQ 2 according to the second order stochastic monotonic dominance criterion (written
as xQ 1
M Q 2 ) if U.Qx1 / U.Qx2 / holds for every non-decreasing concave utility
SSD x
function u W R ! R.
In the spirit of Proposition 2.10, the following result gives three equivalent
characterizations of second order stochastic monotonic dominance (we omit the
proof, which is similar to those of the previous propositions in this section).
(i) xQ 1
M
SSD Rx Q 2 ;
y
(ii) G.y/ D 0 F1 .z/ F2 .z/ dz 0, for every y 2 Œ0; 1;
(iii) xQ 2 d xQ 1 C Q , where Q is a random variable such that EŒQ jQx1 0;
(iv) xQ 2 d xQ 1 C
Q C v,
Q where
Q is a non-positive random variable and vQ is a random
variable such that EŒvQ jQx1 C
Q D 0.
Note that part (ii) of Proposition 2.12 implies that EŒQx1 EŒQx2 . Indeed, since
the random variables xQ 1 and xQ 2 take values in Œ0; 1, this simply follows from the fact
that
Z 1 Z 1

EŒQx1 EŒQx2 D 1 F1 .z/ dz 1 F2 .z/ dz D G.1/ 0:
0 0
By definition, second order stochastic dominance implies second order stochastic

monotonic dominance, in the sense that, if a random variable xQ 1 dominates another
random variable xQ 2 according to
SSD , then xQ 1 also dominates xQ 2 according to
M
SSD .
2.4 Mean-Variance Analysis
As we have seen at the beginning of this chapter by means of the Saint Petersburg
paradox, the expected value fails to provide an adequate representation of a
preference relation in a risky environment. In order to resolve the paradox, expected
utility theory has been introduced under the assumption that the preference relation
R satisfies Assumptions 2.1–2.3. Evaluating a gamble-random variable xQ via the
expected utility paradigm consists in taking the expected value EŒu.Qx/ of a utility
2.4 Mean-Variance Analysis 39
function u W R ! R of the gamble-random variable xQ . If the agent is risk averse,

then the function u is concave. In general, the expected utility function of a gamble-
random variable depends on the whole distribution of the latter.
In this section, we aim at answering the question of when a preference relation R
can be fully described in terms of the first two moments (the mean and the variance)
of a gamble-random variable. According to this paradigm and motivated also by
the approximation (2.8) of the risk premium of a random variable xQ , the variance is
taken as the most natural candidate in order to measure the riskiness of a gamble
xQ . This leads us to introduce the mean-variance criterion: an agent prefers random
variables with higher expected values and dislikes random variables with higher
variances. More precisely:
xQ 1
MV xQ 2 ” EŒQx1 EŒQx2 and 2 .Qx1 / 2 .Qx2 /:
We also say that an agent is characterized by mean-variance preferences if his

expected utility EŒu.Qx/ of a gamble xQ can be represented as

EŒu.Qx/ D V EŒQx; 2 .Qx/
for some function V W R RC ! R increasing with respect to the first argument

and decreasing with respect to the second argument. Clearly, if xQ 1
MV xQ 2 , then
the gamble xQ 1 is preferred to the gamble xQ 2 by any agent with mean-variance
preferences. Note that the mean-variance criterion only yields a partial order among
random variables, since it does not allow to compare random variables characterized
by differences with the same sign in both the expected value and the variance.
Not every risk averse agent exhibits a preference structure compatible with the
mean-variance criterion, as shown by the following counterexample. Let us consider
the couple of random variables xQ 1 and xQ 2 described by
xQ 1 D f0:5; 3; 4; 7I 1=4; 1=4; 1=4; 1=4g; xQ 2 D f1:5; 3; 8I 1=2; 1=4; 1=4g;
and a logarithmic utility function u.x/ D log.x/. Elementary computations lead to
EŒQx1 D 3:625; 2 .Qx1 / D 5:422 and EŒQx2 D 3:5; 2 .Qx2 / D 7:125;
while
EŒu.Qx1 / 0:406 and EŒu.Qx2 / 0:433:
Therefore, the gamble xQ 1 has a higher mean and a lower variance than the gamble
xQ 2 , but an agent with a logarithmic utility function will prefer xQ 2 to xQ 1 . Therefore,
such an agent does not act according to the mean-variance criterion.
This example shows that, in general, considering the class of risk averse agents
and a random variable xQ with mean EŒQx and variance 2 .Qx/, it is not possible to
express the expected utility U.Qx/ in the form V.EŒQx; 2 .Qx//. However, the mean-
variance representation of preferences holds true under additional assumptions on
the utility function u or on the random variable involved. We now discuss some of
the most important cases.
a) Quadratic utility function.
If the utility function u is quadratic, i.e., u.x/ D x b2 x2 (for b > 0), then the
expected utility U.Qx/ of xQ becomes
b b b
U.Qx/ D EŒQx EŒQx 2 D 1 EŒQx EŒQx 2 .Qx/ DW V EŒQx; 2 .Qx/ :
2 2 2
Since b > 0, the expected utility is decreasing in the variance 2 .Qx/ and, moreover,
for EŒQx 1=b, it is increasing in the expected value. However, representing
preferences by means of a quadratic utility function is somewhat problematic. In
particular, a quadratic utility function exhibits increasing absolute risk aversion
(IARA, see Sect. 2.2) and, for EŒQx > 1=b, the expected utility is decreasing in
the expected value.
b) Second order approximation.
Let us consider a random variable xQ and a twice differentiable increasing and
concave utility function u W R ! R. The expected utility function U.Qx/ D EŒu.Qx/
can be approximated through its Taylor expansion up to the second order centered
on the expected value EŒQx:
1
U.Qx/ u EŒQx C u00 EŒQx 2 DW V EŒQx; 2 .Qx/ :
2
Since the function u is increasing, this approximation implies that, in the case
of a random variable xQ with a small dispersion around its mean (so that u.EŒQx/
dominates) or in the case of a utility function with positive third derivative, the
expected utility function is increasing with respect to the expected value and
decreasing with respect to the variance, due to the concavity of u. However, it should
be noted that the higher order terms in the Taylor expansion depend on the higher
order moments of xQ .
c) Normal distribution.
The mean-variance criterion finds a justification within the expected utility
paradigm if a random variable xQ is normally distributed. In that case, moments of
arbitrary order of xQ can be expressed in terms of the mean EŒQx and of the variance
2 .Qx/. As a consequence, the expected utility function EŒu.Qx/ can be written as
V.EŒQx; 2 .Qx// for some function V W R RC ! R. If the utility function u is
increasing and concave, then the function V will be increasing with respect to its
first argument and decreasing with respect to the second. More precisely, we have
the following proposition.
Proposition 2.13 Let u W R ! R be an increasing and concave utility function

and let xQ be a normally distributed random variable with mean and variance 2 .
Then there exists a function V W R RC ! R, increasing with respect to the first
argument and decreasing with respect to the second, such that EŒu.Qx/ D V.; 2 /.
Proof Let zQ denote a normally distributed random variable with zero mean and
unitary variance and define the function V W R RC ! R by
p
V.x; y/ WD EŒu.x C y zQ/; for x 2 R and y 2 RC :
The representation EŒu.Qx/ D V.; 2 / follows by noting that C zQ d xQ . The

sign of the two partial derivatives of V is established in Exercise 2.14. t
u
As a particular case, let us consider the case of an exponential utility function
u.x/ D 1a exp.ax/, for a > 0, and a normally distributed random variable xQ
with mean and variance 2 . Recall that the random variable exp.aQx/ is distributed
according to a log-normal distribution with expected value exp.a C a2 2 =2/. This
implies that the expected utility admits the following representation:
1 1 a2
U.Qx/ D E Œexp.aQx/ D exp a C 2 DW V.; 2 /:
a a 2
In particular, the function V is increasing with respect to the expected value
and decreasing with respect to the variance 2 (mean-variance preferences). The
coefficient of absolute risk aversion a represents the mean-variance trade-off.
Proposition 2.13 implies an interesting property: the indifference curves of the
expected utility function in the expected return-standard deviation plane are ordered
in an increasing sense, have a positive slope and are convex, see Fig. 2.3. Let us
consider the case of a normally distributed random variable xQ with mean and
variance 2 . The slope of the indifference curves of the expected utility function
can be determined by setting equal to zero the total differential of the function V W
R RC ! R defined in the proof of Proposition 2.13:
@V.; 2 / @V.; 2 /
dV.; 2 / D d C d
@ @

xQ
D EŒu0 .Qx/d C E u0 .Qx/ d D 0:

The indifference curves in the expected return-standard deviation plane have a

positive slope. Indeed:
h i
d E u0 .Qx/ xQ

D 0; (2.14)
d EŒu0 .Qx/
Fig. 2.3 Indifference curves of a mean-variance expected utility
where the inequality can be shown as in Exercise 2.14. The convexity of the
indifference curves can be shown by a similar argument, see Ingersoll [1000,
Chapter 4].
As an example, if the utility function is exponential, i.e., u.x/ D 1a exp.ax/,
for a > 0, then we have (see Exercise 2.15)
d
D a (2.15)
d
and the convexity of the indifference curves is ensured since a > 0. Similarly, in the
case of a quadratic utility function u.x/ D x b2 x2 , with < 1=b, we have that the
slope of the indifference curves is (see Exercise 2.15)
d b
D > 0: (2.16)
d 1 b
As long as < 1=b, the slope is increasing in and, therefore, the indifference
curves are convex in the expected return-standard deviation plane.
The Diversification and Insurance Principles
If an agent’s preferences satisfy the mean-variance criterion, then the variance

becomes the sole risk indicator. In this context, we can introduce two key princi-
ples in financial decision making problems: the diversification and the insurance
principles.
Consider an agent who holds one unit of wealth and wants to determine his
optimal exposure to two random variables xQ 1 and xQ 2 . In other words, the agent needs
to choose the optimal w 2 R such that the linear combination w xQ 1 C .1 w /Qx2
will be preferred to any other linear combination of the two gambles xQ 1 and xQ 2 , i.e.,
w xQ 1 C .1 w /Qx2 R wQx1 C .1 w/Qx2 ; for every w 2 R:
Equivalently, we can interpret this optimal choice problem as the problem of

determining the optimal proportion w of wealth invested in the two lotteries xQ 1
and xQ 2 . According to this interpretation, we can distinguish three cases:
a) 0 < w < 1: the agent invests a positive amount of wealth in both gambles;
b) w > 1: the agent invests an amount larger than his endowment of wealth in the
first gamble and goes short on the second gamble;
c) w < 0: the agent invests an amount larger than his endowment of wealth in the
second gamble and goes short on the first gamble.
The expected utility function of our agent with mean-variance preferences in
correspondence of a generic gamble xQ satisfies U.Qx/ D V.EŒQx; 2 .Qx//, for a function
V W R RC ! R increasing with respect to the first argument and decreasing with
respect to the second. In the present context, this means that our agent will choose
the optimal value w achieving the optimal trade-off between the expected value
and the variance of the linear combination wQx1 C .1 w/Qx2 .
We denote i WD EŒQxi and i2 WD 2 .Qxi /, for i D 1; 2, and let 2 Œ1; 1 be
the correlation coefficient between xQ 1 and xQ 2 . For every w 2 R, we also define the
random variable xQ w WD wQx1 C .1 w/Qx2 , with expected value and variance
EŒQxw D w1 C .1 w/2 and 2 .Qxw / D w2 12 C .1 w/2 22 C 2 w.1 w/1 2 :
Let us first consider the case where the two gambles have the same expected
value and are uncorrelated, i.e., 1 D 2 DW and D 0. In this case, since every
linear combination of xQ 1 and xQ 2 has expected value , the agent’s problem reduces
to the problem of minimizing the variance of the linear combination xQ w :
min w2 12 C .1 w/2 22 : (2.17)

w2R
Since 12 ; 22 > 0, the first order condition for the minimization problem (2.17) is
both necessary and sufficient, yielding the optimal value
22
w D :
12 C 22
In this case, the optimal choice w of the agent belongs to the interval .0; 1/,
meaning that a positive amount of wealth is invested in both gambles xQ 1 and xQ 2 .
This result highlights the diversification principle: in order to minimize the risk (the
variance), the agent will choose to diversify his exposure between the two gambles.
Note also that w .1 w / ” 12 22 , so that the agent will choose
a greater exposure to the gamble with lower variance. In the particular case where
12 D 22 DW 2 , the agent will achieve a perfect diversification between the two
gambles, i.e., the optimal choice will be w D 1=2. In this case, the minimal
variance 2 .Qxw / is equal to one half of the original variance ( 2 =2).
Having discussed the case of two uncorrelated gambles xQ 1 and xQ 2 , let us study
the optimal choice problem when ¤ 0. Let us first consider the case of perfectly
correlated gambles, i.e., jj D 1. The random variables xQ 1 and xQ 2 are perfectly
correlated if and only if there exist a; b 2 R such that
xQ 1 D a C bQx2
and, moreover, we have D 1 if and only if b > 0 (and, analogously, D 1

if and only if b < 0). Let us start with the case D 1, always assuming that
1 D 2 DW . As above, in this case the agent’s optimal choice problem reduces
to the problem of finding the optimal w 2 R which minimizes the variance of xQ w :
min w2 12 C .1 w/2 22 C 2 w.1 w/1 2 : (2.18)

w2R
Note that the first two terms are positive, the third one instead becomes negative
when w < 0 or w > 1. Since 12 C 22 21 2 , the first order condition of the above
minimization is both necessary and sufficient and yields the optimal value
2
w D :
2 1
In correspondence of the optimal choice w , it holds that 2 .Qx w / D 0, meaning

that the agent can completely eliminate the risk. It is interesting to observe that, in
the case D 1, the optimal choice w does not belong to the interval .0; 1/. Indeed,
if 12 < 22 , we have w > 1, meaning that the agent does not diversify his exposure
between the two sources of risk xQ 1 and xQ 2 , but invests an amount of wealth greater
than his endowment of wealth in the first gamble xQ 1 (the one with smaller variance)
and sells short the gamble xQ 2 . A similar argument applies to the case 12 > 22 .
Let us now consider the case of perfect negative correlation, i.e., D 1. In that
case, the minimization problem corresponding to (2.18) is
min w2 12 C .1 w/2 22 2 w.1 w/1 2 : (2.19)

w2R
In this case, the third term of (2.19) becomes negative when 0 < w < 1. It is easy
to check that the optimal solution is given by
2
w D :
2 C 1
As in the case of perfect positive correlation, we have 2 .Qx w / D 0, i.e., the

agent can completely eliminate the risk of his exposure to the two gambles xQ 1 and
xQ 2 . However, unlike in the case of perfect positive correlation, here w 2 .0; 1/,
meaning that the agent will diversify his exposure between the two gambles, putting
more weight on the gamble with smaller variance.
The analysis of the two cases of perfect positive and negative correlation allows
us to introduce the insurance principle. Indeed, in the case of perfect negative
correlation, the agent interested in minimizing the variance chooses to invest a
positive amount of wealth in both gambles xQ 1 and xQ 2 . Intuitively, since the two
gambles are perfectly negatively correlated, values of xQ 1 above its mean will be
compensated by values of xQ 2 below its mean and vice-versa. Hence, by combining
the two gambles with positive weights, the agent can reduce to zero the riskiness of
his overall portfolio. Analogously, in the case of perfect positive correlation, the
agent chooses to invest a positive amount of wealth in the less risky gamble and
a negative amount in the riskier gamble, in such a way to create a compensation
between the two sources of randomness. This argument takes the name of insurance
principle, since the exposure on the gamble xQ 1 insures the exposure on the gamble
xQ 2 .
Let us now consider the case where jj ¤ 1. In this case, the variance
minimization problem corresponding to problems (2.17)–(2.18) is
min w2 12 C .1 w/2 22 C 2 w.1 w/1 2 :

w2R
It can be readily checked that the optimal choice is given by
22 1 2
w D :
12 C 22 21 2
Suppose that 12 > 22 . In this case, since 12 > 1 2 , it holds that
22 1 2 D 12 C 22 21 2 .12 1 2 / < 12 C 22 21 2 ;
thus showing that w < 1. This means that, if the agent is interested in minimizing
the variance, then he will never invest more than his endowment of wealth in the
riskier gamble. The sign of w depends on the numerator 22 1 2 . Indeed, if
1 < 2 =1 then we have w 2 Œ0; 1, while if 2 =1 < < 1 then we have
w < 0. This means that, if the correlation is negative or positive but not too high,
then the agent will choose to diversify his exposure between the two gambles xQ 1 and
xQ 2 , while, if the correlation is close to 1, then he will adopt a strategy similar to the
one obtained in the case of perfect positive correlation, investing in the less risky
gamble more than his endowment of wealth and selling short the riskier gamble.
Note that, unlike the cases of perfect positive and negative correlation, it is not
possible to reduce to zero the total variance if jj ¤ 1.
The above discussion can be illustrated by means of Fig. 2.4, which considers
the case of two gambles xQ 1 and xQ 2 with means 1 and 2 and variances 12 and 22
(with 1 > 2 and 12 > 22 ). The figure represents the possible combinations of
expected value and standard deviation obtained in the cases D 1, D 1 and
Fig. 2.4 Mean-variance of portfolios

1 < < 1. We can observe that, in the case of perfect negative correlation, it is
possible to eliminate the risk with a diversified investment in xQ 1 and xQ 2 , while this is
possible by choosing w < 0 in the case of perfect positive correlation. In the case
where 1 < < 1, the risk can never be perfectly eliminated.
Let us conclude this section by generalizing the above problem to the case of
N > 2 gambles-random variables xQ n , for n D 1; : : : ; N. Let us assume that an
agent needs to choose his exposure/investment
P wn 2 R in the n-th gamble, for each
n D 1; : : : ; N, subject to the constraint NnD1 wn D 1 (i.e., the agent must allocate
P
all his wealth among the N gambles). Let us denote xQ w D NnD1 wn xQ n . Then
X
N X
N X
N
EŒQxw D wn n and 2 .Qxw / D wn wm Cov.Qxn ; xQ m /:
nD1 nD1 mD1
Let us consider the special case where the N gambles are uncorrelated, meaning that
Cov.Qxi ; xQ j / D 0 for all i; j 2 f1; : : : ; Ng with i ¤ j. Then, the problem of minimizing
the variance of the linear combination xQ w becomes
X
N
min w2n 2 .Qxn / :
Pw2RN
nD1
s.t. NnD1 wn D1
We introduce the Lagrangian

!
X
N X
N
L.w1 ; : : : ; wN ; / D w2n 2
.Qxn / C 1 wn : (2.20)
nD1 nD1
As shown in Exercise 2.21, the optimal solution is given by
1= 2 .Qxn /
wn D PN ; for all n D 1; : : : ; N: (2.21)
2 x /
nD1 1= .Q n
In other words, in order to minimize the overall variance, the exposure to each
gamble xQ n should be proportional to the (normalized) reciprocal of the variance
of the gamble xQ n , for n D 1; : : : ; N. In particular, if the N gambles have the same
variance 2 , then the agent will equally distribute his wealth among the N gambles,
achieving a perfect diversification, i.e.,
1
wn D ; for all n D 1; : : : ; N:
N
Note that, with a perfectly diversified portfolio, the overall variance amounts to
2 =N. Therefore, thanks to the diversification principle, in the limit the variance
will tend to zero for N ! 1 (perfect diversification).
We refer the reader to Kreps [1136] for a comprehensive account of decision

theory in a risky environment. Assumptions 2.1–2.3 play a key role in representing
preferences via an expected utility function. In particular, referring to Chap. 9 for
a more detailed discussion, the independence axiom (Assumption 2.3) turns out to
be rather controversial on an empirical and theoretical ground. The most famous
argument goes back to the Allais Paradox (see Allais [41]). Consider four gambles
with monetary values f0; 1000; 5000g and corresponding probabilities
A D f0; 1; 0g; B D f0:01; 0:89; 0:1g;

C D f0:9; 0; 0:1g; D D f0:89; 0:11; 0g:
It is empirically observed that people express the following preference order:

A P B and C P D. It is easy to show that this preference relation does not satisfy
the independence axiom. Indeed, let X; Y; Z be the gambles paying the amounts
0; 1000; 5000 with certainty, respectively. A contradiction to the independence
axiom can be easily obtained as follows:

11 89 11 1 10 89
AI YC Y and BI XC Z C Y:
100 100 100 11 11 100
The transitivity of the preference relation P implies that

11 89 11 1 10 89
YC YP XC Z C Y:
100 100 100 11 11 100
By the independence axiom, it holds that
1 10
AP X C Z;
11 11
and

11 89 11 1 10 89
YC XP XC Z C X:
100 100 100 11 11 100
This means that D P C; and, therefore, a contradiction is obtained. One can also
exhibit other situations analogous to the Allais paradox: in general, they require
agents to evaluate gambles with small/large probabilities. These paradoxes highlight
the key role of the independence axiom, which amounts to assume that an agent is
able to identify the common part of two gambles and to evaluate them only by
considering what is different (i.e., an agent is able to “decontextualize” his choices).
However, this is not observed in reality: an agent typically takes different decisions
in relation with the context surrounding him, since the context itself affects the
preferences.
A version of the expected utility representation theorem according to a sub-
jectivist interpretation of probability has been provided in Savage [1499]. In this
setting, agents do not know the true probability of the events. Savage [1499] starts
from a complete preorder defined on M without pre-specified probabilities and
shows that, if the preorder satisfies some assumptions, then there exists both a
subjective probability distribution and a utility function such that the preorder can
be represented by means of the expected value of a utility function, where the
expectation is taken with respect to the subjective probability distribution. Agents
behave as if those probabilities were assigned to the events (on this approach, see
also Sect. 9.1 and Anscombe & Aumann [72]). Another interesting extension of the
expected utility theory is represented by the situation where an agent exhibits state
dependent preferences, i.e., his utility function u depends on the state of the world
! 2 ˝. A generalization of the expected utility representation result (Theorem 2.1)
can be provided in this setting, relaxing the independence axiom.
In Sect. 2.1, we have identified a set of conditions on a preference relation R
so that the latter can be represented by an expected utility function. In applications,
instead of describing the behavior of the agent through his preference relation, it is
often directly assumed that the agent is characterized by a given utility function. The
drawback of this approach is that it can lead to consider a utility function and a class
of gambles which are not compatible with a representation through an expected
utility function, because the utility function is unbounded and/or the expected value
of the gamble is not finite. This drawback is avoided in the following cases: a
bounded utility function; gambles taking values on a finite set; gambles with a finite
expected value together with an increasing and concave utility function. The last
case deserves to be verified. Given an increasing, concave and differentiable utility
function u W R ! R, we have that u.x/ u.z/ C u0 .z/.x z/ for every x; z 2 R. If
the expected value of the gamble xQ is finite, then
EŒu.Qx/ u.z/ C u0 .z/.EŒQx x/ < C1
and, hence, the preference relation admits a representation through the expected
utility.
Note that the risk aversion hypothesis is not consistent with the observation that
many agents stipulate insurance contracts and, at the same time, accept gambles
which are far from being actuarially fair (think of casino games, national lotteries,
etc.). This behavior can be explained by assuming that agents are characterized by
a concave utility function for small levels of wealth, then convex and then again
concave for large values of wealth, as proposed in Friedman & Savage [744].
Kahneman & Tversky [1060] instead conclude that investors maximize a function
which is convex for positive outcomes and concave for negative outcomes, i.e.,
agents are loss averse. We will return to this type of utility functions in Chap. 9.
In Ross [1468], a different measure of risk aversion is proposed: an agent a is

strongly more risk averse than an agent b if
00 0
ua .z/ ua .z/
inf b00 sup b0 :
z2R u .z/ z2R u .z/
This condition implies that agent a is more risk averse than agent b (in the sense
of Sect. 2.2), but the converse implication does not hold. It can be shown that the
above condition is verified if and only if there exists a strictly positive constant
and a non-increasing and concave function g W R ! R such that
ua .z/ D ub .z/ C g.z/; for every z 2 R:
Let agent a be strongly more risk averse than agent b. If the random variable xQ
represents the initial random endowment of an agent (background risk) then the
risk premium of a gamble xQ C Q such that EŒQ jQx D 0 for agent a is larger than
that for agent b, see Ross [1468]. Note that risk aversion alone does not suffice
for this property to hold true. In a multiple goods setting, risk aversion has been
characterized in Kihlstrom & Mirman [1089, 1090] (see also Gollier [800]). More
recently, new characterizations of risk aversion have been proposed referring to the
utility that an agent gets from two gambles. For instance, the definition of proper
risk aversion (see Pratt & Zeckauser [1433]) can be summarized as follows: if
two independent gambles are individually unattractive, then the compound lottery
offering both together is less attractive than either alone. The definition of standard
risk aversion (see Kimball [1101]) extends the above definition to “loss aggravating”
independent risks (see also Gollier & Pratt [803]). For a survey of the recent
literature on risk aversion we refer to Gollier [800] (see also Eeckhoudt & Gollier
[630] on decision making problems in a risky setting). We refer the interested reader
to Levy [1205], Sriboonchitta et al. [1560] for two recent accounts on stochastic
dominance and its implications for the purposes of financial modeling.
The reconciliation of the mean-variance criterion with the expected utility
approach represented a challenging problem in the past. Assuming an increasing
and concave utility function, what are the probability distributions compatible with
the mean-variance criterion? In Tobin [1596], besides verifying that the normal mul-
tivariate distribution satisfies this property, the author claimed that every probability
distribution characterized by two parameters makes the mean-variance criterion
compatible with the expected utility. This conjecture turns out to be incorrect: the
missing point is that a linear combination of random variables with probability
distributions belonging to a given common family does not necessarily have a
probability distribution belonging to the same family. The normal multivariate
distribution obviously enjoys this property. More generally, the distributions which
are compatible with the mean-variance are the elliptical distributions, see Ingersoll
[1000], Owen & Rabinovitch [1389] and also Meyer [1334] for the characterization
of the class of linear distribution functions. In Chamberlain [395], the class of
2.6 Exercises 51
probability distributions such that the expected utility is a function only of the
mean and of the variance is characterized. An important property turns out to be
that of spherical symmetry: a vector of random variables is spherically distributed
(around the origin) if its distribution is invariant with respect to orthogonal linear
transformations that leave the origin fixed. Note however that for this class of
distributions the expected utility function for an increasing and concave utility
function u is not necessarily increasing in the expected value and decreasing in
the variance. Some counterexamples concerning mean-variance preferences are
presented in Exercises 2.22 and 2.23.
2.6 Exercises
Exercise 2.1 In the context of a finite probability space, as in Theorem 2.1, show
that the expected utility representation is determined up to a strictly positive linear
transformation.
Exercise 2.2
(i) Show that the logarithmic utility function u.x/ D log.a C x/ is the limiting
1
case of a power utility function u.x/ D b1 ..a C bx/.b1/=b 1/ for b ! 1.
(ii) Show that the exponential utility function u.x/ D exp.x=a/ is the limiting
1 b1
case of a power utility function u.x/ D b1 .1 C bx
a
/ b for b ! 0.
(iii) Show that the utility function u.x/ D 1ea tends to a linear utility function as
ax
a ! 0.
Exercise 2.3 Let xQ be a random variable with expected value EŒQx and variance
k2 2 , admitting the representation xQ D EŒQx C k Q , where Q is a random variable with
zero mean and variance 2 . Show that, for any strictly increasing and differentiable
utility function u W R ! R, the risk premium u .Qx; k/ parameterized by k satisfies
u .Qx; 0/ D 0 and @u@k
.Qx;k/
jkD0 D 0.
Exercise 2.4 Given a quadratic utility function u.x/ D x b2 x2 , for b > 0, and
a random variable xQ with expected value and variance 2 , determine the risk
premium u .Qx/.
Exercise 2.5 Given the exponential utility function u.x/ D 1a eax ; a > 0; and a
normally distributed random variable xQ N .; 2 /, determine the risk premium.
1a
Exercise 2.6 Given a power utility function u.x/ D x1a , for a > 0, and a random
variable xQ such that log xQ is distributed as a Normal N .; 2 /, determine the risk
premium.
Exercise 2.7 Consider an agent a with piecewise linear utility function ua .x/ D x
for x x0 and ua .x/ D x0 C a.x x0 / otherwise, for some a 2 .0; 1/, and an agent
b with utility function ub .x/ D x for x x0 and ub .x/ D x0 C b.x x0 / otherwise,
for some b 2 .0; 1/. Show that agent b is more risk averse than agent a if and only
if b a.
Exercise 2.8 Let us consider a three times differentiable utility function u W
R ! R.
(i) Prove that, if u is increasing and DARA, then u000 .x/ > 0 for every x 2 R.
(ii) For an increasing and concave utility function u W R ! R, define the degree
of absolute prudence as pau .x/ WD u000 .x/=u00 .x/, for x 2 RC . Show that the
utility function u is DARA if and only if pau .x/ > rua .x/, for every x 2 R.
(iii) Show that pau .x/ > 0, for all x 2 R, is a necessary condition to have a DARA
utility function.
Exercise 2.9 Let u W R ! R be a twice differentiable utility function and denote
by rua .x/ the coefficient of absolute risk aversion. Show that, for suitable c; k 2 R:
Rx
u0 .x/ D ke c rua .y/dy
; for every x 2 R:
Exercise 2.10 Consider the HARA class of utility functions introduced in Sect. 2.2
(rua .x/ D 1=.a C bx/, for some a; b 2 R). Verify that:
(i) a HARA utility function shows a decreasing coefficient of absolute risk aversion
if and only if b > 0 and a decreasing coefficient of relative risk aversion if and
only if a < 0;
(ii) the three specifications of utility function log.x C a/ for b D 1, a exp.x=a/
1 b1
for b D 0 and b1 .a C bx/ b in the remaining cases provide an exhaustive
representation of the HARA class of utility functions. Determine the domain of
these functions.
Exercise 2.11 An agent has to make a choice between two different gambles: a)
receive 25000 euro with certainty; b) receive 32000 euro with probability 0:2,
10000 euro with probability 0:7 and 1000 euro with probability 0:1. The agent is
characterized by a power utility function u.x/ D x , with 2 .0; 1/, and he is
indifferent between two gambles: c) receive 20:25 euro with certainty; d) receive 16
euro with probability 1=2 and 25 euro with probability 1=2. Determine and which
of the two gambles a) and b) will be preferred by the agent.
Exercise 2.12
(i) Show that a utility function u W R ! R is characterized by a constant relative
risk aversion coefficient different from 1 if and only if u.x/ D ˛x1b C ˇ for
˛ > 0 and ˇ 2 R, with b ¤ 1.
(ii) Show that a utility function u W R ! R is characterized by a constant relative
risk aversion coefficient equal to one if and only if u.x/ D ˛ log.x/ C ˇ for
˛ > 0 and ˇ 2 R.
2.6 Exercises 53
Exercise 2.13 Show that if agent a is strongly more risk averse than agent b, in
the sense of Sect. 2.5, then agent a is also more risk averse than agent b, but the
converse is not true.
Exercise 2.14 In the context of Proposition 2.13, prove that the mean-variance
function V W R RC ! R is increasing with respect to the first argument and
decreasing with respect to the second one.
Exercise 2.15 Prove that, in the case of a normally distributed random variable xQ ,
the slope d=d of the indifference curves in the expected return-standard deviation
plane for an exponential utility function u.x/ D 1a exp.ax/, for a > 0, is given
by formula (2.15) and, in the case of a quadratic utility function u.x/ D x b2 x2 , for
< 1=b, by formula (2.16).
Exercise 2.16 Prove that, if two random variables xQ 1 and xQ 2 are normally dis-
tributed and have the same mean , then xQ 1
SSD xQ 2 if and only if 2 .Qx1 / 2 .Qx2 /.
Exercise 2.17 Consider two gambles xQ 1 D f0; 1I 0:3; 0:7g and xQ 2 D f0; xI ; 1g,
with x 2 Œ0; 1. Provide necessary and sufficient conditions on x and so that
(i) xQ 1
FSD xQ 2 ;
(ii) xQ 1
SSD xQ 2 ;
(iii) xQ 1
M
SSD xQ 2;
(iv) xQ 1
MV xQ 2 .
Exercise 2.18 Let xQ 1 be a random variable taking values f1; 1; 0g with probabili-
ties f; ; 1 2g, respectively, and xQ 2 a random variable taking values f1; 1; 0g
with probabilities f; ; 1 2 g, respectively. Show that xQ 1
SSD xQ 2 if and only if
.
Exercise 2.19 Show that if xQ 1
FSD xQ 2 then EŒQx1 EŒQx2 and provide a
counterexample showing that the converse implication does not hold.
Exercise 2.20
(i) Does xQ 1
M
SSD xQ 2 imply 2 .Qx1 / 2 .Qx2 /?
(ii) Does xQ 1
SSD xQ 2 imply xQ 1
MV xQ 2 ?
M
Exercise 2.21 In the context of Sect. 2.4, prove that the solution to the variance
minimization problem (2.20) is given by (2.21).
Exercise 2.22 (Hens & Rieger [938], Theorem 2.30) Suppose that the expected
utility function EŒu.Qx/ of a risk averse agent can be represented through a
continuous mean-variance function V.EŒQx; 2 .Qx//, strictly increasing with respect
to the mean and strictly decreasing with respect to the variance. Then, there exist
two random variables xQ 1 ; xQ 2 such that P.Qx1 xQ 2 / D 1 and P.Qx1 > xQ 2 / > 0, but
EŒu.Qx1 / < EŒu.Qx2 /.
Exercise 2.23 (Hens & Rieger [938], Corollary 2.31) Suppose that the expected
utility function EŒu.Qx/ of a risk averse agent can be represented through a
continuous mean-variance function V.EŒQx; 2 .Qx//, strictly increasing in the first
argument, for any 2 .Qx/. Then the preference relation represented by EŒu.Qx/ does
not satisfy the independence axiom.
Chapter 3
Portfolio, Insurance and Saving Decisions
The most important questions of life are, for the most part,
really only problems of probability.
Pierre Simon de Laplace (Théorie Analytique des Probabilités)
To withdraw is not to run away, and to stay is no wise action
when there is more reason to fear than to hope. ‘Tis the part of a
wise man to keep himself today for tomorrow, and not venture
all his eggs in one basket.
Miguel de Cervantes (Don Quixote)
In the previous chapter, we have presented a general framework for decision making
problems under risk, based on expected utility theory. In that context, decisions
concerned abstract random variables. In the present chapter, we will consider
random variables representing returns of assets/insurance contracts. This allows us
to analyse optimal portfolio, insurance and saving problems in a risky setting and,
in particular, to characterize the optimal choice of a risk averse agent.
This chapter is structured as follows. In Sect. 3.1, we study the optimal portfolio
problem, first in the case of a single risky asset and then in the presence of multiple
risky assets. In Sect. 3.2, we develop the theory of mean-variance portfolio selection
and study the properties of the mean-variance portfolio frontier. Section 3.3 deals
with optimal insurance problems, while Sect. 3.4 lays the foundations of the analysis
of optimal saving-consumption problems. At the end of the chapter, we provide a
guide to further readings as well as a series of exercises.
3.1 Portfolio Theory
In this section, we consider the problem of an agent aiming at optimally investing

his wealth by trading on a set of risk free and risky assets, where the optimality
criterion is represented by the maximization of expected utility. As in the previous
chapter, we consider a two-period model, where at the initial date t D 0 the agent
has to make his investment choices without knowing the state of the world ! 2 ˝,
which will only become known at the future date t D 1. In this context, the agent’s
choices concern random variables representing the random returns of a set of traded
assets, where the return of an asset is defined as the ratio between the (random)

56 3 Portfolio, Insurance and Saving Decisions
dividend yielded by the asset at t D 1 and its current price at t D 0. The optimal
choice problem of an agent consists in determining the portfolio of the traded assets
which maximizes the expected utility of the random wealth obtained at t D 1.
We assume the existence of a financial market where each asset can be bought
and sold without liquidity constraints, transaction costs or other market frictions.
More specifically, we assume the existence of a risk free asset (denoted as asset 0
and generically interpreted as a bank account), the return of which is constant over
all states of the world and denoted by rf > 0. This asset represents the possibility
of investing-borrowing at the risk free rate rf . Besides the risk free asset, the market
contains N 2 N risky assets. The return of asset n at t D 1 is denoted by the random
variable rQn , where rQn WD dQ n =pn , with dQ n denoting the random dividend of asset n
delivered at date t D 1 and pn > 0 denoting the price of asset n at date t D 0, for
n D 1; : : : ; N. According to this notation, rQn represents the random wealth obtained
at t D 1 by investing one unit of wealth in the n-th asset at the initial date t D 0.
Let us denote by w0 the initial endowment/wealth of an agent at date t D 0.
The agent’s problem consists in allocating the initial wealth w0 among the N C 1
available assets (i.e., the risk free asset together with the N risky assets) in order to
maximize the expected utility of the random wealth obtained at t D 1. We denote by
wn , for n D 1; : : : ; N, the amount of wealth invested at time t D 0 in the n-th asset.
For each n 2 f1; : : : ; Ng, the inequality wn > 0 means that the agent is investing
a positive amount of wealth in the n-th asset, while wn < 0 means that the agent
is selling short the n-th asset for an amount of wealth equal to jwn j (i.e., he sells
the amount jwn j ofP the asset without owning it). Since the budget constraint has to
be satisfied, w0 NnD1 wn represents the amount of initial wealth invested in the
P
risk free asset. As before, the inequality w0 NnD1 wn > 0 means that the agent
is investing a positive amount of wealth in the bank account, while the converse
inequality means that the agent is borrowing a positive amount of wealth from the
bank account. Starting from initial wealth w0 , a vector w D .w1 ; : : : ; wN /> 2 RN
represents a portfolio, i.e., an investment strategy in the N C 1 assets.
The random returns of the N risky assets are represented by different ran-
dom variables. Without loss of generality, we assume that the random variables
.rf ; rQ1 ; : : : ; rQN / are almost surely linearly independent, meaning that no random
return can be expressed as the linear combination of the other N 1 random returns
and of the risk free return. We denote by e D .e1 ; : : : ; eN /> 2 RN the vector of
the expected returns of the risky assets, i.e., en D EŒQrn , for all n D 1; : : : ; N, and
by V 2 RNN the variance-covariance matrix of the N random returns of the risky
assets:
2 3
2 .Qr1 / Cov.Qr1 ; rQ2 / : : : Cov.Qr1 ; rQN /
6 Cov.Qr2 ; rQ1 / 2 .Qr2 / : : : Cov.Qr2 ; rQN / 7
6 7
VD6 :: :: :: :: 7 :
4 : : : : 5
2
Cov.QrN ; rQ1 / ::: : : : .QrN /
3.1 Portfolio Theory 57
The matrix V is symmetric and, since asset returns are assumed to be linearly
independent, it is also positive definite.
Since the returns of the N risky assets are random, the wealth at time t D 1
generated by a portfolio w D .w1 ; : : : ; wN /> 2 RN will also be random. We
e the random variable representing the wealth obtained at t D 1 from an
denote by W
investment strategy w 2 RN , starting from the initial wealth w0 2 R. More precisely:
X
N X
N X
N

e D w0
W wn rf C wn rQn D w0 rf C wn rQn rf : (3.1)
nD1 nD1 nD1
The second equality in (3.1) shows that the portfolio w D 0 2 RN (i.e., the strategy
investing the total amount of the initial wealth w0 in the risk free asset) yields
a portfolio return equal to rf (this represents the equivalent to the status quo in
Chap. 2). On the contrary, if the agent decides to invest an amount of wealth equal
to wn > 0 on the n-th asset, for n D 1; : : : ; N, then his wealth at t D 1 will increase
or decrease depending on whether the random return rQn of the n-th asset will be
greater or smaller than the risk free rate rf .
The agent’s optimal portfolio problem consists in solving the following maxi-
mization problem, starting from a given initial wealth w0 2 R:

e ;
max E u W (3.2)
w2RN
with We being given as in (3.1) and where we tacitly assume that the expected value
e is finite for all w 2 RN . Note that, if the agent is risk averse, then his
EŒu.W/
e Indeed, risk aversion implies that the
expected utility function is concave in W.
utility function u W R ! R is concave (see Proposition 2.4) and, hence, given two
portfolios w1 ; w2 2 RN and two initial wealths w10 ; w20 2 R, by the linearity of the
expectation it holds that
1 1 2
e C.1/W
E u W e C.1/E u W
e 2 E u W e ; 8 2 Œ0; 1; (3.3)
where W e i denotes the random wealth at t D 1 associated to the portfolio wi and

starting from the initial wealth wi0 , for i D 1; 2. Moreover, if u is strictly concave,
e Similarly, the function
then the expected utility function is strictly concave in W.
e
w0 7! EŒu.W/ is also concave.
By relying on representation (3.1), the optimal portfolio problem can be rewritten
as follows, for a given initial wealth w0 2 R:
" #
X
N

max E u w0 rf C wn rQn rf :
w2RN
nD1
Assuming an increasing utility function u, a necessary condition for the existence

of a solution to problem (3.2) is that the random return rQn is lower than the
risk free rate rf in some states of the world and higher than rf in some other
states, for all n D 1; : : : ; N. In other words, if P.Qrn ¤ rf / > 0, we must have
P.Qrn < rf / > 0 and P.Qrn > rf / > 0, for all n D 1; : : : ; N. Indeed, if this was not
the case, then in the absence of trading constraints it would be possible to attain
arbitrarily large values of expected utility. This condition is related to the absence
of arbitrage opportunities in the market. Moreover, it turns out that, under suitable
assumptions, the same condition is also sufficient for the existence of a solution
to the portfolio problem (3.2) (see Sect. 4.4 for more details). We call EŒQrn rf
the return risk premium of asset n, for n D 1; : : : ; N. Concerning the uniqueness
of the optimal solution to problem (3.2), it is easy to show that, if the utility
function u is strictly concave, then the solution in terms of wealth is necessarily
unique. Indeed, if there were two optimal wealths W e 1 ; W
e 2 , any strictly convex
e 1 e 2
combination W C .1 /W would yield a strictly higher expected utility,
as shown in (3.3), thus contradicting the optimality of W e 1 ; W
e 2 . Moreover, under
the assumption of linearly independent returns, it can be shown that the optimal
portfolio w 2 RN is unique (see Exercise 3.3).
If the utility function u is differentiable, then the first order necessary conditions
(i.e., the gradient of the expected utility function has to be equal to zero) for an
optimal portfolio w 2 RN implies that

e /.Qrn rf / D 0;
E u 0 .W for all n D 1; : : : ; N; (3.4)
where u0 denotes the first derivative of the utility function u and W e denotes the
optimal wealth resulting from the solution of problem (3.2). If the utility function
u is strictly concave and twice differentiable, so that u00 < 0, then condition (3.4)
is also sufficient for the optimality of the portfolio w 2 RN . We denote by rQ the
random return of a generic portfolio w (so that W e D w0 rQ ) and by rQ the optimal
e D w0 rQ .
portfolio return, so that W
The analysis of optimal portfolio problems changes depending on the number

of risky assets available in the market. In the following, we shall first consider the
simple case of a single risky asset (N D 1) and then the more general case of
multiple risky assets (N > 1). The key difference between the two cases is that, in
the presence of multiple risky assets, each individual asset is evaluated not only for
its own riskiness but also for its contribution to the risk of a portfolio as a whole.
In the latter case, the diversification and the insurance principles introduced in the
previous chapter will play a crucial role.
The Case of a Single Risky Asset
In the case of a single risky asset (N D 1), the optimal portfolio problem can
be analysed in an explicit way, establishing an important relationship between
the return risk premium of the risky asset and the optimal portfolio, as shown
in the following proposition. We always assume that utility functions are twice
differentiable and denote by rQ the random return of the single risky asset.
Proposition 3.1 If an agent is strictly risk averse and characterized by a strictly
increasing utility function u and initial wealth w0 , then the following hold:
(i) w D 0 if and only if EŒQr rf D 0;
(ii) w > 0 if and only if EŒQr rf > 0;
(iii) w < 0 if and only if EŒQr rf < 0.
Proof If w D 0, condition (3.4) directly implies that EŒQr rf D 0, since u is
assumed to be strictly increasing and, hence, u0 .w0 rf / > 0. Conversely, suppose
that EŒQr rf D 0. Recall that strict risk aversion implies strict concavity of the
utility function u. Then, by the optimality of We , it holds that

e
u.w0 rf / D u.w0 rf / C u0 .w0 rf / EŒQr rf w E u W u.w0 rf /;
thus showing that EŒu.W e / D u.w0 rf /. In turn, this implies that w D 0, since

the optimal portfolio w is unique, by the strict concavity of u and the assumption
of linearly independent returns. We have thus proved claim .i/. Moreover, since
u0 .w0 rf / > 0, the sign of the first derivative of the expected utility function with
respect to w in correspondence of w D 0 is strictly negative or strictly positive
depending on whether EŒQr rf > 0 or EŒQr rf < 0, respectively. Since the utility
function u is concave, this implies claims .ii/-.iii/. t
u
According to Proposition 3.1, a risk averse agent invests a strictly positive
amount of wealth in the risky asset if and only if its return risk premium is strictly
positive. Observe that the riskiness of the asset itself does not matter for the sign of
the optimal portfolio w . Furthermore, if the return risk premium of the risky asset is
zero, then any investment is less attractive than investing the total amount of wealth
in the risk free asset.
We now proceed to derive several comparative statics results. Comparative statics
results aim at understanding the impact on the optimal portfolio w of changes of
the coefficient of risk aversion, of the initial wealth and of the distribution of the
return of the risky asset. We start with the following proposition, which compares
the optimal portfolios of two risk averse agents characterized by different degrees
of risk aversion.
Proposition 3.2 Let a and b denote two risk averse agents characterized by strictly
increasing and strictly concave utility functions ua and ub , respectively, with the
same initial wealth w0 > 0. Suppose that EŒQr rf > 0. If agent a is more risk
averse than agent b, then the demand of the risky asset by agent a is smaller than
that by agent b, i.e., wa wb .
Proof Let wa and wb denote the optimal portfolios for agents a and b, respectively.
Since EŒQr rf > 0, Proposition 3.1 implies that wa > 0 and wb > 0.
By Proposition 2.5, there exists an increasing, concave and twice differentiable
function g./ such that ua .x/ D g.ub .x//. The optimality condition (3.4) for agent a
is
0
E ua w0 rf C wa .Qr rf / .Qr rf / D 0:
Let
0
f .w/ WD E ua w0 rf C w.Qr rf / .Qr rf / :
By the concavity of ua , it can be readily verified that w 7! f .w/ is decreasing. To

prove that wa wb it is enough to show that f .wb / 0. Observe that
h 0 i
f .wb / D E g0 ub w0 rf C wb .Qr rf / ub w0 rf C wb .Qr rf / .Qr rf / (3.5)
and
0
E ub w0 rf C wb .Qr rf / .Qr rf / D 0: (3.6)
The difference between f .wb / in (3.5) and the expression in (3.6) is due to the
positive and decreasing function g0 ./ evaluated at ub .w0 rf C wb .Qr rf //. It is easy
to see that

g0 ub w0 rf C wb .Qr rf / .Qr rf / g0 ub .w0 rf / .Qr rf /;
which in turn implies that

0
f .wb / g0 ub .w0 rf / E ub w0 rf C wb .Qr rf / .Qr rf / D 0;

u
The result of Proposition 3.2 is in line with economic intuition. Indeed, it shows
that a more risk averse agent will invest a larger amount of wealth in the risk free
asset, in order to reduce the riskiness of his overall portfolio. In Proposition 2.5,
we have shown that an agent’s risk premium increases as his absolute risk aversion
coefficient increases. Proposition 3.2 is coherent with this result, since it shows that
the optimal investment in the risky asset decreases as the risk aversion increases.
In a setting with a single risky asset, it is also possible to derive comparative
statics results on the risky asset demand as the initial wealth, the risk premium or
the risk free rate change. We start by considering the effects of a change in the initial
wealth on the risky asset demand by a risk averse agent. We denote by w .w0 / the
risky asset demand (i.e., the optimal portfolio) of an agent endowed at date t D 0
with the initial wealth w0 2 RC .
Proposition 3.3 Let u be a strictly increasing and strictly concave utility function.
If w .w0 / > 0 for all w0 2 RC and rf > 0, then:
(i) if rua W RC ! RC is decreasing then w W RC ! RC is increasing;
(ii) if rua W RC ! RC is increasing then w W RC ! RC is decreasing;
(iii) if rua W RC ! RC is constant then w W RC ! RC is constant.
Proof We only prove the first claim, the proof of .ii/ and .iii/ being similar. The
optimal portfolio w .w0 / for an agent endowed with wealth w0 at t D 0 is implicitly
defined by the optimality condition (3.4). By the implicit function theorem,
dw e /.Qr rf /rf

EŒu00 .W
D :
dw0 e /.Qr rf /2
EŒu00 .W
Due to the strict concavity of u, the denominator of the above expression is negative
and, therefore, the sign of dw =dw0 depends on the sign of the numerator. If
w .w0 / > 0 and u is DARA, then it can be easily shown that
e /.Qr rf / rua .w0 rf /u0 .W

u00 .W e /.Qr rf /: (3.7)
Indeed, since w .w0 / > 0, we observe that W e w0 rf on the event fQr rf g and
therefore the fact that u is DARA implies that u00 .W e /=u0 .W
e / rua .w0 rf / on
fQr rf g. On the complementary event fQr < rf g, the converse inequality holds. By
multiplying by rQ rf , we obtain inequality (3.7). In turn, inequality (3.7) implies
that

e /.Qr rf / rua .w0 rf /E u0 .W
E u00 .W e /.Qr rf / D 0;
where the last equality follows from the optimality condition (3.4). Therefore, if
rua W RC ! RC is decreasing (i.e., the utility function u is DARA) and w .w0 / > 0,
then dw =dw0 0. t
u
Proposition 3.3 shows that, if the utility function is DARA, then the risky asset is
a normal good, in the sense that the agent’s optimal demand is an increasing function
of wealth. On the contrary, if the utility function is IARA, then the risky asset is an
inferior good, meaning that an agent’s optimal demand is a decreasing function
of wealth. Finally, if the utility function is CARA, then the optimal investment
in the risky asset is independent of wealth. This result suggests that a DARA
utility function is the most plausible hypothesis for the majority of the agents in
an economy (see Arrow [78]).
According to Proposition 3.3, the coefficient of absolute risk aversion gives
information about changes in the risky asset demand. However, it does not give
information on changes in the proportion of wealth invested in the risky asset. In this
regard, the coefficient of relative risk aversion matters and we have the following
proposition.
If w .w0 / > 0 for all w0 2 RC and rf > 0, then:

(i) if rur W RC ! RC is decreasing then dwdw.w0 0 / ww.w0 0 / 1 and, in particular, the
function w0 7! w .w0 /=w0 is increasing;

(ii) if rur W RC ! RC is increasing then dwdw.w0 0 / ww.w0 0 / 1 and, in particular, the
function w0 7! w .w0 /=w0 is decreasing;

(iii) if rur W RC ! RC is constant then dwdw.w0 0 / ww.w0 0 / D 1 and, in particular, the
function w0 7! w .w0 /=w0 is constant.
Proof We will only prove part .i/, the proof of parts .ii/ and .iii/ being similar. By
the implicit function theorem applied to the optimality condition (3.4), we have that
the elasticity of the optimal demand of the risky asset is given by

dw .w0 / w0 e /.Qr rf / rf w0
E u00 .W
D 00
dw0 w .w0 / E u .W e /.Qr rf /2 w .w0 /

E u00 .We /W
e .Qr rf /
D1 ;
w .w0 /E u00 .W e /.Qr rf /2
where the second equality follows from W e D w0 rf C w .Qr rf /, see (3.1).

Therefore, since it is assumed that w .w0 / > 0 for all w0 2 RC and the function u
is strictly concave, it holds that

dw .w0 / w0
sign 1 e /W
D sign E u00 .W e .Qr rf / :

dw0 w .w0 /
Then, a reasoning analogous to that used in the proof of Proposition 3.3 yields that
e /W
EŒu00 .W e .Qr rf / 0 (see Exercise 3.7) if the mapping x 7! rur .x/ is decreasing.
To complete the proof, it suffices to note that

d w =w0 w .w0 / dw .w0 / w0
D 1 :
dw0 w20 dw0 w .w0 /
t
u
In particular, if an agent is characterized by a constant relative risk aversion
coefficient (CRRA utility function), then the proportion of wealth invested in the
risky asset is independent of the initial wealth.
We now consider comparative statics results concerning the behavior of the risky
asset demand with respect to changes in the risk free rate rf and in the expected
return EŒQr . In general, when rf and/or EŒQr change, there is an interplay between a
substitution and a income effect on the optimal demand of the risky asset.1 Which
one of the two effects will prevail depends on further assumptions on the utility
function, as shown in the following proposition (see Fishburn & Porter [707]).
If w .w0 / > 0 for all w0 2 RC , then the following hold:
(i) if u is DARA and the distribution of rQ rf remains unchanged as rf changes,
then @w =@rf 0;
(ii) if u is IARA and 0 w .w0 / w0 , then @w =@rf 0;
(iii) if rur .x/ 1, for all x 2 RC , and rQ is non-negative, then @w =@rf 0;
(iv) if u is DARA and the distribution of rQ EŒQr does not change as EŒQr changes,
then @w =@EŒQr 0.
Proof For brevity of notation, we denote zQ WD rQ rf .
.i/: Since the distribution of zQ is assumed not to change when rf changes, the
implicit function theorem applied to the optimality condition (3.4) gives
(omitting the dependence of the optimal demand on the initial wealth w0 )

@w E u00 w0 rf C w zQ zQ
D w0 00 :
@rf E u w0 rf C w zQ zQ2
Since u is strictly concave, the numerator of the above expression is non-

negative if u is DARA, as shown in the proof of Proposition 3.3.
.ii/: The implicit function theorem applied to the optimality condition (3.4) allows
us to establish that

@w E u00 w0 rf C w zQ zQ .w0 w / E u0 w0 rf C w zQ
D
@rf E u00 w0 rf C w zQ zQ2

E rua w0 rf C w zQ u0 w0 rf C w zQ zQ .w0 w /
D (3.8)
E u00 w0 rf C w zQ zQ2

E u0 w0 rf C w zQ
C 00 :
E u w0 rf C w zQ zQ2
The denominator of the above expression is negative, since u is strictly con-

cave. The last term of (3.8) is therefore also negative, since u is strictly
increasing, and is related to the substitution effect on the demand of the risky
asset: if the risk free rate increases, then the risk free asset becomes more
1
In general terms, the substitution effect is related to changes in the relative prices of goods,
while the income effect is due to changes in purchasing power. Technically, the substitution effect
consists in a price change that alters the slope of the budget constraint but leaves the consumer on
the same indifference curve, while the income effect represents the change in purchasing power
due to price movements (a parallel shift of the budget constraint).
attractive and, hence, the optimal demand of the risky asset decreases. On the
contrary, the numerator of the term on the second line of (3.8) is related to
the income effect associated with a change in the risk free rate. Similarly as
in the proof of Proposition 3.3, if the utility function u is IARA, then it can
be shown that
e /Qz rua .w0 rf /Qz;

rua .W
and, hence,

E rua w0 rf C w zQ u0 w0 rf C w zQ zQ rua .w0 rf /E u0 w0 rf C w zQ zQ D 0;
where the last equality follows from the optimality condition (3.4). Since we
assume that w 2 Œ0; w0 , part .ii/ is established.
.iii/: Note that the numerator in (3.8) can be rewritten as

e / 1 C rua .W
E u 0 .W e /Qz.w0 w /

D E u 0 .W e / 1 rua .We /W
e C rua .We /w0 rQ

D E u 0 .W e /.1 rur .W
e / C rua .We /w0 rQ :
e / 1 and rQ is non-negative, then the above quantity is also non-

If rr .W
negative. Since u is strictly concave and, hence, the denominator in (3.8) is
negative, part .iii/ then follows.
.iv/: Let us denote xQ WD rQ EŒQr , so that rQ rf D xQ C EŒQr rf . Then, the implicit
function theorem applied to the optimality condition (3.4) gives

@w e /w .Qx C EŒQr rf / C u0 W
E u00 .W e /
D :
@EŒQr e /.Qx C EŒQr rf /2
E u00 .W
If the utility function u is DARA, then part .iv/ follows by analogous

arguments as for part .ii/, making use of the fact that w .w0 / > 0 for all
w0 2 RC , together with the strict concavity of u. t
u
In view of Proposition 3.1, the assumption w .w0 / > 0, for all w0 2 RC ,
appearing in the statement of Proposition 3.5 is satisfied if and only if EŒQr rf > 0.
In all the cases considered in Proposition 3.5, the impact of a change in the return
risk premium EŒQr rf on the optimal demand of the risky asset is determined
by the interplay of two effects: the substitution and the income effect. While the
substitution effect has a clear sign, the income effect depends on the coefficient of
absolute risk aversion. For instance, in part .i/ of Proposition 3.5, the fact that u is
DARA implies that the demand of the risky asset increases as rf increases because,
under the assumption of part .i/, the distribution of rQ rf remains unchanged with
no substitution effect. On the contrary, in the case considered in part .ii/, both the
income and the substitution effects play a role. In particular, if the utility function
u is IARA, then both effects go in the same direction and lead to a decrease of the
demand of the risky asset in correspondence of an increase of the risk free rate rf . If
in part .ii/ of Proposition 3.5 we consider a DARA utility function, then the income
effect would go in the opposite direction of the substitution effect and the overall
impact on the demand of the risky asset could be positive or negative. The situation
considered in part .iv/ is also similar and shows that, if u is DARA, then both the
income and the substitution effects go in the same direction, leading to an increase
of the investment in the risky asset as long as the expected return EŒQr increases,
while leaving unchanged the distribution of rQ EŒQr .
The Case of Multiple Risky Assets
In general, the comparative statics results obtained in a single risky asset setting
cannot be easily extended to the case of N > 1 assets. Indeed, without further
assumptions on the distribution of asset returns or on the utility function, general
results on optimal portfolio choices with multiple risky assets are less informative.
In particular, it is difficult to obtain an explicit relation between the optimal portfolio
and the assets’ risk premia.
We start by providing a simple necessary and sufficient condition for an arbitrary
risk averse agent to invest the total amount of wealth in the risk free asset. We denote
by w D .w1 ; : : : ; wN /> 2 RN the optimal portfolio.
Then wn D 0 for all n D 1; : : : ; N if and only if EŒQrn D rf for all n D 1; : : : ; N.
Proof The optimality condition (3.4) gives that
" #
X
N

0
E u w0 rf C wk .Qrk rf / rQn rf D 0; for all n D 1; : : : ; N:
kD1
If EŒQrn D rf holds for all n D 1; : : : ; N, then the portfolio w D 0 2 RN satisfies

the above optimality condition. As a matter of fact, for w D 0, it becomes

u0 .w0 rf / EŒQrn rf D 0; for all n D 1; : : : ; N
which is always satisfied if EŒQrn D rf , for all n D 1; : : : ; N. By the strict concavity

of u, the portfolio w is then the unique optimal solution to the portfolio choice
problem (compare with Exercise 3.3). Conversely, if wn D 0 for all n D 1; : : : ; N,
then the same optimality condition, together with the strict increasingness of u,
shows that we must have EŒQrn D rf for all n D 1; : : : ; N. t
u
According to the above proposition, any risk averse agent will invest the total
amount of wealth in the risk free asset if and only if every risky asset has a null
return risk premium. However, this result is not really interesting for the sufficiency
part, since it is indeed uncommon for every risky asset traded in the market to have a
null risk premium. Instead, it is interesting for its converse implication: in reality, it
is observed that only a small fraction of market participants invests in stocks, while
many investors hold all their wealth in the bank account. According to the above
proposition, it is difficult to reconcile this behavior with the hypothesis of perfect
rationality.
Let us consider a risk averse agent. It is reasonable to guess that the expected
return of his optimal portfolio will be greater or equal than the risk free rate rf ,
since the agent can always choose to invest all his wealth in the risk free asset. This
intuition is formalized in the following proposition. We recall that W e denotes the
optimal wealth resulting from the solution of problem (3.2).
Proposition 3.7 Let u be a strictly increasing and concave utility function and
e =w0 ,
denote by rQ the random return of the optimal portfolio w , i.e., rQ D W

for w0 > 0. Then EŒQr rf .
Proof We argue by contradiction. Suppose that EŒQr < rf . Then:

e / D EŒu.w0 rQ / u EŒw0 rQ < u.w0 rf /;
U.W
where the first inequality follows from Jensen’s inequality and the second from the
strict increasingness of the utility function u together with the fact that w0 > 0. This
contradicts the optimality of w . t
u
Note also that, if in Proposition 3.7 we assume that the utility function u is strictly
concave, then the same result holds true with a strict inequality, i.e., EŒQr > rf .
As shown in Proposition 3.1, in the context of simple financial markets with
a single risky asset the demand of the risky asset is positive, negative or null
according to whether the return risk premium of the asset if positive, negative or
null, respectively. In the presence of multiple risky assets, we no longer have such a
general and clear relation between risky assets’ demand and risk premia.
In the absence of trading restrictions, if an agent is risk neutral (i.e., with a linear
utility function), then he will invest in an unlimited way on the asset with the highest
return risk premium, thus excluding the existence of an optimal portfolio. However,
if a risk neutral agent is subject to a short sale constraint and cannot borrow from the
bank account, then his optimal portfolio is well-defined and will consist of investing
the total amount of wealth in the asset with the highest risk premium. The rationale
of this result is very simple: a risk neutral agent only cares about the expected wealth
and the latter is linear in the assets’ expected returns. Therefore, a risk neutral agent
will invest only in the asset with the highest risk premium.
On the contrary, if an agent is risk averse, then he aims at maximizing
the expected value of an increasing and concave transformation (i.e., the utility
function) of the random wealth obtained at t D 1. In this case, the agent will not
only consider the risk premia of the assets, but also the distribution of the random
wealth. Note that only in some specific cases (see Sect. 2.4) the expected utility
is increasing in the expected value of the wealth and decreasing in its variance. In
a general context, it may happen that a risk averse agent will choose to invest a
positive amount of wealth in a risky asset with a negative return risk premium if
that asset helps to reduce the riskiness of the overall portfolio. This suggests that,
in a general setting, the diversification and insurance principles play a key role in
portfolio selection.
Proposition 3.8 Let u be a strictly increasing and concave utility function. Then,
for any w0 > 0, the following hold, for all n D 1; : : : ; N:

e 0 if and only if EŒQrn rf 0;
(i) Cov rQn ; u0 .W

e / 0 if and only if EŒQrn rf 0.
(ii) Cov rQn ; u0 .W
Proof By the optimality condition (3.4) and recalling that for two random variables
xQ 1 ; xQ 2 it holds that Cov.Qx1 ; xQ 2 / D EŒQx1 xQ 2 EŒQx1 EŒQx2 , we obtain

e /; rQn C EŒQrn EŒu0 .W
Cov u0 .W e / D rf EŒu0 .W
e /; for all n D 1; : : : ; N:
This optimality condition implies the following return risk premium implicit in the
optimal portfolio choice:

e /; rQn
Cov u0 .W
EŒQrn rf D ; for all n D 1; : : : ; N:
e /
EŒu0 .W
The denominator on the right-hand side is positive, due to the fact that u is
increasing, and the claim follows. u
t
While it is not possible to derive general relations between asset risk premia and
optimal portfolios, Proposition 3.8 establishes a relation between the expected risk
premium of an asset and the covariance of the random return of that asset with
the marginal utility of the optimal wealth. Note that this result relates risk premia
to optimal wealth and not to optimal portfolios. Moreover, the relation is not fully
explicit due to the presence of the marginal utility function u0 . The main message
of Proposition 3.8 is that a risk averse agent will invest in such a way that the
covariance of the marginal utility of the optimal wealth and the return of an asset is
negative if and only if the risk premium of that asset is positive. In this sense, the
result yields an implicit relation between asset returns and optimal wealth.
In some special cases, it is possible to obtain more explicit results on the relation
between risk premia and optimal wealth (see Exercises 3.4 and 3.5). In particular,
this is the case if the utility function is quadratic or if asset returns are distributed as
a multivariate normal random variable. Let us first consider the case of a quadratic
utility function u.x/ D x b2 x2 , with b > 0. It holds that:

Cov rQn ; u0 .w0 rQ / D bw0 Cov rQn ; rQ :
Assuming that the expected marginal utility of optimal wealth is strictly positive, i.e.
EŒu0 .w0 rQ / D 1b w0 EŒQr > 0, and w0 > 0, we have that, for all n D 1; : : : ; N,
(i) EŒQrn rf 0 if and only if Cov.Qrn ; rQ / 0;
(ii) EŒQrn rf 0 if and only if Cov.Qrn ; rQ / 0.
In the case where asset returns are distributed as a multivariate normal random
variable, we can rely on the following result, known as Stein’s lemma.
Lemma 3.9 Let .Qx1 ; xQ 2 / be a bivariate normal random vector and let g W R ! R
be a differentiable function. Then, as soon as the expectations are well-defined, it
holds that

Cov g.Qx1 /; xQ 2 D EŒg0 .Qx1 / Cov.Qx1 ; xQ 2 /:
Letting g./ be the marginal utility function u0 ./, we can then obtain a result
analogous to that obtained above in the case of a quadratic utility function.
These results directly relate the risk premium of an asset to the covariance
between its return and the optimal wealth. The key feature that allows us to obtain
such a relation is represented by the fact that, both in the case of a quadratic utility
and of normally distributed returns, the expected utility function is an increasing
function of the expected wealth and a decreasing function of its variance (provided
that EŒWe 1=b in the case of a quadratic utility function). This observation allows
us to explain the role of the insurance and diversification principles in the context
of portfolio choice problems. Indeed, an asset whose return is positively correlated
with the optimal wealth does not contribute to reduce the variability of the optimal
wealth and, therefore, a risk averse agent will require a positive risk premium for
investing in such an asset. Conversely, an asset whose return is negatively correlated
with the optimal wealth contributes to reduce the variability of the optimal wealth,
so that a risk averse agent may accept a negative risk premium because such an
asset provides insurance against wealth variability. In other words, an agent aiming
to reduce the overall variance will choose a portfolio such that wealth is positively
correlated with assets with a positive risk premium and negatively correlated with
assets with a negative risk premium. We remark that all these results only concern
the optimal wealth and not the optimal portfolio itself. Indeed, there is no direct and
general relation between optimal portfolios and asset risk premia.
Concerning the diversification principle, risk aversion implies that an agent will
diversify his wealth among several sources of risk, instead of investing all his wealth
in a single risky asset. However, perfect diversification is a rare situation. In a setting
without a risk free asset, the optimal portfolio is perfectly diversified (i.e., the agent
will invest the same amount of wealth in each asset) if the risky asset returns are
independent and identically distributed (i.i.d.), as shown in the next result (see
Samuelson [1492]).
Proposition 3.10 Suppose that the random returns .Qr1 ; : : : ; rQN / of N risky assets
are independent and identically distributed and suppose that there exists no risk free
asset. Then, the optimal portfolio of every risk averse agent is given by wn D w0 =N,
for all n D 1; : : : ; N.
Proof The proof is based on a stochastic dominance argument. Denote by
X
N
e D w0 1
W rQn
N nD1
the wealth associated with the fully diversified portfolio, starting from a given initial
wealth w0 . We show that the fully diversified portfolio dominates any other portfolio
according to the second order stochastic dominance criterion.
PN Indeed, let w 2 RN be
any other portfolio investing in the N assets, so that nD1 wn D w0 and denote by
e 0 the random wealth associated to such a portfolio. The random variable W
W e 0 can
be equivalently written as follows:
N
e0 D W
e C
X w0
W wn rQn :
nD1
N
e e0
In view
Pof Proposition
2.10, in order to show that W
SSD W , it suffices to prove
that E N
nD1 wn w0
Q
r n j e
W
D 0. This can be easily shown as follows:
N
"N ˇ # X
w0 ˇˇ e
N
X w0 e
E wn rQn ˇW D wn EŒQrn jW
nD1
N nD1
N
N
e
X w0
D EŒQr1 jW wn D 0;
nD1
N
where we have used the fact that the random returns .Qr1 ; : : : ; rQN / are i.i.d. and, hence,
e D EŒQr1 jW
EŒQrn jW e , for every n D 1; : : : ; N. t
u
Closed Form Solutions and Mutual Fund Separation
As we have mentioned in the previous section, in the context of a financial

market with multiple risky assets, it is difficult to establish general and explicit
results on optimal portfolio choices. In what follows, we aim at investigating some
special cases where, under additional assumptions on the utility function or on the
distribution of the asset returns, explicit results can be obtained. In particular, we
first focus on the two classical cases of quadratic and exponential utility functions
with multivariate normal returns. Then, in a more general setting, we investigate
the class of utility functions for which only the overall amount invested in the risky
assets, and not the composition of the optimal portfolio, changes as the initial wealth
changes (mutual fund separation).
Let us consider the case of a quadratic utility function u.x/ D x b2 x2 , for some
parameter b > 0. In this case, the optimality condition (3.4) becomes

e /.Qrn rf / D 0;
E .1 bW for all n D 1; : : : ; N:
Using the definition of covariance, the above optimality condition implies that, for
all n D 1; : : : ; N,
X
N
e ; rQn / D b
1 b w0 EŒQr EŒQrn rf D b Cov.W wk Cov.Qrk ; rQn /:
kD1
Equivalently, in vector notation, the optimal portfolio w 2 RN satisfies
1 b w0 EŒQr 1
w D V .e rf 1/; (3.9)
b
where we have used the assumption that the risky assets are non-redundant (i.e., the
variance-covariance matrix V is positive definite) and 1 denotes the column vector
.1; : : : ; 1/> 2 RN . Note that, up to a multiplicative factor, the optimal portfolio
w is given by the inverse of the variance-covariance matrix multiplied by the
vector of the risky assets’ return risk premia. If w0 EŒQr < 1=b, then the scale
factor .1 b w0 EŒQr /=b is strictly positive and corresponds to the reciprocal of
the coefficient of absolute risk aversion evaluated at the expected optimal wealth
e D w0 rQ ). Note that (3.9) characterizes the optimal portfolio only in a semi-
(W
explicit form.
In the simpler case of a single risky asset (N D 1), with expected return EŒQr and
variance 2 .Qr/, the optimal portfolio can be written in a fully explicit form:

.1 b w0 rf / EŒQr rf
w D 2 : (3.10)
b 2 .Qr/ C EŒQr rf
Under the assumption that w0 rf < 1=b, the optimal investment in the risky asset
jw j is decreasing with respect to the risk-aversion parameter b, to the initial wealth
w0 and to the variance 2 .Qr/.
Exponential Utility Function and Normal Returns
Let us now suppose that the risky asset returns rQ are distributed as a multivariate
random variable with mean vector e and variance-covariance matrix V and, as in
Sect. 2.4, let us consider an exponential utility function u.x/ D 1a exp.ax/, for
a given risk aversion parameter a > 0. Recall that, for any vector z 2 RN , the
random variable exp.z> rQ / is log-normally distributed with mean exp.z> eCz> Vz=2/.
The random wealth W e obtained at t D 1 by investing in a portfolio w 2 RN is given
by We D w0 rf C w .Qr rf 1/ and, since the normal distribution is stable with respect
>
to linear transformations, W e is also normally distributed, with mean and variance

given by
e D w0 rf C w> .e rf 1/
EŒW and e D w> Vw:
2 .W/
Due to the above observations, maximizing the expected exponential utility is

equivalent to the following optimization problem:
a
max w> .e rf 1/ w> Vw :
w2RN 2
Since the matrix V is assumed to be positive definite, the optimal portfolio w is

then explicitly determined as
1 1
w D V .e rf 1/: (3.11)
a
Note that, since the exponential utility function is CARA (see section “Classes of
Utility Functions”), the optimal investment in the risky assets does not depend on
the initial wealth w0 (compare with Proposition 3.3). As can be seen from (3.11),
the investment in the risky assets is decreasing in the coefficient of absolute risk
aversion a.
The appearance of the matrix V 1 makes the relation between the optimal
portfolio w and the assets’ risk premia e rf rather complex, due to the presence of
correlation effects. Indeed, if the random returns of the risky assets are uncorrelated
(and, hence, due to the normality assumption, independent), then the matrix V
reduces to a diagonal matrix whose elements are given by the variances of the
individual returns 2 .Qr1 /; : : : ; 2 .QrN / . Hence, the inverse matrix V 1 has elements
1
Vn;n D 1= 2 .Qrn /, for n D 1; : : : ; N. In this special case, we recover the result
obtained in the case of a single risky asset in Proposition 3.1: wn > 0 if and only if
EŒQrn > rf , for every n D 1; : : : ; N.
In the simpler case of a single risky asset with return rQ , it holds that
1
w D 2
EŒQr rf : (3.12)
a .Qr /
We see that the amount of wealth invested in the risky asset is increasing in the
risk premium EŒQr rf , decreasing in the coefficient of absolute risk aversion a
(compare with Proposition 3.3) and its absolute value is decreasing in the variance
2 .Qr/. Moreover, the optimal portfolio does not depend on the initial wealth w0 .
In the present context, the diversification and insurance principles drive the
optimal portfolio decisions. To illustrate the effects of these two principles in a
simple setting, let us consider the case of two risky assets with returns rQ1 ; rQ2 jointly
distributed as normal random variables, with expected returns e1 ; e2 , variances

12 ; 22 and correlation coefficient 2 .0; 1/. Without loss of generality, we suppose
that 12 22 . The variance-covariance matrix Vand its inverse V 1 then become

" 1 #
12 1 2 12 .12 /
2/
1 1 2 .1
VD and V D 1 :
1 2 22 1 2 .1 2/
22 .12 /
By formula (3.11), up to a scale factor of 1=a, the optimal portfolio is given by
1
w1 D .e1 rf / .e2 rf /;
12 .1 2
/ 1 2 .1 2 /
1
w2 D 2
.e1 rf / C 2 .e2 rf /:
1 2 .1 / 2 .1 2 /
From the above expressions, it is evident that the correlation coefficient , in

particular its sign, plays a crucial role in the determination of the optimal portfolio,
see Fig. 3.1. Let us first consider the case when > 0. Then, it is easy to see that:
1
w1 > 0 ” e1 rf > .e2 rf /;
2
1
w2 > 0 ” e1 rf < .e2 rf /:
2
As a consequence, the optimal portfolio is diversified (i.e., w1 ; w2 > 0) if and only if
1 1
.e2 rf / < e1 rf < .e2 rf /: (3.13)
2 2
If at least one of the two assets has a negative risk premium, then at least one
asset will be sold short and the optimal portfolio will not be diversified. In other
words, having positive risk premia represents a necessary condition in order to have
a diversified optimal portfolio. The optimal portfolio is diversified if both assets are
characterized by a positive risk premium and if the difference between the two risk
premia is not too large, as follows from (3.13).
Let us now consider the case of negative correlation, i.e., < 0. In this case, it
holds that
1
w1 > 0 ” e1 rf > .e2 rf /;
2
1
w2 > 0 ” e1 rf > .e2 rf /:
2
In this case, the fact that both assets have a positive risk premium is a sufficient
condition to have a diversified optimal portfolio. However, the optimal portfolio
Fig. 3.1 Portfolios in the risk premium space (positive correlation, top; negative correlation,
bottom)
may be diversified also in the case where the risk premium of one of the two assets
is negative (but not too large in absolute value). In particular, note that the optimal
portfolio is more easily diversified in the case of negative correlation, as can be seen
by comparing the two cases of positive and negative correlation. This is illustrated
in Fig. 3.1.
Mutual Fund Separation
In a setting with a single risky asset, we have presented several comparative statics
results on the optimal demand of the risky asset. As already mentioned, analogous
general results cannot be established in the case of multiple risky assets. In
particular, there is no simple and explicit connection between the optimal portfolio
and changes in the initial wealth.
An exception is represented by the case where an agent always chooses to invest
in the same portfolio of risky assets, changing only the proportion of wealth invested
in that portfolio as a whole. In this case, this “invariant” portfolio composed by
the risky assets represents a mutual fund and we say that a two mutual funds
(monetary) separation result holds. In other words, for any level of wealth, an agent
will always choose to hold a linear combination of the risk free asset and of the
mutual fund. For any level of wealth, the portfolio choice problem consists only in
determining the proportion of wealth invested in the mutual fund, without modifying
the composition of the mutual fund itself.
The following proposition shows that two mutual funds monetary separation
always holds in the case of utility functions characterized by hyperbolic absolute
risk aversion (HARA utility functions, see section “Classes of Utility Functions”),
in the context of a general financial market with N risky assets and a risk free asset.
Proposition 3.11 Let u be a strictly increasing and strictly concave utility function
with hyperbolic risk aversion, so that rua .x/ D 1=.a C bx/, for suitable parameters
a; b 2 R. Then the optimal portfolio w 2 RN is a linear function of the initial
wealth w0 , i.e.,
w D .a C b w0 rf /;
where the vector 2 RN is independent of w0 .

Proof As shown in part b) of Exercise 2.10, a utility function u belongs to the
HARA class if and only if it is of one of the three following forms, for suitable
parameters a; b:
1 b1 x
u.x/ D .a C bx/ b ; u.x/ D log.x C a/; u.x/ D ae a :
b1
This implies that, if u belongs to the HARA class, then the marginal utility u0 takes
one of the two following forms (with b D 1 in the case of logarithmic utility):
1
u0 .x/ D .a C bx/ b ; u0 .x/ D e a :
x
For wealth w0 , let us denote by w .w0 / the associated optimal portfolio. In the
remaining part of the proof, following LeRoy & Werner [1191, Theorem 13.6.1],
we fix an arbitrary wealth wN 0 and aim at showing that the optimal portfolio w .w0 /
associated to any other value of wealth w0 is given by
a C b w0 rf
w .w0 / D w .wN 0 /:
aCbw N 0 rf
The result will then follow by defining WD w .wN 0 /=.a C b w N 0 rf /.

Let us first consider the case where u0 .x/ D .a C bx/1=b , for some parameter
b > 0. In this case, for wealth wN 0 , the optimality condition (3.4) takes the form, for
all n D 1; : : : ; N,
2 ! 1b 3
X
N
E4 a C bw
N 0 rf C b wn .wN 0 /.Qrn rf / .Qrn rf /5 D 0:
nD1
Dividing the above equation by .a C b w N 0 rf /1=b and multiplying by

.a C b w0 rf /1=b , we get, for all n D 1; : : : ; N,
2 ! 1b 3
X
N
a C b w0 rf
E 4 a C b w0 rf C b wn .wN 0 /.Qrn rf / .Qrn rf /5 D 0;
nD1
aCbw
N 0 rf
aCb w r
thus showing that the portfolio w .w0 / D aCb wN 00 rff w .wN 0 / satisfies the optimality
condition for an arbitrary wealth w0 , thus proving the claim.
In the case where u0 .x/ D exp.x=a/, the reasoning is analogous. Indeed, the
optimality condition (3.4) gives, for the fixed level wN 0 of wealth,
" ! #
1X
N
1
E exp wN 0 rf w .w
N 0 /.Qrn rf / .Qrn rf / D 0; for all n D 1; : : : ; N:
a a nD1 n

Then, multiplying the above expression by exp .w0 wN 0 /rf =a gives
" ! #
1X
N
1
E exp w0 rf w .w
N 0 /.Qrn rf / .Qrn rf / D 0; for all n D 1; : : : ; N:
a a nD1 n
This shows that the portfolio w .w0 / D w .wN 0 / solves the optimality condition for
an arbitrary wealth w0 , thus proving the claim. t
u
The above result implies that ratios of the optimal demands of the risky assets do
not depend on the level of initial wealth. Indeed, for any w0 , it holds that
wn .w0 / n
D ; for every n; m 2 f1; : : : ; Ng:
wm .w0 / m
In other words, for different levels of initial wealth, the optimal portfolios differ
only by the amount of wealth invested in the risky assets as a whole and not
by the composition of the portfolio of risky assets. Indeed, the optimal portfolio
corresponding to an initial wealth w0 is given by
>
w .w0 / D .a C b w0 rf /1 ; : : : ; .a C b w0 rf /N 2 RN ;
while the amount of wealth invested in the risk free asset is given by
X
N
w0 .a C b w0 rf / n :
nD1
As a further consequence, for the HARA class of utility functions, the comparative
statics results obtained in the case of a single risky asset can be extended to the case
of N risky assets by considering as a single risky asset the mutual fund, whose
composition does not change as wealth changes. In particular, if a D 0 in the
representation rua .x/ D 1=.a C bx/, then the relative risk aversion coefficient is
constant (CRRA utility), thus implying that the proportion of wealth invested in the
mutual fund of risky assets does not change with wealth, due to Proposition 3.4.
On the other hand, the exponential utility function is characterized by an increasing
coefficient of relative risk aversion, so that the proportion of wealth invested in the
mutual fund will decrease as wealth increases.
3.2 Mean-Variance Portfolio Selection
In this section, we present one of the most popular and historically important
portfolio selection approach, consisting in the determination of the portfolio which
achieves the minimum variance among all portfolios with a given expected return.
This problem has been originally introduced by Markovitz [1303] and represents
one of the cornerstones of modern portfolio theory. Furthermore, it is also at the
basis of the CAPM and is related to classical asset pricing models, as will be
discussed in Chap. 5.
3.2 Mean-Variance Portfolio Selection 77
Some of the main reasons for the popularity of the mean-variance approach
are represented by its analytical tractability and by the possibility of deriving
clear empirical implications. However, as we have already discussed in Sect. 2.4,
mean-variance preferences are consistent with expected utility maximization only
in special cases, notably in the case of a quadratic utility function (which however
suffers from some drawbacks) and in the case of normally distributed returns.
Mean-variance portfolio selection is also related to second order stochastic
dominance. Indeed, in view of Proposition 2.10, if two portfolios w1 and w2 have
the same expected return EŒQr and portfolio w1 dominates portfolio w2 according to
the second order stochastic dominance criterion, then portfolio w1 must have a lower
variance than w2 . In this sense, if there exists a portfolio w with expected return EŒQr
that dominates according to the second order stochastic dominance criterion any
other portfolio with the same expected return, then the portfolio w also achieves
the minimum variance among all portfolios with expected return EŒQr .
In this section, we shall first consider the case of an economy with N 2 N risky
assets with random returns .Qr1 ; : : : ; rQN / and then the case of an economy with N
risky assets together with a risk free asset with deterministic return rf .
The Case of N Risky Assets
Let us consider an economy with N > 1 risky assets with returns .Qr1 ; : : : ; rQN /. As in
the previous sections, we denote by e D .e1 ; : : : ; eN /> 2 RN the vector of expected
returns and by V 2 RNN the variance-covariance matrix of the random vector rQ ,
assumed to be positive definite.
P In the present context, a portfolio is represented by
a vector w 2 RN such that NnD1 wn D 1, where wn represents the proportion of
wealth invested in the n-th asset (equivalently, the initial wealth is normalized to
1), for n D 1; : : : ; N. We denote by N the set of all portfolios. PFor any portfolio
N
w 2 N , we denote by rQw its random return, given by rQw D nD1 wn rQn , with
expected value and variance given by
X
N X
N X
N
EŒQrw D w> e D wn en and 2 .Qrw / D w> Vw D wn wm Cov.Qrn ; rQm /:
nD1 nD1 mD1
Note also that the condition w 2 N can be simply expressed as w> 1 D 1, where
1 WD .1; : : : ; 1/> 2 RN . We base our presentation on the derivation of the portfolio
frontier proposed in Merton [1329] (compare also with Huang & Litzenberger [971,
Chapter 3]).
We are interested in determining the portfolio w 2 N which achieves the
minimum variance among all possible portfolios with a given expected return
2 R. The optimal portfolio w is determined as the solution to the following
problem:
min w> Vw; (3.14)

w2RN
under the constraints
w> e D and w> 1 D 1: (3.15)
Note that the two constraints simply amount to require that the vector w is a
portfolio (i.e., w 2 N ) and yields the required expected return . The solution
to the above problem is given in the following theorem.
Theorem 3.12 Let .Qr1 ; : : : ; rQN / be random returns with expected value e 2 RN and
positive definite variance-covariance matrix V 2 RNN . Then, for any 2 R, the
solution to the constrained variance minimization problem (3.14) is given by
w D g C h ; (3.16)
where
BV 1 1 AV 1 e CV 1 e AV 1 1
gD ; hD ;
D D
A D 1> V 1 e; B D e> V 1 e; C D 1> V 1 1; D D BC A2 :
Proof Clearly, problem (3.14) is equivalent to the minimization of w> Vw=2 over
all w 2 RN subject to the constraints (3.15). To this minimization problem, we can
associate the Lagrangian
1 >
L.w; ; / WD w Vw C . w> e/ C .1 w> 1/:
2
The necessary conditions for the optimality of w are given by
@L.w ;; /
@w
D Vw e 1 D 0; (3.17)
@L.w ;; /
@
D w> e D 0; (3.18)
@L.w ;; /
@
D 1 w> 1 D 0: (3.19)
In (3.17), @L.w ; ; /=@w denotes the gradient of the function L evaluated at

.w ; ; /. Note that, since the matrix V is assumed to be positive definite, then
the first order conditions (3.17)–(3.19) are also sufficient for the solution to the
variance minimization problem (3.14). We now determine explicitly the optimal
portfolio w 2 N . First, by (3.17),
w D V 1 e C V 1 1: (3.20)
Replacing this expression in (3.18) and in (3.19), we obtain

CA
D D ; (3.21)
BA
D D
; (3.22)

w D g C h; (3.23)
where A; B; C; D and g; h are defined as in the statement of the theorem. Observe

that, since the inverse of a positive definite matrix is also positive definite, it holds
that B > 0 and C > 0. Moreover, D > 0, since BD D .Ae B1/> V 1 .Ae B1/ > 0
and B > 0. t
u
Theorem 3.12 explicitly characterizes the optimal portfolio w which minimizes
the variance among all portfolios with a given expected return . Note that the
optimal portfolio is unique and is explicitly given by (3.16), where the terms
A; B; C; D; g; h are derived by the data of the problem. Theorem 3.12 solves the
problem of an agent who is interested in minimizing the risk (as measured by
the variance) of his portfolio while ensuring a given level of expected return.
We call portfolio frontier (denoted by PF) the set of all portfolios which solve
problem (3.14) as the required expected return varies. Since the optimality
conditions employed in the proof of Theorem 3.12 are both necessary and sufficient,
it follows that a portfolio belongs to the portfolio frontier if and only if it admits the
representation (3.16) for some expected return 2 R. Moreover, Theorem 3.12 also
provides the basis for the solution of several related mean-variance optimization
problems. For instance, as shown in Exercise 3.19, it can be used for determining
the portfolio maximizing the Sharpe ratio with respect to a given reference rate of
return.
Note that, as a consequence of Proposition 2.10, portfolios belonging to the
portfolio frontier cannot be dominated by other portfolios in the sense of second
order stochastic dominance. However, it is not generally true that a portfolio
belonging to the portfolio frontier dominates all other portfolios with the same
expected return according to the second order stochastic dominance criterion (see
Sect. 2.3).
In the following, we present several fundamental properties of the portfolio
frontier.
Property 1: Linear Combinations of Frontier Portfolios
Given two arbitrary portfolios w1 ; w2 belonging to the portfolio frontier, with
EŒQrw1 ¤ EŒQrw2 , any other portfolio w belonging to the portfolio frontier can
be represented as a linear affine combination of w1 and w2 .
Indeed, let us consider two frontier portfolios w1 ; w2 with expected returns
equal to 1 D 0 and 2 D 1, respectively. Then, due to Theorem 3.12, it holds
that w1 D g and w2 D g C h. Then, by Theorem 3.12, given any other frontier
portfolio w with expected return EŒQrw , we can write

w D gCh EŒQrw D 1EŒQrw gCEŒQrw .gCh/ D 1EŒQrw w1 CEŒQrw w2 ;
thus proving that w is an linear affine combination of w1 and w2 . The same
reasoning can be extended to two arbitrary portfolios w1 ; w2 belonging to the
portfolio frontier with EŒQrw1 ¤ EŒQrw2 .
Moreover, for any M 2 N, if the portfolios w1 ; : : : ; wM all belong to
the portfolio frontier, then any linear combination of w1 ; : : : ; wM with weights
summing up to one will also belong to the portfolio frontier, with expected return
given by the linear combinationP of the expected returns
PM associated to the portfolios
w1 ; : : : ; wM . Indeed, if w D M i
iD1 ˛i w with iD1 ˛i D 1, then it holds that
X
M X
M
X
M
w D ˛i wi D ˛i g C h EŒQrwi D g C h ˛i EŒQrwi :
iD1 iD1 iD1
PM
Hence, by Theorem 3.12, the portfolio w D i
iD1 ˛i w solves problem (3.14)
PM
with respect to the expected return D iD1 ˛i EŒQrwi .
Property 2: Covariance and Variance of Frontier Portfolios
For any two portfolios w1 and w2 belonging to the portfolio frontier, with expected
returns EŒQrw1 and EŒQrw2 , respectively, it holds that

1 > 2 C A A 1
Cov rQw1 ; rQw2 D .w / V w D EŒQrw1 EŒQrw2 C
D C C C
(3.24)
and, in particular, for any portfolio w belonging to the portfolio frontier,

C A 2 1 1
2 .Qrw / D EŒQrw C D C EŒQrw 2 2 AEŒQrw C B ; (3.25)
D C C D
where A; B; C; D are defined as in Theorem 3.12.

The expressions (3.24)–(3.25) can be easily obtained by means of elementary

computations, see Exercise 3.10. In particular, by (3.25), we can obtain an explicit
characterization of the portfolio wMVP 2 N which achieves the minimum
variance among all portfolios, without any constraint on its expected return. Indeed,
expression (3.25) is minimized by EŒQrw D A=C. Hence, letting D A=C (recall
that C > 0) in (3.16), the portfolio
A V 1 1
wMVP D g C h D (3.26)
C C
minimizes the variance, so that 2 .QrwMVP / D 1=C, with a corresponding expected

return of EŒQrwMVP D A=C. The second equality in (3.26) easily follows by
substituting the explicit expressions for g; h; A; C given in Theorem 3.12.
Moreover, the covariance between a frontier portfolio w1 and an arbitrary
portfolio w2 (i.e., a portfolio w2 2 N not necessarily belonging to the portfolio
frontier) can be expressed as follows, making use of the representation (3.20) of a
frontier portfolio:

Cov rQw1 ; rQw2 D .w1 /> Vw2 D 1 e> V 1 V w2 C 1 1> V 1 Vw2 D 1 EŒQrw2 C 1 ;
(3.27)
where 1 and 1 are the optimal Lagrange multipliers associated to the frontier
portfolio w1 and where EŒQrw2 is the expected return of the portfolio w2 .
Formula (3.27) can also be used to prove the following property of the minimum
variance portfolio wMVP : for any portfolio w 2 N (not necessarily belonging to the
portfolio frontier), it holds that
Cov.QrwMVP ; rQw / D 2 .QrwMVP /:
Indeed, wMVP is the solution to Problem (3.14) for the expected return D A=C,
so that, according to the notation used in Theorem 3.12, we have MVP D 0 and
MVP D 1=C. Hence, applying expression (3.27) with w1 D wMVP gives that
1
Cov rQwMVP ; rQw D MVP D D 2 .QrwMVP /: (3.28)
C
Property 3: Shape of the Portfolio Frontier
In the standard deviation - expected return plane .; / 2 .0; C1/ R, the
by a hyperbola centered in the point .0; A=C/ and
portfolio frontier is representedp
with asymptotes D A=C ˙ D=C . In the variance - expected return plane
. 2 ; / 2 .0; C1/ R, the portfolio frontier is represented by a parabola with
vertex .1=C; A=C/.
A/C

1/C
Fig. 3.2 Portfolio frontier in the standard deviation-expected return plane
Indeed, the fact that the portfolio frontier can be represented by a hyperbola in
the standard deviation - expected return plane follows from equation (3.25), which
can be equivalently rewritten in the following form, for any portfolio w 2 PF:
2
2 .Qrw / EŒQrw A=C
D 1:
1=C D=C2
The above equation defines a hyperbola in the standard deviationp - expected return
plane, centered in .0; A=C/ and with asymptotes D A=C C D=C , see Fig. 3.2.
In the variance - expected return plane, the last part of equation (3.25) clearly
shows that the portfolio frontier is represented by a parabola, see Fig. 3.3. In
particular, the vertex of the parabola coincides with the minimum variance portfolio
wMVP , which is characterized by expected return EŒQrwMVP D A=C and variance
2 .QrwMVP / D 1=C, as shown in Property 2 above. In particular, this graphical
representation of the portfolio frontier makes clear that, in order to achieve a higher
expected return, the investor needs to tolerate a greater risk (as measured by the
variance) as there is a trade-off between variance and expected return.
Fig. 3.3 Portfolio frontier in the variance-expected return plane
Property 4: Efficient Portfolio Frontier
All portfolios not dominated according to the mean-variance criterion belong to the
following subset of the portfolio frontier:
˚
EPF D w 2 PF such that EŒQrw A=C : (3.29)
As we have seen in Theorem 3.12, given any expected return 2 R, there

exists a unique portfolio w 2 N with expected return EŒQrw D which
minimizes the variance. On the contrary, as can be deduced from equation (3.25),
for a fixed value N 2 of the variance such that N 2 > 1=C D 2 .QrwMVP /, we can
find two frontier portfolios w1 ; w2 with expected returns EŒQrw1 ¤ EŒQrw2 such that
2 .Qrw1 / D 2 .Qrw2 / D N 2 . More specifically, we have EŒQrw1 > A=C > EŒQrw2 ,
where we recall that A=C is the expected return corresponding to the minimum
variance portfolio wMVP . We call efficient portfolio frontier (EPF) the subset of
the portfolio frontier defined in (3.29). In this case, w1 2 EPF and w2 … EPF.
The efficient portfolio frontier is composed of those frontier portfolios that are
not dominated according to the mean-variance criterion (see Sect. 2.4). We say
that a frontier portfolio is inefficient if it does not belong to EPF. In the case of
agents with mean-variance preferences (see Sect. 2.4), we can restrict our attention
to the efficient part of the portfolio frontier, since for every inefficient portfolio
there exists a portfolio with the same variance but with a higher expected return. In
Figs. 3.2 and 3.3, the efficient portfolio frontier is represented by the upper branch of
the hyperbola/parabola representing the whole portfolio frontier. The lower branch
contains instead inefficient portfolios.
The efficient portfolio frontier is closed under convex linear combinations,
i.e., the set of all efficient portfolios is convex. This means that a linear convex
combination of efficient frontier portfolios is a portfolio belonging to the efficient
portfolio frontier. The proof of this claim follows by a direct application of Property
1.
Property 5: Zero-Correlation Portfolio
Let w be an arbitrary frontier portfolio other than the minimum variance portfolio.
There exists a unique frontier portfolio wzc (which depends on w ) such that
Cov.Qrw ; rQwzc / D 0.
Indeed, let w be an arbitrary frontier portfolio other than the minimum variance
portfolio wMVP . We can identify the zero-correlation portfolio wzc by setting
expression (3.24) equal to zero (see Exercise 3.12), thus yielding
A D=C2
EŒQrwzc D : (3.30)
C EŒQrw A=C
By Theorem 3.12, the portfolio wzc is identified as the unique portfolio which
minimizes the variance given an expected return equal to (3.30). The portfolio wzc is
the unique frontier portfolio that achieves zero correlation with the given portfolio
w . Moreover, as can be easily deduced from (3.30) together with Property 4 above,
the portfolio wzc is efficient if and only if the portfolio w is inefficient. Indeed,
since D > 0, it holds that EŒQrwzc > A=C if and only if EŒQrw < A=C.
The zero-correlation portfolio wzc admits an interesting geometrical interpreta-
tion. Indeed, in the standard deviation - expected return plane, the expected return
of the portfolio wzc is identified by the intersection
with thevertical axis of the
tangent to the portfolio frontier at the point .Qrw /; EŒQrw . In the variance -
expected return plane, the expected return of the portfoliowzc is identified by the
intersection with the vertical axis of the line connecting 2 .Qrw /; EŒQrw to the
point .1=C; A=C/. We refer the reader to Exercise 3.14 for a proof of these claims.
The properties established above allow us to prove the following proposition
which represents one of the key results on the portfolio frontier.
Proposition 3.13 Let wq 2 N be an arbitrary portfolio (with corresponding
return rQwq ) and wp 2 N a portfolio belonging to the portfolio frontier (with
corresponding return rQwp ) other than the minimum variance portfolio wMVP . Denote
by wzc.p/ the zero-correlation portfolio with respect to wp . Then the following

hold:
(i) if wq 2 PF then
rQwq D .1 ˇqp /Qrwzc.p/ C ˇqp rQwp I
(ii) if wq … PF then
rQwq D .1 ˇqp /Qrwzc.p/ C ˇqp rQwp C Qqp ; (3.31)

where Cov rQwp ; Qqp D Cov rQwzc.p/ ; Qqp D EŒQ qp D 0 and
Cov.Qrwq ; rQwp /
ˇqp D :
2 .Qrwp /
Moreover, for any portfolio wq 2 N , it holds that
EŒQrwq D .1 ˇqp /EŒQrwzc.p/ C ˇqp EŒQrwp : (3.32)
Proof We start by proving the last part of the proposition. Let wp and wq be two
arbitrary portfolios, with wp belonging to the portfolio frontier and different from the
minimum variance portfolio wMVP . By (3.21)–(3.22) together with property (3.27),
which holds for any portfolio wq , we obtain
A EŒQrwp B D
EŒQrwq D C Cov.Qrwq ; rQwp /
C EŒQrwp A C EŒQrwp A
!
A D=C2 Cov.Qrwq ; rQwp / 1 .EŒQrwp A=C/2 D
D C 2
C
C EŒQrwp A=C .Qrwp / C D=C C EŒQrwp A

D EŒQrwzc.p/ C ˇqp EŒQrwp EŒQrwzc.p/ ;
where ˇqp is defined as in the statement of the proposition. This proves (3.32).
In order to prove part .ii/ of the proposition, observe that the relationship between
the random variables rQwq , rQwp and rQwzc.p/ can always be written as
rQwq D ˛ C ˇ1 rQwzc.p/ C ˇ2 rQwp C Qqp ;
where ˇ1 and ˇ2 are obtained as the regression coefficients of rQwq on .Qrwzc.p/ ;rQwp /
and Qqp is a random variable satisfying Cov rQwp ; Qqp D Cov rQwzc.p/ ; Qqp D
EŒQ qp D 0. Note that, since the two random variables rQwp and rQwzc.p/ are uncorrelated
(Property 5), we have
ˇ2 D Cov.Qrwq ; rQwp /= 2 .Qrwp / D ˇqp and ˇ1 D ˇq zc.p/ :

Moreover, note that relation (3.32) can also be written taking wzc.p/ as the reference
portfolio, instead of the portfolio wp , thus yielding
EŒQrwq D .1 ˇq zc.p/ /EŒQrwp C ˇq zc.p/ EŒQrwzc.p/ ; (3.33)
where we have used the fact that the zero-correlation portfolio with respect to wzc.p/
is the portfolio wp . Comparing (3.32) with (3.33), we obtain ˇq zc.p/ D 1 ˇqp , so
that ˇ1 D 1 ˇqp and
rQwq D ˛ C .1 ˇqp /Qrwzc.p/ C ˇqp rQwp C Qqp :
Finally, taking the expectation of the last expression and comparing to (3.32), we
obtain that ˛ D 0. We have thus proved part .ii/ of the proposition. Part .i/ then
follows from part .ii/ together with Property 1, since any frontier portfolio wq can
be written as the linear affine combination of two arbitrary frontier portfolios, in
this case wp and wzc.p/ . This implies that, if wq 2 PF, then the random variable Qqp is
identically equal to zero. t
u
In view of the above proposition, we have established the following properties:
• The return of any portfolio wq (not necessarily belonging to the portfolio frontier)
can be written as a linear combination of the return of an arbitrary portfolio wp
belonging to the portfolio frontier and the return of the frontier portfolio wzc.p/
uncorrelated with wp plus a random variable Qqp with zero mean, uncorrelated
with the two frontier portfolios wp and wzc.p/ . Moreover, if (and only if) the
portfolio wq belongs to the frontier, the last component is null.
• The coefficients of the linear combination (3.31) can be interpreted as the
coefficients of the linear regression of rQwq on rQwp and rQwzc.p/ .
• The expected return of any portfolio wq (not necessarily belonging to the
portfolio frontier) is given by the linear affine combination of the expected return
of an arbitrary portfolio wp belonging to the portfolio frontier and of the expected
return of the frontier portfolio wzc.p/ uncorrelated with wp , with weights given by
ˇqp and 1 ˇqp , respectively.
• The coefficient ˇqp is a linear combination of the ˇ coefficients associated to the
individual N risky assets with respect to the portfolio wp . Indeed, for any portfolio
wq D .w1 ; : : : ; wN /> , due to the bilinearity of the covariance operator, it holds
q q
PN
that ˇqp D nD1 wn ˇnp , where ˇnp WD Cov.Qrn ; rQwp /= 2 .Qrwp /.
q
We close this section by illustrating the concepts presented so far in the simple
case of two risky assets with random returns rQ1 and rQ2 . We denote by e1 , e2 and
12 , 22 the expected values and the variances of the random variables rQ1 and rQ2 ,
respectively. In this context, a portfolio is simply identified by the couple .w; 1 w/,
with w 2 R. The expected return and the variance of the return associated to a
portfolio .w; 1 w/, denoted by EŒQrw and 2 .Qrw /, respectively, are given by
EŒQrw D w e1 C .1 w/e2 ; (3.34)

2 .Qrw / D w2 12 C .1 w/2 22 C 2w.1 w/1 2 ;
where denotes the correlation coefficient between rQ1 and rQ2 .

In this simple situation, given a required expected return 2 R, the identification
of the portfolio with expected return achieving the minimum variance becomes
trivial, since there will be a unique portfolio .w; 1 w/ with expected return
EŒQrw D . Indeed, due to (3.34), it holds that w D . e2 /=.e1 e2 /, provided that
e1 ¤ e2 . In this case, the portfolio frontier can be easily described in the standard
deviation - expected return plane. The shape of the portfolio frontier crucially
depends on the correlation coefficient and we can distinguish three different cases
(see also Fig. 2.4):
a) If D 1 (perfect positive correlation), then we have, for all w 2 R,
EŒQrw D w e1 C .1 w/e2 ;
.Qrw / D w 1 C .1 w/2 :
The above equations imply that, in the standard deviation - expected return plane
.; /, the portfolio frontier is described by the linear equation
2
D e2 C .e1 e2 /:
1 2
This means that the portfolio frontier is represented by the straight line connect-
ing the point .1 ; e1 /, which represents the first risky asset, to the point .2 ; e2 /
representing the second risky asset.
b) If D 1 (perfect negative correlation), then we have, for all w 2 R,
EŒQrw D w e1 C .1 w/e2 ;
2
.Qrw / D w 1 .1 w/2 ; if w ;
1 C 2
2
.Qrw / D w 1 C .1 w/2 ; if w < :
1 C 2
In this case, the portfolio frontier is identified on the standard deviation

- expected return plane by the two half-lines originating from the point
.0; e2 C 1C
2
2
.e1 e2 // with slopes ˙ .e1 e2 /=.1 C 2 /.
c) If 1 < < 1, then the portfolio frontier is represented on the standard

deviation - expected return plane as a hyperbola in an intermediate position with
respect to the two cases considered above. For 0 w 1, the portfolio frontier
belongs to the interior of the region identified by the straight line connecting the
points .1 ; e1 / and .2 ; e2 / and the half-lines described in case b).
The Case of N Risky Assets and a Risk Free Asset
So far we have considered an economy with N risky assets with random returns
.Qr1 ; : : : ; rQN /. We now extend the analysis to the case of N C 1 assets, adding to the
original economy with N risky assets a risk free asset with deterministic return rf .
As before, we are interested in determining the portfolio composed of the N C 1
assets achieving the minimum variance for a given expected return. In the present
context, a portfolio is simply identified by a vector w D .w1 ; : : : ; wN /> 2 RN ,
where wn determines the proportion of Pwealth invested in the n-th risky asset, for
each n D 1; : : : ; N. The quantity 1 NnD1 wn represents the proportion of wealth
P
invested in the risk free asset. According to whether 1 NnD1 wn is greater or less
than zero, this represents the possibility of investing or borrowing, respectively, in
the bank account. We denote by PF the portfolio frontier obtained with N risky
assets plus a risk free asset. To exclude trivial cases, we assume that EŒQrn ¤ rf > 0
for at least one n 2 f1; : : : ; Ng.
To identify the portfolio frontier in the presence of a risk free asset, we need to
solve the following problem:
min w> Vw; (3.35)

w2RN
under the constraint
w> e C .1 w> 1/rf D ;
where 2 R represents the required expected return of the portfolio. Since we

now have an additional asset, namely the risk free asset with return rf , the set of
investment possibilities is larger than the one considered in the case of N risky
assets. Therefore, the minimum variance that can be achieved in Problem (3.35)
for a given expected return 2 R will always be less or equal than the minimum
variance that can be achieved in Problem (3.14) for the same expected return .
Actually, in the present context, it is always possible to reduce to zero the variance
of a portfolio by simply investing the total amount of wealth in the risk free asset.
Similarly as in the case of N risky assets, we can establish the following proposition.
Proposition 3.14 Let .Qr1 ; : : : ; rQN / be N random returns with expected value e 2 RN
and positive definite covariance matrix V 2 RNN and let rf > 0 be the risk free
return. Then, for any 2 R, the solution to the constrained variance minimization
problem (3.35) is given by
rf 1
w D V .e rf 1/; (3.36)
K
where K D .e 1rf /> V 1 .e 1rf / D B 2Arf C Crf2 > 0.

Proof As in the case of Theorem 3.12, the problem can be solved through the
associated Lagrangian
1 >
L.w; / WD w Vw C w> e .1 w> 1/rf :
2
The first order conditions amount to
@L.w ; /
D Vw .e 1rf / D 0;
@w
@L.w ; /
D w> e .1 w> 1/rf D 0:
@
Solving for w we obtain w D V 1 .e 1rf /, so that, solving for , we get

D . rf /=K, with K > 0 defined as in the statement of the proposition. Since
the matrix V is assumed to be positive definite, the first order conditions are both
necessary and sufficient for the optimality of w . t
u
In view of (3.36), the variance and the standard deviation of a portfolio w
belonging to the portfolio frontier composed by the N risky assets together with
the risk free asset can be explicitly computed as
2
2 > EŒQrw rf
.Qrw / D w Vw D (3.37)
K
and
EŒQrw rf
.Qrw / D p ; if EŒQrw rf I
K
(3.38)
EŒQrw rf
.Qrw / D p ; if EŒQrw rf :
K
The portfolio frontier PF can be represented

p in the p standard deviation - expected
return plane by two half-lines with slope K and K, respectively, originating
from the point .0; rf /. Analogously to the discussion in Property 4 above, the
half-line with positive slope represents the efficient part of the portfolio frontier
(EPF ). Moreover, as in the case of Property 1, it can be easily shown that two
arbitrary portfolios belonging to the portfolio frontier generate through linear affine
combinations all portfolios belonging to the portfolio frontier and linear affine
combinations of frontier portfolios belong to the portfolio frontier.
It is interesting to study the relation between the two portfolio frontiers PF
and PF obtained with and without, respectively, the risk free asset. Such a relation
crucially depends on rf and A=C, using the notation introduced in Theorem 3.12,
where we recall that A=C is the expected return associated to the minimum variance
portfolio constructed from the N risky assets.
Proposition 3.15 Let denote by PF and PF the portfolio frontiers obtained,
respectively, with and without the risk free asset with return rf and let A and C
be defined as in Theorem 3.12. Then, if rf ¤ A=C, there exists a unique portfolio we
which belongs to both PF and PF . Moreover, we 2 EPF if and only if rf < A=C.
Proof Let us suppose that rf ¤ A=C and consider the portfolio we 2 PF having
zero correlation with the portfolio belonging to PF with expected return equal to rf .
Then, in view of (3.30), it holds that
A BC A2 A rf B
EŒQrwe D D : (3.39)
C C.C rf A/ C rf A
Therefore, due to Theorem 3.12,
BV 1 1 AV 1 e CV 1 e AV 1 1 A rf B
we D g C h EŒQrwe D 2
C
BC A BC A2 C rf A
e rf 1
D V 1 :
1> V 1 .e
rf 1/
It can be easily checked that this expression coincides with (3.36) for an expected
return equal to D .A rf B/=.C rf A/. We have thus shown that the portfolio we
belongs to both PF and PF . We now show that the intersection of the two portfolio
frontiers PF and PF contains at most one element, thus yielding the uniqueness of
we . Indeed, arguing by contradiction, suppose that there exists another portfolio w0
belonging to both PF and PF . Note first that EŒQrwe ¤ EŒQrw0 , since each expected
return identifies a unique frontier portfolio. Let ˛ WD .rf EŒQrw0 /=.EŒQrwe rQw0 /
and define the portfolio wN WD ˛we C .1 ˛/w0 . Then, since PF and PF are both
closed under linear affine combinations, the portfolio wN belongs to both PF and PF .
Moreover, due to the definition of ˛, we have that EŒQrwN D rf and, by formula (3.37),
it also holds that 2 .QrwN / D 0 < 1=C, thus contradicting the minimality property
of the minimum variance portfolio wMVP . We have thus shown that we is unique.
Finally, recalling that BC A2 > 0 (see Theorem 3.12), it follows from (3.39) that
EŒQrwe > A=C if and only if rf < A=C. In view of Property 4, this implies that we
belongs to EPF if and only if rf < A=C. t
u
We can also explicitly compute the variance of the portfolio we as follows:
.e rf 1/> V 1 .e rf 1/ K
2 .Qrwe / D we> Vwe D 2 D : (3.40)
> 1
1 V .e rf 1/ .A Crf /2
In the context of Proposition 3.15, if rf ¤ A=C, then the portfolio we admits an

interesting geometric characterization: in the standard deviation-expected return
plane, the portfolio we is identified by the tangency condition between one of the
two half-lines (3.38) and the portfolio frontier PF composed by the risky assets
only. In view of expression (3.38), in order to confirm this claim, it suffices to show
we
p slope of the tangent line to PF in correspondence of ..Qr /; EŒQrwe / is equal
that the
to ˙ K (compare also with Exercise 3.14). By Property 3, the line tangent to PF at
e
the point ..Qrw /; EŒQrwe / has a slope equal to
d .Qrwe /D
D˙ :
d C EŒQrwe A
Substituting EŒQrwe and 2 .Qrwe / with the explicit expressions p given in (3.39)
and (3.40), respectively, we then obtain that d=d D ˙ K. In view of this
observation, we call the portfolio we (when it exists, i.e., when rf ¤ A=C) the
tangent portfolio.
Thanks to Property 1, which also holds for the portfolio frontier PF , any frontier
portfolio can be obtained as the linear affine combination of the tangent portfolio we
and of the portfolio w0 investing in the risk free asset alone. Note that the tangent
portfolio satisfies we> 1 D 1, while the portfolio investing only in the risk free asset
satisfies w0> 1 D 0. Depending on the relation between rf and A=C (where the latter
quantity is the expected return of the minimum variance portfolio wMVP ), we can
distinguish three cases (the first one is illustrated in Fig. 3.4):
a) The PF is composed of two half-lines, see (3.38). If rf < A=C, due to
Proposition 3.15, the tangent portfolio we is efficient and the frontier is divided
into three regions. The first region corresponds to the segment connecting the two
points that correspond to w0 and we . This region contains all portfolios that are
linear convex combinations of the two portfolios w0 and we , i.e., all portfolios
characterized by a positive investment in the risk free asset as well as in the
tangent portfolio we , so that 0 w> 1 1. The second region corresponds to the
part at the right of the point representing we : any portfolio w belonging to this
second region involves short selling the risk free asset and investing in the tangent
portfolio we more than the total amount of wealth, i.e., w> 1 > 1. Finally, the third
region is located below the risk free rate rf and contains all inefficient portfolios.
Such portfolios involve short selling the tangent portfolio we and investing more
than the total amount of wealth in the risk free asset w0 , i.e., w> 1 < 0.
b) If rf > A=C, due to Proposition 3.15, the tangent portfolio we is inefficient and, as
in case a), the frontier is divided into three regions. The first region corresponds to
the segment connecting the two points that correspond to w0 and we . This region
we
E [r̃ew ]
A/C
w0
rf
(r̃ew )
Fig. 3.4 Portfolio frontier with a risk free asset (A=C > rf )
contains all portfolios that are linear convex combinations of the two portfolios
w0 and we , i.e., all portfolios characterized by a positive investment in the risk
free asset as well as in the tangent portfolio we , so that 0 w> 1 1. Any
such portfolio is inefficient. The second region corresponds to the part at the
right of the point that represents we . Any portfolio w belonging to this second
region is inefficient and involves short selling the risk free asset and investing
in the tangent portfolio we more than the total amount of wealth, i.e., w> 1 > 1.
Finally, the third region contains all efficient portfolios and is located above rf .
All portfolios belonging to this region involve short selling the tangent portfolio
we and investing an amount of money larger than the total amount of wealth in
the risk free asset w0 , i.e., w> 1 < 0.
c) Finally, it may happen that rf D A=C. In this case K D D=C > 0 and the
portfolio frontier PF is represented in the standard deviation-expected return
plane by the asymptotes of the portfolio frontier PF. In this case, there does
not exist a portfolio belonging to both frontiers PF and PF and every portfolio
belonging to PF involves a net investment in the risk free asset equal to the total
amount of wealth and, consequently, the overall net amount of wealth invested in
the risky assets is equal to zero. Indeed, let w be an arbitrary portfolio belonging
to PF , thus admitting the representation (3.36). Then, setting rf D A=C, we get

A EŒQrw A=C
1> w D 1> V 1 e 1 D 0:
C K
The previous discussion on the efficiency of portfolios belonging to the portfolio

frontier PF leads us to introduce the Sharpe ratio, an index frequently used to
evaluate the performance of an asset or portfolio. Given a portfolio w 2 RN , the
corresponding Sharpe ratio SR.w/ is defined as follows:
EŒQrw rf
SR.w/ WD ;
.Qrw /
i.e., the ratio between the return risk premium of a portfolio and the standard
deviation of the associated random return. In the standard deviation - expected
return plane, the Sharpe ratio represents the slope of the straight line which connects
the point .0; rf / to the point ..Qrw /; EŒQrw / representing the portfolio w. Observe

that, due to expression (3.38),
p the Sharpe ratio of all portfolios belonging to EPF
is constant and equal to K. Moreover, since efficient frontier portfolios are not
dominated according to the mean-variance
p criterion, there does not exist a portfolio
with a Sharpe ratio greater than K. On the Sharpe ratio, see also Exercise 3.19.
We can obtain an explicit expression for the covariance of an arbitrary portfolio
w 2 RN (investing in the N C 1 assets and not necessarily belonging to the portfolio
frontier) and a portfolio w belonging to the portfolio frontier PF . Indeed, by
relying on representation (3.36) and on the fact that EŒQrw D w> e C .1 w> 1/rf ,
we get

> EŒQrw rf > EŒQrw rf EŒQrw rf
Cov.Qrw ; rQw / D w Vw D w .e 1rf / D :
K K
In the case of an economy with N risky assets and a risk free asset with return rf ,
we can establish the following proposition (we omit the proof, which is analogous
to that of Proposition 3.13, noting that the riskless portfolio w0 has zero correlation
with any other portfolio).
Proposition 3.16 Let wq 2 RN be an arbitrary portfolio and wp 2 RN a portfolio
belonging to the portfolio frontier PF such that EŒQrwp ¤ rf . Then the following
hold:
(i) if wq 2 PF then
rQwq D .1 ˇqp /rf C ˇqp rQwp I
(ii) if wq … PF then
rQwq D .1 ˇqp /rf C ˇqp rQwp C Qqp ;


where Cov rQwp ; Qqp D EŒQ qp D 0 and ˇqp D Cov.Qrwq ; rQwp /= 2 .Qrwp /.
Moreover, for any portfolio wq 2 RN , it holds that

EŒQrwq rf D ˇqp EŒQrwp rf : (3.41)
Relation (3.41) expresses the risk premium of any portfolio wq (not necessarily
belonging to the portfolio frontier PF ) in terms of the risk premium of an arbitrary
portfolio wp 2 PF . The two risk premia have the same sign if and only if the
coefficient ˇqp is positive, where ˇqp is the regression coefficient of the random
return rQwq on rQwp .
Extensions: Constraints on Borrowing
The portfolio frontier can also be constructed in the presence of additional con-
straints on the financial market. In what follows, we limit our attention to two
possible situations: the case where agents are not allowed to borrow at the risk free
rate rf and the case where there are transaction costs on the risk free asset (i.e., there
are different interest rates for lending and for borrowing).
Let us first consider the case where agents cannot borrow at the risk free rate
rf , i.e., all portfolios must satisfy the additional requirement w> 1 1. As we
have discussed above, if rf < A=C, then the efficient portfolio frontier EPF
(obtained by including the risk free asset) consists of the tangent line to the portfolio
frontier PF (with no risk free asset) in correspondence of an efficient portfolio we .
Moreover, any portfolio belonging to the portfolio frontier can be represented as a
linear combination of we and of the risk free asset (i.e., of the portfolio w0 D 0).
According to the analysis developed above (see Fig. 3.4), on the efficient part of
the portfolio frontier, i.e., on the straight line with positive slope originating from
the point .0; rf /, borrowing occurs only at the right of the point ..Qrwe /; EŒQrwe /,
since along the segment that connects the point .0; rf / to the point ..Qrwe /; EŒQrwe /
an agent will invest a positive amount of wealth both in the risk free asset and in the
tangent portfolio (i.e., 0 w> 1 1), while along the half-line with negative slope
he will sell short the tangent portfolio (i.e., w> 1 0) and will invest in the risk free
asset. As a consequence, the portfolio frontier in the presence of the additional no-
borrowing constraint coincides with the unconstrained portfolio frontier PF up to
the point ..Qrwe /; EŒQrwe /, which corresponds to the tangent portfolio we . However,
at the right of that point the constrained portfolio frontier will coincide with the
portfolio frontier PF obtained without the risk free asset. In the latter part of the
portfolio frontier, due to the no-borrowing constraint, the agent will only invest in
the N risky assets. Of course, a similar analysis can be developed in the symmetric
case where rf > A=C.
we, b
C
e,
w
A/C
rbf
rf
Fig. 3.5 Portfolio frontier with transaction costs
Let us now suppose that agents face two different risk free rates depending on
whether they borrow or invest in the risk free asset. More specifically, we assume
the existence of a risk free rate rfb (rf` , resp.) which represents the rate at which it
is possible to borrow from (to lend to, resp.) the bank account. We suppose that
rfb > rf` and that A=C > rfb . Considering two risk free assets with returns rfb and rf` ,
respectively, we can build two portfolio frontiers as in Proposition 3.14. We denote
by we;b the tangent portfolio of the portfolio frontier obtained with a risk free asset
with return rfb and, analogously, by we;` the tangent portfolio of the frontier obtained
with risk free rate rf` . Note that, as can be seen from (3.38), the absolute value of
the slope of the two half-lines defining the portfolio frontier with a risk free asset
is decreasing in the risk free rate. In the present context, the portfolio frontier is
illustrated in Fig. 3.5. In particular, the portfolio frontier with different borrowing
and lending rates is composed of four regions. In region A, agents sell short the
portfolio of the risky assets we;` (i.e., w> 1 0) and invest in the risk free asset
(with a corresponding rate rf` ). In region B agents invest a positive amount of wealth
both in the risk free asset (with corresponding rate rf` ) and in the risky portfolio we;` .
In region C agents neither invest in the risk free asset nor borrow money, holding a
portfolio composed by the risky assets only. In particular, such a portfolio is a linear
convex combination of we;b and we;` . Finally, in region D agents invest more than
the total wealth in the risky portfolio we;b and borrow money at the risk free rate rfb .
Mutual Fund Separation and Mean-Variance Portfolio Selection
As we have discussed in section “The Case of Multiple Risky Assets”, in the case
of multiple risky assets, it is generally difficult to establish explicit and informative
results on optimal portfolios. An exception is represented by the case when investors
choose their optimal portfolios by constructing portfolios of a family of mutual
funds. In that case, we can reduce the dimension of the original portfolio choice
problem by only considering portfolios composed of the mutual funds.
At the end of Sect. 3.1, we have provided conditions ensuring that a portfolio
choice problem in the presence of N risky assets and a risk free asset can be reduced
to the analysis of an optimal choice problem with two mutual funds, represented
by a fixed portfolio of the risky assets and the risk free asset. In that context, we
have provided an answer by formulating sufficient conditions on the utility function
in order to ensure the two mutual funds separation property. In particular (see
Proposition 3.11), we have shown that two mutual funds (monetary) separation
always holds for utility functions belonging to the HARA class, regardless of the
distribution of the returns.
In the present section, we consider the mutual fund separation problem from
a different perspective. Instead of restricting the class of utility functions, we
shall consider restrictions on the probability distribution of the returns, aiming at
establishing necessary and sufficient conditions for the validity of the mutual fund
separation property for every risk averse agent, following Ingersoll [1000] and Ross
[1467]. We aim at answering the following question: given a set of N assets, is it
possible to identify K < N portfolios (mutual funds) of the N assets such that,
for any portfolio composed of the N original assets and for any utility function
satisfying some basic properties, there exists a portfolio composed of the K mutual
funds that is preferred to the first portfolio? More precisely, let us formulate the
following definition.
Definition 3.17 We say that a set of N assets with returns rQ D .Qr1 ; : : : ; rQN / satisfies
the separation property through K mutual funds if there exist K portfolios (mutual
funds) .w1 ; : : : ; wK /, with wk 2 N and return rQ k , for all k D 1; : : : ; K, such that, for
any portfolio w 2 P N of the N assets with return r Qw , there exists a vector of weights
.1 ; : : : ; K /, with KkD1 k D 1, such that, for every concave utility function u,
" !#
X
K

E u k rQ k
E u.Qrw / : (3.42)
kD1
According to the above definition, a set of N assets satisfies the separation

property through K mutual funds if, for any portfolio w 2 N of the original N
assets, there exists a linear combination 1 w1 C : : : C K wK , with weights adding
up to one, of the K mutual funds which dominates the portfolio w in the sense of
second order stochastic dominance (see Sect. 2.3). Definition 3.17 is in a strong
sense because the weights 1 ; : : : ; K are assumed not to depend on the utility
function. If the weights are allowed to depend on the utility function u, then one
can define the separation property in a weak sense (however, the two definitions
coincide for K 2, see Ross [1467]). We also mention that an alternative definition
of separation property can be obtained by formulating Definition 3.17 with respect
to all concave non-decreasing functions (see Ross [1467]).
In what follows, we shall focus our attention on the case K D 2 (see Exercise 3.24
for the simple case K D 1). As in the case of portfolio choice problems, the
conditions which ensure the validity of the two mutual funds separation property
depend on whether a risk free asset is present in the economy or not.
Let us first consider an economy with N risky assets with returns rQ D .Qr1 ; : : : ; rQN /
together with a risk free asset with return rf . In order to establish the two mutual
funds separation property, we can assume that the risk free asset acts as one
of the two mutual funds (monetary separation). Indeed, since Definition 3.17 is
formulated with respect to any concave utility function, with possibly unbounded
risk aversion, we can consider risk averse agents for whom investing only in
the risk free asset is the optimal choice. A necessary and sufficient condition on
the probability distribution of the random vector rQ in order to satisfy the two
mutual funds (monetary) separation property is given in the following proposition
(see Ingersoll [1000, Chapter 6] for a proof based on the second order stochastic
dominance criterion).
Proposition 3.18 The two mutual funds (monetary) separation property holds if
and only if the random vector rQ admits the representation
rQ D rf 1 C bQr C Q ;
for some random variable rQ , where b 2 RN and Q is an N-dimensional random

PN such that EŒQ
vector jQr D 0, and there exist weights .a1 ; : : : ; aN / such that
P N
iD1 ai Qi D 0 and iD1 ai D 1.
In the setting of the above proposition, the two separating mutual funds are
represented by the risk free asset and the portfolio .a1 ; : : : ; aN /> 2 N composed
by the N risky assets.
If we instead consider an economy with N risky assets and without a risk
free asset, then a necessary and sufficient condition in order to satisfy the two
mutual funds separation property can be formulated as follows (see Ross [1467,
Corollary 1]).
Proposition 3.19 Suppose that EŒQri ¤ EŒQr1 , for at least one i 2 f2; : : : ; Ng. Then
the two mutual funds separation property holds if and only if the random vector rQ
admits the representation
rQ D rQ 1 C bQr C Q ;
for some random variables rQ and rQ , where b 2 RN and Q is an N-dimensional

random vector such that EŒQ jQr C rQ D 0, for every 2 R, and there exist two
vectors of weights a; c 2 RN such that a> 1 D c> 1 D 1 and a> Q D c> Q D 0, with
a> b ¤ c> b.
If the asset returns are distributed according to a distribution belonging to the
elliptical class, as well as any distribution implying a mean-variance utility function
(see Sect. 2.4), then the two mutual funds separation property holds, see Owen &
Rabinovitch [1389] and Chamberlain [395]. In particular, the normal multivariate
distribution satisfies the conditions for the two mutual funds separation property,
see Ross [1467].
Let us now explore the relations between the mutual funds separation property
and the mean-variance portfolio frontier. Let us suppose that the economy consists
of N risky assets, without a risk free asset. It is easy to see that, if the two
funds separation property holds, then the two separating portfolios w1 and w2 must
belong to the portfolio frontier. Indeed, suppose on the contrary that one of the
two portfolios does not belong to the portfolio frontier. Then, for any linear affine
combination w1 C .1 /w2 of the two portfolios w1 ; w2 , there exists a portfolio
w0 such that

EŒQrw0 D EŒQrw1 C .1 /Qrw2 and 2 .Qrw0 / < 2 Qrw1 C .1 /Qrw2 :
This implies that the portfolio w0 is not dominated according to the second order
stochastic dominance criterion by the linear combination of the two portfolios
w1 ; w2 . Clearly, this contradicts the assumption that the two funds separation
property holds with respect to the couple w1 ; w2 .
Due to the properties of the portfolio frontier, if the N assets satisfy the mutual
funds separation property, then any pair of two portfolios belonging to the portfolio
frontier can be chosen as a pair of mutual funds. In particular, as mutual funds we
may take an arbitrary frontier portfolio wp (different from the minimum variance
portfolio wMVP ) and the associated zero-covariance portfolio wzc.p/ . Recall that, in
view of Proposition 3.13, the return of any portfolio wq 2 N can be written as
rQwq D ˇqp rQwp C .1 ˇqp /Qrwzc.p/ C Qqp ; (3.43)
where Cov.Qrwp ; Qqp / D Cov.Qrwzc.p/ ; Qqp / D EŒQ qp D 0. Exploiting this represen-

tation, we can establish the following result (see Huang & Litzenberger [971] and
Litzenberger & Ramaswamy [1222]).
Proposition 3.20 Consider N risky assets with returns rQ D .Qr1 ; : : : ; rQN / and let wp
be a frontier portfolio, with corresponding zero-correlation portfolio wzc.p/ . The two
funds separation property holds with respect to the pair of mutual funds .wp ; wzc.p/ /
if and only if
EŒQ qp j ˇqp rQwp C .1 ˇqp /Qrwzc.p/ D 0;
for any portfolio wq 2 N .

Proof We show the sufficiency of the above condition (we refer to Huang &
Litzenberger [971, Chapter 4] for the proof of the necessity part). Due to the tower
property of the conditional expectation together with Jensen’s inequality, it holds
that, for any portfolio wq 2 N ,

EŒu.Qrwq / D E EŒu.Qrwq /jˇqp rQwp C .1 ˇqp /Qrwzc.p/

E u EŒQrwq jˇqp rQwp C .1 ˇqp /Qrwzc.p/

D E u ˇqp rQwp C w0 .1 ˇqp /Qrwzc.p/ ;
for any concave utility function u. Hence, by Definition 3.17, the two portfolios wp
and wwzc.p/ act as separating mutual funds. t
u
By relying on the above proposition, we can also establish the following result.
Corollary 3.21 Consider N assets with normally distributed returns rQ D
.Qr1 ; : : : ; rQN /, with EŒQri ¤ EŒQrj for some i; j 2 f1; : : : ; Ng, and let wp be a frontier
portfolio, with wzc.p/ being the corresponding zero-correlation portfolio. Then
the two funds separation property holds with respect to the pair of mutual funds
.wp ; wzc.p/ /.
Proof Since the normal distribution is stable with respect to linear transformations,
.Qrwp ; rQwzc.p/ / is jointly normally distributed. Moreover, due to the properties of the
normal distribution (and noting that Qqp is also normally distributed), the fact that
Cov.Qrwp ; Qqp / D Cov.Qrwzc.p/ ; Qqp / D 0 implies that rQwp , rQwzc.p/ and Qqp are mutually
independent. In turn, this implies that EŒQ qp j ˇqp rQwp C.1ˇqp /Qrwzc.p/ D EŒQ qp D 0.
Proposition 3.20 then implies that the two funds separation property holds. t
u
The above results are in line with Proposition 2.13, where we have shown that, if
asset returns are normally distributed, then the expected utility of a risk averse agent
with a strictly increasing utility function can be written as a function of the mean
and of the variance, increasing in the first argument and decreasing in the second
argument. These features imply that the agent’s optimal portfolio belongs to the
efficient part of the portfolio frontier. A result analogous to Corollary 3.21 can be
established in the case where the risky assets have normally distributed returns with
equal expectations. In this case, as shown in Exercise 3.26, the one fund separation
property holds with respect to the minimum variance portfolio wMVP .
Let us now consider an economy with a risk free asset with return rf > 0 together
with N risky assets with returns rQ D .Qr1 ; : : : ; rQN /. In this setting, by Proposition 3.16,
the random return rQw associated to a portfolio w composed of the N risky assets plus
the risk free asset can be represented as
rQw D ˇqe rQwe C .1 ˇqe /rf C Qqe ;
where rQwe 2 PF and Cov.Qrwe ; Qqe / D EŒQ qe D 0. We have the following version of
Proposition 3.20 (see also Huang & Litzenberger [971, Chapter 4]).
Proposition 3.22 Consider N risky assets with returns rQ D .Qr1 ; : : : ; rQN / and a
risk free asset with return rf ¤ A=C and let we be the tangent portfolio (see
Proposition 3.15). The two funds (monetary) separation property holds with respect
to the risk free asset and the tangent portfolio we if and only if
EŒQ qe j rQwe D 0;
for any portfolio wq 2 N , using the notation of (3.43).

Similarly to the analysis performed in the case of N risky assets, if asset returns
are normally distributed and have different expected returns then the two funds
separation property holds true.
We close this section with a simple example, which shows that, in the case of
N risky assets with normally distributed returns and a risk free asset with return
rf < A=C, the optimal portfolio of an agent characterized by an exponential utility
function is composed of the risk free asset and of the tangent portfolio we . Of course,
such a result is not surprising, in view of the above discussion and of the link
between exponential utility maximization and mean-variance portfolio selection in
the case of normally distributed returns (see section “Closed Form Solutions and
Mutual Fund Separation”). Indeed, for aP given initial wealth w0 and a risk aversion
parameter a > 0, letting We D w0 rf C NnD1 wn .Qrn rf /, we consider the optimal
portfolio choice problem
1 e :

max E exp.aW/
w2R N a
Due to expression (3.11), the optimal portfolio w 2 RN is explicitly given in terms

of wealth invested in the N risky assets by
1 1
w D V .e rf 1/:
a
On the other hand, due to Proposition 3.14, the frontier portfolio w associated to
the expected return can be written in terms of proportions of wealth invested in
the N risky assets as
rf
w D V 1 .e rf 1/ ;
K
where K is defined as in Proposition 3.14. By comparing the two above portfolios,

we see that the optimal portfolio w for the exponential utility corresponds to the
frontier portfolio w for D rf C K=.aw0 /. Recall also that, by Proposition 3.15,
the tangent portfolio we is given by
.e rf 1/
we D V 1 ;
1> V 1 .e rf 1/
so that
1> V 1 .e rf 1/ e
w D w:
a
In particular, note that the proportion of wealth invested in the tangent portfolio we is
decreasing in the coefficient of absolute risk aversion a. In Fig. 3.6 the indifference
curves of an exponential utility function are depicted together with the optimal
portfolios, which are shown to belong to the efficient part of the portfolio frontier,
given by a linear combination of the risk free asset and the tangent portfolio we .
we
rf
Fig. 3.6 Mean-variance utility and portfolio frontier

3.3 Insurance Demand
In this section, we shall consider the optimal choice problem of a risk averse agent
facing a risk (i.e., a potential loss) and having the possibility of buying an insurance
contract in order to mitigate the risk exposure. Insurance problems can be addressed
with tools analogous to those used in the context of portfolio selection problems,
since the insurance contract can be thought of as a risky asset whose payoff depends
on the realization of the loss event. Indeed, while in portfolio choice problems agents
need to decide on the optimal allocation of wealth, in insurance problems agents
need to determine the optimal demand of insurance contracts in order to hedge the
risk.
Let us first study an insurance problem in a very simple setting, where the
randomness is reduced to two elementary states of the world. Consider a risk averse
agent endowed with initial wealth w0 > 0 at date t D 0, facing the possibility of
a loss event at the future date t D 1. The loss event is represented by a random
variable taking value D > 0 with probability 2 .0; 1/ and 0 with probability
1 (i.e., at t D 1 there is the risk of losing an amount equal to D). If there does
not exist any insurance market, then the agent’s expected utility is simply given by
u.w0 D/ C .1 /u.w0 /;
for some given utility function u W R ! R.

We now introduce an insurance market, in a very simple form. At time t D 0,
the agent has the possibility of buying a particular type of insurance contract, which
delivers one unit of wealth at time t D 1 only if the loss occurs. Such an insurance
contract can be viewed as an Arrow security, i.e., a security paying 1 if an elementary
event is realized at t D 1 and 0 otherwise. In other words, the agent can trade in
the insurance market and buy one unit of wealth at t D 1 contingent on the event
“the loss happens” (for simplicity, we assume that the interest rate between date
t D 0 and date t D 1 is equal to zero). We denote by p > 0 the price at t D 0
of this insurance contract. Observe that the latter can be interpreted as the gamble
Œ1 p; pI ; 1 . This gamble is fair (i.e., it has zero expectation) if and only
if p D , meaning that the price of one unit of wealth at t D 1 contingent on the
occurrence of the loss is equal to the probability of the loss occurring.
Let us denote by w 2 RC the optimal quantity of insurance bought by a risk
averse agent at time t D 0, assuming perfect liquidity in the insurance market. In
this simple setting, the following proposition characterizes the optimal insurance
demand (we tacitly assume that utility functions are continuously differentiable),
see Mossin [1358].
3.3 Insurance Demand 103
Proposition 3.23 Let u be strictly increasing and strictly concave utility function
and let w0 > 0. Then the following hold:
(i) if p D then w D D;
(ii) if p > then w < D;
(iii) if p < then w > D.
Proof The agent’s optimization problem amounts to

max u.w0 D wp C w/ C .1 /u.w0 wp/ :
w2RC
By the Kuhn-Tucker conditions, w is a solution to the above problem if and only if
.1 p/u0 .w0 D w p C w / p.1 /u0 .w0 w p/ 0; (3.44)
with equality holding in the case of an interior solution (i.e., w > 0). If p D ,
then condition (3.44) becomes

u0 w0 D C w .1 p/ u0 .w0 w p/ 0:
Since the function u is assumed to be strictly concave (so that u0 is strictly

decreasing), this condition can only be verified for w > 0, meaning that the optimal
solution w must satisfy

u0 w0 D C w .1 p/ D u0 .w0 w p/:
Again, the strict concavity of u implies that the unique solution is given by w D D,
thus proving part (i) of the proposition. If p > , then the optimality condition (3.44)
gives, for an optimal choice w > 0,
u0 .w0 D w p C w / p.1 /
D > 1:
u0 .w0 w p/ .1 p/
By the strict concavity of u, this implies that w < D, thus proving part (ii). The
proof of part (iii) is analogous. t
u
In view of the above proposition, if the insurance contract is fairly priced
(i.e., p D ), then a risk averse agent will insure himself completely, thus reducing
to zero the variability of his wealth at the future time t D 1. On the contrary, if
the price p of the insurance contract is higher than the probability of the loss
occurring, then the agent will only partially insure himself (w < D). Finally, in the
case p < , the agent will over-insure himself, buying an insurance coverage w
greater than the potential loss D.
Let us now extend the above analysis to the case of several possible events at
time t D 1. For simplicity, we shall restrict our attention to the case of two possible
events, removing the risk free status where the initial wealth remains unaltered. This
means that at the initial time t D 0 an agent faces a gamble xQ D Œx1 ; x2 I ; 1 .
The expected utility associated to such a gamble is
U.Qx/ D u.x1 / C .1 /u.x2 /:
For instance, we can think of a farmer endowed with a field of wheat, which provides
a crop x1 if the weather conditions are favorable and a crop x2 if the weather
conditions are unfavorable.
We first characterize the indifference curves associated to the expected utility
U in the space of wealth in correspondence of the first event and wealth in
correspondence of the second event, keeping the probability fixed. In this context,
we denote a generic gamble xQ through the pair .x1 ; x2 / and the space R2C is called the
state space. The marginal rate of substitution SMS.x1 ; x2 / of an agent with utility
function u in correspondence of the gamble xQ D .x1 ; x2 / (given by the ratio between
the marginal utilities in the two states of the world weighted by the corresponding
probabilities of occurrence) is equal to minus the slope of the tangent line to the
indifference curve at the point .x1 ; x2 /, i.e.,
dx2 u0 .x1 /

SMS.x1 ; x2 / D D : (3.45)
dx1 .1 /u0 .x2 /
The bisectrix of the state space R2C admits an interesting interpretation, see
Fig. 3.7. Indeed, it represents the certainty line, i.e., the locus of all the gambles
characterized by the same amount of wealth in correspondence of both elementary
events. In other words, along the certainty line the agent faces no risk and the
following equality holds:

SMS.x1 ; x2 / D :
1
This means that, along the certainty line, the marginal rate of substitution is equal
to the ratio of the probabilities of the two elementary events. This allows us to show
an important property of utility functions of risk averse agents: if u is concave, then
the indifference curves in the state space R2C are convex. To show this property, it
suffices to prove that
dSMS.x1 ; x2 /
0; for all .x1 ; x2 / 2 R2C :
dx1
x2
45◦ − /(1 − )◦
x1
Fig. 3.7 Expected utility in the state space
Since SMS.x1 ; x2 / > 0, due to the strict increasingness of u, the above inequality
holds if and only if d log SMS.x1 ; x2 /=dx1 0. Differentiating the logarithm of the
marginal rate of substitution and using (3.45), we obtain
d log.SMS.x1 ; x2 // d u00 .x1 / u00 .x2 / dx2

D log u0 .x1 / log u0 .x2 / D 0 0
dx1 dx1 u .x1 / u .x2 / dx1
u00 .x1 / u00 .x2 / u0 .x1 /
D 0
C 0 0;
u .x1 / u .x2 / .1 / u0 .x2 /
where the last inequality comes from the risk aversion hypothesis (i.e., the concavity
of u). Analogously, one can show that if an agent is risk lover then his indifference
curves in the state space R2C are concave.
As illustrated by Fig. 3.7, risk aversion implies that agents aim at diversifying
risks. Indeed, since the indifference curves are convex in the state space R2C , the
convex linear combination of two gambles will always be preferred (or, at most,
indifferent) to the worse of the two gambles, while if two gambles are indifferent
among themselves then their convex linear combination will be preferred (or, at
most, indifferent) to both of them. Of course, the opposite happens in the case of
risk loving agents: they will concentrate their wealth in only one of the two gambles.
Let us now consider an agent who can trade in a market with two securities:
a security that represents wealth contingent upon the occurrence of the first
elementary event and a security representing wealth contingent upon the occurrence
of the second elementary event (Arrow securities). We denote by p1 and p2 the price
of the first and of the second security, respectively, and assume that p1 ; p2 > 0. Let
also .e1 ; e2 / denote an agent’s endowment in terms of wealth in correspondence of
the two possible states of the world. In the present context, the optimal consumption
problem for an agent characterized by an utility function u becomes

max u.x1 / C .1 /u.x2 / ;
.x1 ;x2 /2R2C
subject to the budget constraint
p 1 x1 C p 2 x2 p 1 e1 C p 2 e2 :
As can be readily verified, necessary and sufficient conditions for an interior solution
.x1 ; x2 / 2 R2CC amount to
u0 .x1 / p1
D : (3.46)
.1 /u0 .x2 / p2
Observe the analogy between condition (3.46) and the classical condition (1.1)
obtained for the optimal consumption problem under certainty. The main difference
between these two conditions is that, under uncertainty, the marginal utilities of
optimal consumption in correspondence of the two states of the world are weighted
by the respective probabilities of occurrence.
If the two Arrow securities are priced fairly in relative terms, meaning that
p1 =p2 D =.1 /, then the optimal consumption will be the same in both states
of the world (i.e., x1 D x2 ), analogously to the case of a fair insurance contract
described at the beginning of this section. Hence, if the two Arrow securities are
priced fairly, then the agent will choose an optimal consumption vector belonging
to the certainty line, thus reducing to zero his overall risk exposure. Note also that, if
the Arrow securities are priced fairly, then the budget constraint (when satisfied as an
equality) identifies consumption vectors .x1 ; x2 / with an expected value equal to that
of the initial endowment .e1 ; e2 /. Hence, choosing among lotteries with the same
expected value, a risk averse agent will optimally achieve a risk free consumption
plan by suitably investing in the two Arrow securities.
It can be easily shown that if p2 =p2 ¤ =.1 / (i.e., the two securities are not
fairly priced in relative terms), then an agent’s optimal consumption plan .x1 ; x2 /
does not belong to the certainty line. Indeed, as can be deduced from (3.46), if wealth
contingent on the first elementary event has a relative price with respect to wealth
contingent on the second elementary event higher than the fair price =.1 /, then
the optimal consumption x1 in correspondence of the first event will be smaller than
the optimal consumption x2 in correspondence of the second event:
p1
> ” x1 < x2 :
p2 1
So far, we have considered losses represented by random variables taking a finite

number of possible values. However, the above results can be extended to general
random variables. We shall think of a random variable xQ with EŒQx > 0 as the
potential loss incurred by an agent and assume that the agent can buy an insurance
contract to offset the loss xQ . The price of one unit of insurance is assumed to be
given by .1 C /EŒQx, where 0 can be interpreted as an insurance premium, and
the payoff of one unit of the insurance contract is exactly xQ . The agent can choose
the optimal coverage w (i.e., the number of units of the insurance contract bought
at t D 0): he pays w .1 C /EŒQx at time t D 0 and he will receive w xQ at time
t D 1. In the current setting, the optimal insurance problem of an agent endowed
with initial wealth w0 > 0 can be formulated as follows:

max E u.w0 .1 w/Qx w.1 C /EŒQx/ : (3.47)
w2RC
The solution to problem (3.47) is given in the following proposition, where we

implicitly assume enough regularity in order to ensure the existence of an interior
solution (compare with Proposition 3.23).
Proposition 3.24 Let u be a strictly increasing and strictly concave utility function
and let w0 > 0. Then the following hold:
(i) if D 0 then w D 1;
(ii) if > 0 then w < 1;
(iii) if < 0 then w > 1.
Proof The first order condition for an interior solution w > 0 to problem (3.47) is
given by

E u0 w0 .1 w /Qx w .1 C /EŒQx xQ .1 C /EŒQx D 0:
Since the function u is assumed to be strictly concave, this condition is necessary

and sufficient for a solution w > 0 to problem (3.47). If D 0, then the above
condition becomes

0 D E u0 w0 .1 w /Qx w EŒQx xQ EŒQx

D EŒu0 w0 .1 w /Qx w EŒQx E xQ EŒQx

C Cov u0 w0 .1 w /Qx w EŒQx ; xQ EŒQx

D Cov u0 w0 .1 w /Qx w EŒQx ; xQ EŒQx :
Clearly, w D 1 satisfies the latter condition, since the covariance between a random
variable and a deterministic quantity is always null. The strict concavity of u ensures
that w D 1 is the unique solution. Let us now consider the case > 0. In this case,
the derivative of the expected utility with respect to w evaluated at w D 1 is negative:

E u0 w0 .1 C /EŒQx xQ .1 C /EŒQx D u0 w0 .1 C /EŒQx EŒQx < 0;
since EŒQx > 0. As a consequence, it holds that w < 1, thus proving part (ii) of the
proposition. The proof of part (iii) is analogous. t
u
Note that, similarly to Proposition 3.23, it the insurance contract is priced fairly
(i.e., D 0), then a risk averse agent will choose a full insurance coverage (w D 1),
thus eliminating the exposure to the risk generated by the random loss xQ . On the
contrary, it can be shown that, for a sufficiently large insurance premium , the
agent will choose not to buy insurance at all (see Exercise 3.29).
As in the case of portfolio choice problems with a single risky asset (see
Sect. 3.1), several comparative statics results can be established on the optimal
insurance demand with respect to changes in the initial wealth w0 , in the insurance
premium and in the degree of risk aversion, as shown in the following proposition
(where we always assume that the utility function u is sufficiently differentiable).
Then the following hold:
(i) given the same initial wealth w0 , if agent a is more risk averse than agent b,
then the optimal insurance demand of agent a is higher than that of agent b,
i.e., wa wb ;
(ii) if the utility function u is DARA (IARA, resp.), then the optimal insurance
demand w is decreasing (increasing, resp.) with respect to the initial wealth
w0 ;
(iii) if the utility function u is CARA or IARA, then the optimal insurance demand
w is decreasing with respect to the insurance premium .
Proof The proposition can be proved by the same arguments used to establish
Propositions 3.2, 3.3 and 3.5 in the context of portfolio choice problems with a
single risky asset by identifying w0 .1 C /EŒQx with w0 rf , .1 w/
(see Sect. 3.1),
with w and xQ .1 C /EŒQx with rQ rf in terms of the notation used in Sect. 3.1.
Part (i) of the proposition then follows from Proposition 3.2, while part (ii) follows
from Proposition 3.3. Finally, part (iii) can be proved by relying on arguments
analogous to those used in the proof of Proposition 3.5 (see Eeckhoudt et al. [631],
Proposition 3.4 for more details). t
u
Note that, in the case of a DARA utility function, the effect on the optimal
insurance demand of an increase in the insurance premium is ambiguous
(see Eeckhoudt et al. [631, Chapter 3] for more details and compare also with
Proposition 3.5).
Finally, let us extend the above analysis by introducing an additional non-

insurable risk yQ ( background risk), which is supposed to be independent of xQ and
with zero mean. In this case, the optimal insurance problem for a risk averse agent
becomes

max E u w0 C yQ .1 w/Qx w.1 C /EŒQx : (3.48)
w2RC
To analyse the solution to the insurance problem (3.48) in the presence of

background risk, we need to introduce the coefficient of absolute prudence pau , for a
given utility function u (tacitly assumed to be three times differentiable)2:
u000 .x/
pau .x/ D : (3.49)
u00 .x/
We say that a risk averse agent is prudent if his marginal utility u0 is convex, i.e., if he
exhibits a positive coefficient of absolute prudence. Let us denote by w the optimal
insurance demand for a risk averse agent with utility function u in the presence of
the background risk yQ . As in the case without background risk, it can be easily
shown that w D 1 if D 0. However, in the case where there exists a strictly
positive insurance premium > 0, it holds that w < w < 1, provided that
pau > 0 and the coefficients of absolute prudence and of absolute risk aversion are
both decreasing in wealth (see Eeckhoudt & Kimball [632]). In other words, under
these assumptions, a prudent and risk averse agent will buy more insurance when
exposed to the additional source of uncertainty represented by the background risk
yQ , in line with economic intuition.
A similar analysis can be developed for the optimal portfolio choice problem
with a single risky asset in the presence of a background risk yQ , independent of the
random return rQ of the risky asset and with EŒ yQ D 0. In this case, the optimal
portfolio choice problem becomes

max E u w0 rf C yQ C w.Qr rf / : (3.50)
w2R
If the utility function u is increasing and concave and EŒQr > rf , then the optimal
portfolio w consists in a positive investment in the risky asset, thus confirming
the result obtained without background risk (see Proposition 3.1). Moreover, in
the presence of background risk, the optimal investment w in the risky asset is
smaller than the optimal investment w obtained without background risk if the
2
The coefficient of absolute prudence provides an approximation of the precautionary premium
x/ for the random variable Qx, defined as follows: u0 .EŒQx u .Qx// D EŒu0 .Qx/. See also
u .Q
Eeckhoudt & Gollier [630] and the following section for more results on the coefficient of absolute
prudence.
coefficient of absolute risk aversion is decreasing and convex with respect to wealth.
A necessary condition for this to hold is that the coefficient of absolute prudence is
larger than the coefficient of absolute risk aversion (see Exercise 3.32).
3.4 Optimal Saving and Consumption
In the analysis developed so far agents were only interested in maximizing

the expected utility of wealth/consumption at the future date t D 1, while
wealth/consumption at the initial date t D 0 did not enter explicitly into the agents’
utility functions. In the present section, we shall consider portfolio problems where
agents also take into account consumption at the initial date t D 0. In that context,
an agent needs to decide the proportion of the initial wealth to be consumed at
t D 0 and the proportion of wealth to be saved for future consumption at t D 1. In
other words, an agent needs to optimally allocate the initial wealth between current
consumption and saving for future consumption (the general multi-period case will
be treated in Chap. 6).
Let us first recall the optimal saving and consumption problem under certainty, as
discussed at the end of Sect. 1.1. We consider an agent endowed with wealth w0 at
time t D 0 and wealth w1 at time t D 1 and we denote by c0 and c1 the consumption
at t D 0 and t D 1, respectively. We denote by s the part of the initial wealth w0
saved for future consumption and invested in a risk free asset yielding the risk free
rate rf . The optimal saving-consumption problem can then be formalized as follows,
for a given utility function u, assumed to be twice differentiable, strictly increasing
and strictly concave:

max u.w0 s/ C u.w1 C srf / : (3.51)
s2R
The first order necessary condition for the optimality of s amounts to the following
equality:
u0 .w0 s / D u0 .w1 C s rf /rf : (3.52)
Moreover, since the function u is assumed to be strictly concave, condition (3.52) is

also sufficient for the optimality of s .
By relying on the optimality condition (3.52) and applying the implicit func-
tion theorem, we can obtain several comparative statics results on the optimal
saving-consumption problem, analogously to the case of portfolio choice problems
discussed in section “The Case of a Single Risky Asset”. In particular, we can
establish that the optimal saving s is increasing in w0 and decreasing in w1 . The
first claim follows by noting that
@s u00 .w0 s /

D 00 > 0;
@w0 u .w0 s / C u00 .w1 C s rf /rf2
3.4 Optimal Saving and Consumption 111
due to the strict concavity of u. The fact that s is decreasing with respect to w1
can be established similarly. This shows the preference of risk averse agents for
consumption smoothing, i.e., risk averse agents prefer to consume a similar amount
of wealth in both time periods t D 0 and t D 1.
With a similar reasoning, we can also study the effect on the optimal saving s
of a change in the risk free rate rf . Indeed, by the implicit function theorem,
@s u00 .w1 C s rf /s rf u0 .w1 C s rf /

D : (3.53)
@rf u00 .w0 s / C u00 .w1 C s rf /rf2
Observe that, as in the case of portfolio choice problems (see Proposition 3.5), the
overall effect of a change in the risk free rate rf is undetermined, since it depends
both on the income and on the substitution effect. However, if s > 0, a sufficient
condition for the optimal saving s to be increasing with respect to rf is that the
coefficient of relative risk aversion is smaller than one (see Exercise 3.33).
Let us now consider a risky setting where at date t D 1 agents face a risk
represented by a random variable yQ . Without loss of generality, we assume that
EŒ yQ D 0. For example, such a random variable can represent the uncertain labor
income received at time t D 1. Intuitively, for a risk averse agent, this additional
uncertainty should lead to an increase in the optimal saving. In the present risky
context, the optimal saving-consumption problem becomes

max u.w0 s/ C EŒu.w1 C srf C yQ / : (3.54)
s2R
We denote by s the optimal saving in the presence of the additional risk yQ in order
to distinguish it from the optimal saving s obtained under certainty as a solution to
problem (3.51). The optimality condition for the optimal saving s gives
u0 .w0 s / D EŒu0 .w1 C s rf C yQ /rf :
By comparing the above optimality condition with the optimality condition (3.52)
under certainty, we see that there exists an “extra saving” term s s induced by
the future uncertainty represented by the random variable yQ . We call precautionary
saving the difference s s . As shown in the following proposition (see also
Kimball [1100]), the precautionary saving is linked to the coefficient of absolute
prudence introduced in (3.49). As before, we always assume that rf > 0.
Proposition 3.26 In problem (3.54), if pau .x/ 0 for all x 2 RC , then s s 0.
Proof Let the function H W s 7! H.s/, denote the expected utility in correspondence
of a saving level s 2 R, i.e.,
H.s/ WD u.w0 s/ C EŒu.w1 C srf C yQ /:

Note that, since u00 < 0, the function H is concave. As a consequence, in order to
prove that s s 0 it is enough to show that H 0 .s / 0. By (3.52), it holds
that, under the standing assumption EŒ yQ D 0,
H 0 .s / D u0 .w0 s / C rf EŒu0 .w1 C s rf C yQ /

D u0 .w1 C s rf C EŒ yQ /rf C rf EŒu0 .w1 C s rf C yQ /:
Therefore,

H 0 .s / 0 ” EŒu0 .w1 Cs rf C yQ / u0 w1 Cs rf CEŒ yQ D u0 .w1 Cs rf /:
Since pau .x/ 0 for all x 2 RC , the function u0 is convex, so that the inequality on
the right-hand side holds true, as a consequence of Jensen’s inequality. In turn, this
implies that H 0 .s / 0, thus proving the claim. t
u
Results similar to those on risk aversion (see, e.g., Propositions 2.5 and 2.6)
can be also established with respect to prudence. In particular, analogously to the
coefficient of risk aversion, the coefficient of absolute prudence can be increasing
or decreasing in wealth. Decreasing absolute prudence implies that wealthier agents
are less inclined towards precautionary saving. Note that a DARA utility function
exhibits prudence, since the fact that the function x 7! rua .x/ is decreasing implies
that pau .x/ rua .x/, for all x 2 RC , as can be easily checked provided that u000 exists,
see Exercise 2.8.
Concerning the behavior of the optimal saving s with respect to changes in
the risk free rate rf , in the presence of the additional risk yQ , if the coefficient of
relative risk aversion is smaller than one then the optimal saving s is increasing
with respect to the risk free rate rf (compare with Exercise 3.33).
Let us now consider what happens if the riskiness of the labor income represented
by the random variable yQ increases. To this end, we replace the random variable yQ
with a random variable zQ such that zQ d yQ C Q , where Q is a random variable with
EŒQ jQy D 0. In this case, the same arguments used in the proof of Proposition 3.26
allow us to show that if the agent is prudent then the optimal saving will increase.
We conclude the present section by considering the optimal saving-consumption
problem in the presence of a risky asset with random return rQ . This means that the
agent has the possibility of saving at time t D 0 an amount s and investing that
amount into the risky asset, yielding the random wealth sQr at the future time t D 1.
The optimal saving-consumption problem then becomes

max u.w0 s/ C EŒu.w1 C sQr / :
s2R
Let s denote the optimal solution to this problem. As before, the optimality
condition for s gives
u0 .w0 s / D EŒu0 .w1 C s rQ /Qr

and an application of the implicit function theorem confirms that the optimal saving
s is increasing with respect to the initial wealth w0 :
@s u00 .w0 s /

D 00 0;
@w0 u .w0 s / C EŒu00 .w1 C s rQ /Qr2

similarly as above. Concerning the effect on the optimal saving of an increase in the
riskiness of the random return of the risky asset, it can be shown that the optimal
saving s increases if 2u00 .x/ C u000 .x/x > 0 for all x 2 RC , i.e., xpu .x/ > 2. If
the coefficient of relative risk aversion is greater than one and decreasing, then s
will increase as the riskiness of the risky asset return increases (see Exercise 3.34).
In particular, this is the case of a power utility function with relative risk aversion
coefficient greater than one (compare also with Eeckhoudt et al. [631, Section 6.3]).
In a multi-period setting, we will often consider a separable power utility function
of the following form, for b > 0:
b 1 1b b 1 1b
u.x0 ; x1 / D x0 C x :
b1 b1 1
In this case, the absolute risk aversion coefficient is given by 1=.bx/ and the
relative risk aversion coefficient is 1=b. The marginal rate of substitution between
consumption at date t D 0 and consumption at date t D 1 is .x1 =x0 /1=b and the
elasticity of intertemporal substitution of consumption is b. Indeed:
1b
@x1 u.x0 ; x1 / x1
D
@x0 u.x0 ; x1 / x0
and
1b
d log.x1 =x0 / x1 x0 d.x1 =x0 /
D D b:
@x1 u.x0 ;x1 /
d log @x u.x0 ;x1 / x0 x1 d.x1 =x0 / 1b
0
Observe also that the elasticity of intertemporal substitution of consumption b is

the reciprocal of the coefficient of relative risk aversion 1=b. This fact represents
a severe limitation of the time-additive power utility function, as we are going to
discuss in more detail in Chaps. 6 and 9.
We refer the interested reader to Leland [1180] concerning the existence of an

optimal solution to problem (3.2). In Chap. 4, we will also relate the existence of an
optimal portfolio to the absence of arbitrage opportunities.
In the case of a single risky asset, as considered in the first part of Sect. 3.1, it is
rather difficult to provide further comparative statics results beyond Proposition 3.5.
In particular, it is difficult to analyse the behavior of the optimal demand of the
risky asset as the riskiness of the risky asset increases or, similarly, when the initial
wealth becomes random. Intuitively, one would expect that the optimal demand of
a risk averse agent decreases as the riskiness of the asset increases. However, this
is not always the case. Indeed, let us consider the portfolio problem (3.2) with two
assets: a risk free asset with return rf and a risky asset with random return rQ1 and
a positive risk premium. Let us now consider the same problem with a risky return
rQ2 such that rQ1
SSD rQ2 , i.e., rQ2 d rQ1 C Q , where EŒQ jQr1 D 0 (the return rQ2 is
equal in distribution to rQ1 plus a noise component which increases the riskiness).
In Rothschild & Stiglitz [1473] and Hadar & Seo [872], it has been shown that
the demand of the risky asset with return rQ1 by a risk averse agent can be greater
or smaller than that of the risky asset rQ2 . A sufficient condition for the (positive)
demand of the asset with return rQ1 to be larger than that of the risky asset with return
rQ2 is that the coefficient of relative risk aversion is increasing and less or equal than
one and that the coefficient of absolute risk aversion is decreasing (see Rothschild
& Stiglitz [1473]). The same effect is observed if relative prudence is positive and
less than two (see Hadar & Seo [872]). Furthermore, if the change in the riskiness of
the risky asset leads to a random return which is dominated according to first order
stochastic dominance, then the optimal demand will decrease if the coefficient of
relative risk aversion is less than one (see Hadar & Seo [872]).
As a related situation, let us consider the optimal portfolio problem (3.2) in the
case where the initial wealth becomes w0 C Q , where Q is a random variable with
zero mean, independent of rQ . We call Q a background risk (which may be interpreted
as labor income) and denote by w the optimal portfolio in the case of random
initial wealth and by w the optimal portfolio without background risk. One would
guess that, if EŒQr rf > 0, a risk averse agent with a strictly increasing utility
function would choose an optimal portfolio w such that 0 w w , since
the random initial wealth increases the overall risk and, hence, the optimal choice
would be to invest less in the risky asset. Actually, this does not always hold true
for a risk averse agent. It can be shown (Eeckhoudt et al. [631, Proposition 4.3])
that the above conjecture is true under the additional assumption that the coefficient
of absolute risk aversion is decreasing and convex (for instance, the coefficient of
absolute risk aversion of a power utility function is decreasing and convex).
As mentioned above, the second order stochastic dominance criterion does not
allow us to establish that all risk averse agents reduce the demand after a shift in
riskiness dominated according to this criterion. Gollier [799] provides a criterion
(central dominance) that allows to establish that all risk averse agents reduce the
demand of a risky asset after a shift in its distribution. According to this criterion, if
the random return of the risky asset changes from the distribution function F1 to the
distribution function F2 , then every risk averse agent Rreduces his optimal demand of
y
the risky asset if and only if there exists such that 0 .F2 .z/ F1 .z//dz 0, for
all y 2 Œ0; 1 (assuming that the distribution is normalized onto Œ0; 1). The central
dominance and the second order stochastic dominance criterion cannot be compared
(see also Gollier [800, Chapter 6]).
Further stochastic dominance results have been established by several authors

(we refer to Levy [1204, 1205], Sriboonchitta et al. [1560] for surveys). If two asset
returns are independently distributed with positive variance and the same mean (in
particular, if they are identically distributed), then both assets must enter the optimal
portfolio with a strictly positive weight, see Hadar & Russell [869], Hadar et al.
[871]. The result has been extended in Hadar & Russell [870] to the case of two
dependent risky assets provided that their marginal distributions are identical.
Ross [1468] establishes the following interesting result, making use of the notion
of strong risk aversion (see Sect. 2.5). Consider two risky assets, the first one with a
higher expected return and riskier than the second one (i.e., two random returns rQ 1
and rQ2 such that EŒQr1 rQ2 jQr2 0). If agent a is strongly more risk averse than agent
b, then the optimal demand of agent a of the first asset will be less than that of agent
b. Agent a, being strongly more risk averse, will choose a less risky portfolio with a
lower expected return. When the initial wealth is stochastic and the asset has a non-
negative risk premium conditional on wealth, the strongly more risk averse agent’s
demand will be smaller than that of a less risk averse agent (see Ross [1468]).
An abstract and general framework allowing for a unified treatment of several
mean-variance optimization problems has been recently proposed in Fontana &
Schweizer [725] by relying on Hilbert space methods. The portfolio frontier has
been derived allowing for short sales (see e.g. Dybvig [606]): imposing a no short
sale constraint on risky assets (i.e., wn 0, for all n D 1; : : : ; N), the portfolio
frontier of course changes. Indeed, if short sale constraints are introduced, then we
can observe kinks in the portfolio frontier.
In Proposition 3.11 we have provided a general two mutual funds separation
result in the case of HARA utility functions, in the presence of a risk free asset
(monetary separation). In the absence of a risk free asset, a two mutual funds
separation result still holds with respect to two risky mutual funds (two risky mutual
funds separation) for any utility function belonging to the HARA class, provided
that a complete set of Arrow securities is traded (i.e., markets are complete), as
shown in Cass & Stiglitz [376, Theorem 4.1]. For a general set of traded securities,
in a possibly incomplete market, the class of utility functions such that the two funds
separation result holds is restricted to the class of utility functions satisfying
u0 .x/ D bxc or u0 .x/ D a C bx;
see Cass & Stiglitz [376, Theorem 5.1]. Quadratic utility functions as well as CRRA
utility functions (power and logarithmic utility functions) satisfy this condition.
Mutual fund separation results have been tested empirically. In particular, two
mutual funds (monetary) separation results have been tested empirically with
negative results, see Lo & Wang [1240] for an analysis based on trading volume
and Canner et al. [360] for an analysis based on portfolio allocations proposed by
financial advisors (for a reconciliation see Bajeux-Besnainou et al. [182]). Kroll
et al. [1137] present experimental evidence showing that people often choose
inefficient portfolios and violate the two mutual funds separation.
3.6 Exercises
Exercise 3.1 Consider a strictly risk averse agent endowed with initial wealth w0
and with a strictly increasing and twice differentiable utility function. Let rf and rQ
denote the return of the risk free asset and of the risky asset, respectively. Show that
the minimum risk premium EŒQr rf of the risky asset required by the agent to invest
the totality of his wealth in the risky asset approximatively satisfies the following
inequality
EŒQr rf rua .w0 rf /w0 EŒ.Qr rf /2 :
Exercise 3.2 Consider a quadratic utility function u.x/ D x b2 x2 , an initial wealth

w0 D 100, a risk free rate rf D 1:1 and a risky asset with expected return EŒQr D 1:3
and variance 2 .Qr/ D 1:5.
(i) Determine the optimal portfolio w for b D 0:006.
(ii) Determine for which values of b we have w < 0.
(iii) How does w change if the risk free rate increases to rf0 D 1:2?
(iv) How does w change if the initial wealth increases to w00 D 150?
(v) How does w change if its expected return decreases to EŒQr0 D 1:2?
Exercise 3.3 Consider the optimal portfolio choice problem in the presence of N
risky assets with returns .Qr1 ; : : : ; rQN / and of a risk free asset with return rf > 0.
Suppose that there are no redundant assets in the economy in the sense that the
random variables .rf ; rQ1 ; : : : ; rQN / are linearly independent. Show that the optimal
portfolio w 2 RN solving problem (3.2), for a strictly increasing and strictly
concave utility function u, is unique.
Exercise 3.4 (LeRoy & Werner [1191], Theorem 13.5.1) Consider a strictly
increasing and strictly concave utility function u and suppose that there exist N
risky assets whose returns .Qr1 ; : : : ; rQN / admit the representation
X
N
rQn D k rQk C Qn ; for all n D 1; : : : ; N;
kD1
k¤n
PN
where kD1;k¤n k D 1, for all n D 1; : : : ; N, and Qn is a random variable satisfying
EŒQ n jQr1 ; : : : ; rQn1 ; rQnC1 ; : : : ; rQN D EŒQ n ; (3.55)
for all n D 1; : : : ; N. Show that wn > 0 if and only if EŒQ n > 0, for all n D 1; : : : ; N.
3.6 Exercises 117
Exercise 3.5 (LeRoy & Werner [1191], Corollary 13.5.2) Consider a strictly
increasing and strictly concave utility function u and suppose that there exist a risk
free asset with return rf > 0 and N risky assets whose returns .Qr1 ; : : : ; rQN / satisfy
the condition
EŒQrn jQr1 ; : : : ; rQn1 ; rQnC1 ; : : : ; rQN D EŒQrn ;
for all n D 1; : : : ; N. Show that wn > 0 if and only if EŒQrn > rf , for all n D
1; : : : ; N.
Exercise 3.6 Consider two risky assets with returns rQ1 ; rQ2 , with corresponding
expected returns e1 ; e2 , variances 12 ; 22 and correlation . Let .w; 1 w/ denote
a portfolio of the two risky assets, with corresponding expected return EŒQr and
variance 2 .Qr/. Verify the following claims:
(i) if 0 w 1 then 2 .Qr/ maxf12 I 22 g;
(ii) if D 1 then 2 .Qr/ D 0 for w D 2 =.1 2 /;
(iii) if D 1 then 2 .Qr/ D 0 for w D 2 =.1 C 2 /;
Exercise 3.7 In the context of Proposition 3.4, show that EŒu00 .W e /W
e .Qr rf / 0
if the utility function u exhibits decreasing relative risk aversion.
Exercise 3.8 Consider an economy with a risk free asset with return rf and a risky
asset whose random return rQ can take two possible values fd; ug with probabilities
f; 1 g, respectively. Assume that d < rf < u. Determine the optimal demand of
the risky asset of an agent endowed with initial wealth w0 according to the following
utility functions:
p
(i) u.x/ D x;
(ii) u.x/ D log.x/;
(iii) u.x/ D x = , with ¤ 1.
Exercise 3.9 Consider an exponential utility function u.x/ D 1a exp.ax/, with
a > 0, and an economy with a risk free asset with return rf D 1:1 and two risky
assets with random returns rQ1 ; rQ2 distributed as normal random variables with means
e1 D 1:2 and e2 D 1:3 and variances 12 D 4 and 22 D 9, respectively, with
correlation coefficient 2 .1; 1/.
(i) Determine when the optimal portfolio is diversified, i.e., when w1 ; w2 > 0.
(ii) Suppose that is such that the optimal portfolio is diversified (take for instance
D 0:5). If increases, how should 1 and 2 vary in order for the optimal
portfolio to remain diversified?
(iii) Suppose that 2 f0:5; 0:5g. Give conditions on a such that the optimal
portfolio invests more than one unit of wealth in the risky assets.
Exercise 3.10 Show that the covariance between the returns rQw1 and rQw2 of two
frontier portfolios w1 and w2 is given by formula (3.24).
Exercise 3.11 Let wMVP denote the minimum variance portfolio. Show that, for any
frontier portfolio w , it holds that Cov.QrwMVP ; rQw / D 1=C D 2 .QrwMVP /.
Exercise 3.12 Given an arbitrary frontier portfolio w , with w ¤ wMVP , there
exists a unique frontier portfolio wzc such that Cov.Qrw ; rQwzc / D 0.
Exercise 3.13 Determine the frontier portfolio w such that its variance is equal
to the variance of its zero correlation portfolio, i.e., determine the frontier portfolio
w such that 2 .Qrw / D 2 .Qrwzc /, where wzc is the zero correlation portfolio with
respect to w .
Exercise 3.14 Given a frontier portfolio w , show that the expected return EŒQrwzc
of its zero correlation portfolio wzc is identified, in the variance-expected return
plane, by the intersection of the line connecting the points . 2 .Qrw /; EŒQrw / and
. 2 .QrwMVP /; EŒQrwMVP / with the vertical axis. Similarly, show that EŒQrwzc is identified,
in the standard deviation-expected return plane, by the intersection of the tangent to
the portfolio frontier at the point . 2 .Qrw /; EŒQrw / with the vertical axis.
Exercise 3.15 Consider a portfolio wp … PF. Show that, in the variance -
expected return plane, the line which connects the two points . 2 .Qrwp /; EŒQrwp /
and . 2 .QrwMVP /; EŒQrwMVP / intercepts the expected return axis at the level EŒQrwq ,
where wq is the portfolio such that Cov.Qrwq ; rQwp / D 0 and with the minimum
variance among all portfolios with zero correlation with wp .
Exercise 3.16 Let us consider an economy with two risky assets with returns rQ1 and
rQ2 . Show that the explicit formula w D . EŒQr2 /=.EŒQr1 EŒQr2 / is a special case
of formula (3.16) in the case N D 2.
Exercise 3.17 Let wp be a frontier portfolio and wq be an arbitrary portfolio (i.e.,
not necessarily belonging to the portfolio frontier) such that EŒQrwq D EŒQrwp . Show
that Cov.Qrwq ; rQwp / D 2 .Qrwp /.
Exercise 3.18 Consider an economy with N risky assets with random returns
.Qr1 ; : : : ; rQN / and a risk free asset with return rf , as in section “The Case of N
Risky Assets and a Risk Free Asset”, and suppose that rf ¤ A=C. Show that any
portfolio w belonging to the portfolio frontier PF can be expressed as the linear
combination of the risk free asset and the tangent portfolio we , so that
e rf 1
w D ˛we D ˛ ;
1> V 1 .e rf 1/
for some ˛, where the second equality follows from the proof of Proposition 3.15
Exercise 3.19 Let > A=C and consider the following optimization problem:
w> e
max p ; (3.56)
w2 N w> Vw
3.6 Exercises 119
corresponding to the maximization of the Sharpe ratio with respect to the reference
rate of return over the set of all portfolios investing in the N risky assets. Prove
that, for any > A=C, the solution w to problem (3.56) is given by
A B V 1 .e 1/
w D g C h D > 1 : (3.57)
C A 1 V .e 1/
Exercise 3.20 Consider an economy with N risky assets with returns .Qr1 ; : : : ; rQN /
and a risk free asset with return rf , with rf < EŒQrn for all n D 1; : : : ; N. Show that
the tangent portfolio we is diversified if the risky asset returns are uncorrelated.
Exercise 3.21 Consider a risk free asset with return rf D 1:1 and two risky assets
with normally distributed returns .Qr1 ; rQ2 / with EŒQr1 D 1:2, EŒQr2 D 1:3 and
2 .Qr1 / D 4, 2 .Qr2 / D 9 and correlation .
(i) Determine the portfolio frontier composed by the two risky assets only.
(ii) In the case where D 0:5, determine the minimum variance portfolio wMVP
and the corresponding variance 2 .QrwMVP /.
(iii) In the case where D 0:5, determine the tangent portfolio we .
(iv) Consider the problem of an agent maximizing the expected exponential utility
(with u.x/ D exp.ax/). Show that the optimal portfolio w .a/ for such
an agent is given by a multiple of the tangent portfolio we , as shown at the
end of Sect. 3.2. Give conditions on the risk aversion parameter a so that the
investment in the portfolio we is greater than one.
(v) Determine the optimal portfolio that minimizes the variance, with and without
the risk free asset, for the given expected return D 1:25.
(vi) Verify that the portfolio w0 D .0:6; 0:5/ (with the remaining proportion of
wealth being invested in the risk free asset) does not belong to the portfolio
frontier.
Exercise 3.22 Consider an economy with a risk free asset with return rf D 1:13
and two risky assets with normally distributed returns with expected values EŒQr1 D
1:16, EŒQr2 D 1:25 and variances 2 .Qr1 / D 2, 2 .Qr2 / D 4 and correlation .
(i) Determine the portfolio frontier composed by the risky assets in the two cases
D 0:5 and D 0:5.
(ii) Determine the minimum variance portfolio wMVP in the two cases D 0:5 and
D 0:5.
(iii) Determine the tangent portfolio in the two cases D 0:5 and D 0:5.
(iv) For D 0:5, consider an agent with quadratic utility function u.x/ D x b2 x2
and determine his optimal portfolio. Give conditions on b in order that the
optimal investment in the tangent portfolio we is greater than one.
Exercise 3.23 Consider two risky assets with random returns rQ1 and rQ2 , with
expected values e1 and e2 , respectively, and variances 12 and 22 (with 12 22 ),
respectively, and correlation . For a portfolio .w; 1w/, denote by EŒQrw and 2 .Qrw /
the expectation and the variance, respectively, of the corresponding random return
rQw . Verify the following claims.
(i) If < 1 =2 then there exists w 2 .0; 1/ such that .Qrw / < 1 and, for all
w … Œ0; 1, it holds that .Qrw / 1 .
(ii) If D 1 =2 then .Qrw / > 1 for every w 2 R.
(iii) If > 1 =2 then there exists w … .0; 1/ such that .Qrw / < 1 and, for all
w 2 .0; 1/, it holds that .Qrw / > 1 .
(iv) Show that, if 1 D 2 DW , then the minimum variance portfolio is given by
wMVP D .1=2; 1=2/>, independently of the value of the correlation coefficient
. What happens in the case D 1?
Exercise 3.24 Show that, if the mutual fund separation property holds for K D 1,
i.e., there exists a portfolio w such that EŒu.Qrw / EŒu.Qrw / for any w 2 N
and for any concave utility function u, then w must coincide with the minimum
variance portfolio.
Exercise 3.25 Consider an economy with a risk free asset with return rf and N risky
assets with i.i.d. random returns rQ D .Qr1 ; : : : ; rQN /. Show that the two mutual funds
separation property holds with respect to the risk free asset and the equally weighted
portfolio .1=N; : : : ; 1=N/> of the risky assets.
Exercise 3.26 Let us consider an economy with N risky assets with normally
distributed returns rQ D .Qr1 ; : : : ; rQN / with EŒQrn D EŒQr1 , for all n D 2; : : : ; N, and
let wMVP denote the minimum variance portfolio. Show that the one fund separation
property holds with respect to wMVP .
Exercise 3.27 Consider the optimal insurance problem of an agent with quadratic
utility function u.x/ D x b2 x2 , with initial wealth w0 , exposed to the possibility of
a loss D > 0 which can occur at time t D 1 with probability 2 .0; 1/. Let p be
the price of one unit of wealth contingent on the occurrence of the loss event and
denote by w the agent’s optimal insurance demand. Verify that w D D if p D
and that w < D if p > .
Exercise 3.28 Consider a risky setting with two possible states of the world at the
time t D 1, with probabilities and 1 , and an agent with power utility function
u.x/ D x , with 2 .0; 1/. Verify that, if the prices of wealth contingent on the
two states of the world are not fair, i.e., p1 =p2 > =.1 /, then the optimal
consumption in the two states of the world satisfies x1 < x2 .
Exercise 3.29 In the setting of Proposition 3.24, for a given utility function u,
define by

Cov xQ ; u0 .w0 xQ /
D :
EŒQxEŒu0 .w0 xQ /
3.6 Exercises 121
Show that, if the insurance premium is greater than , then the optimal insurance
demand w consists in buying zero units of the insurance contract.
Exercise 3.30 Consider a normally distributed loss xQ N .; 2 / with > 0 and
an agent with exponential utility function u.x/ D exp.ax/=a, for a > 0. Verify
that, in the setting of Proposition 3.24, if > 0, then the optimal insurance demand
w is less than one.
Exercise 3.31 (Eeckhoudt et al. [631], Proposition 3.5) Consider a random loss
xQ which can take N Ppossible ordered values 0 < x1 < : : : < xN with probabilities
1 ; : : : ; N , with NnD1 n D 1. Consider an insurance contract paying a non-
negative indemnity I.Qx/ at time t D 1, with price p D .1 C /EŒI.Qx/ at time
t D 0. Show that, for any risk averse agent, an insurance contract with an indemnity
of the form I .Qx/ D maxf0I xQ Kg, for some K > 0 is optimal among all insurance
contracts with price p.
Exercise 3.32 (Eeckhoudt et al. [631], Proposition 4.3) In the same setting as
at the end of Sect. 3.3, show that, in the presence of background risk, the optimal
investment w in the risky asset is smaller than the optimal investment w obtained
in the case without background risk if the coefficient of absolute risk aversion is
decreasing and convex with respect to wealth. Show also that a necessary condition
is that the coefficient of absolute prudence is larger than the coefficient of absolute
risk aversion.
Exercise 3.33 Consider the optimal saving-consumption problem under certainty,
as presented at the beginning of Sect. 3.4. Show that the optimal saving s is
increasing with respect to the risk free rate rf if the coefficient of relative risk
aversion is less than one.
Exercise 3.34 In the context of the optimal saving-consumption problem in the
presence of a risky asset, as described at the end of Sect. 3.4, show that if the
coefficient of relative risk aversion is greater than one and decreasing, then the
condition xpu .x/ > 2, for all x 2 R, is satisfied. In particular, the optimal saving
s increases with respect to an increase in the riskiness of the risky asset’s return.
Chapter 4
General Equilibrium Theory and No-Arbitrage
An economy is in equilibrium when it produces messages which

do not induce the agents to modify the theories they believe in or
the policies which they pursue.
Hahn (1973)
In a rational expectations equilibrium, not only are prices

determined so as to equate supply and demand, but individual
economic agents correctly perceive the true relationship
between the non price information received by the market
participants and the resulting equilibrium market price. This
contrasts with the ordinary concept of equilibrium in which the
agents respond to prices but do not attempt to infer other
agents’ non price information from actual market prices.
Radner (1982)
Relying on the perfect competition hypothesis, similarly to the situation considered

in Chap. 1 under certainty, the analysis of an economy under risk can be decom-
posed into two steps: first, the agents’ individual behavior is studied by taking
prices/returns as given (internal consistency); in a second step, the interaction of the
agents in the market is studied. In this second step, the focus is on the determination
of an equilibrium price vector, i.e., a price vector such that the market demand equals
the total supply and, therefore, agents’ decisions are compatible among themselves
(external consistency). In Chaps. 2 and 3, we have addressed only the first step of
the problem, assuming that at the initial date t D 0 agents trade wealth contingent
on the state of the world realized in t D 1 or assets with dividends described by
generic random variables. In the present chapter, we will address the second step.
In order to illustrate the topics dealt with in this chapter, let us first present
an elementary example. Consider a farmer who owns a field of corn and, at date
t D 0, has to take a consumption-investment decision. At t D 0, the farmer must
decide how to allocate among a set of assets the wealth obtained from the previous
harvest (investment decision), deciding how much to consume today (at t D 0) and
how much to consume at the future date t D 1 in correspondence of the possible
states of the world (consumption decision). We will consider two different settings:
investment decisions taken at t D 0 with consumption only at the future date t D 1
and investment decisions taken at t D 0 with consumption at both dates t D 0
and t D 1. In order to allocate his wealth, our farmer has the possibility of trading

124 4 General Equilibrium Theory and No-Arbitrage
corn in markets open at t D 0 for assets delivering goods at the future date t D 1
( future markets) and markets open at t D 1 for immediate delivery (spot markets).
To simplify the analysis, we will consider two stylized settings: when agents can
consume only at t D 1, there exist L > 1 goods; when agents can consume at
t D 0 as well as at t D 1, then there exists a single good (L D 1), generically
identified as wealth. As in the previous chapter, we shall restrict our attention to a
finite probability space, meaning that we only consider a finite number S > 1 of
elementary states of the world that can be realized at t D 1. For instance, in the
case of our farmer, each elementary event fully describes the weather conditions
prevailing in the time period between t D 0 and t D 1 (since the crop obtained at
t D 1 depends on the weather conditions). Let us suppose that there are I 2 agents
(farmers) in the economy, where every agent is fully described by a preference
relation and by an endowment. We shall always make the crucial assumption that the
probability space describing the uncertainty of the economy as well as the agents’
rationality are common knowledge among the agents.1 Furthermore, we will assume
that the preference relation of each agent satisfies the axioms ensuring the expected
utility representation (see Sect. 2.1). Note that most of the results presented in this
chapter can be also established under the assumption that agents have heterogeneous
beliefs.
To develop our analysis, we introduce the concept of contingent good. Indeed,
goods differ not only for their intrinsic features, but also for the state of the world
in which they become available. For instance, in the case of a farmer owning a
field of corn, a unit of corn at t D 1 in the case where weather conditions have
been unfavorable and a unit of corn at t D 1 in the case where weather conditions
have been favorable are two rather different goods from the point of view of a
farmer who has to take decisions at t D 0. Hence, if there are S elementary
events and the farmer cultivates L > 1 goods, then the economy comprises L S
contingent goods. When consumption is only allowed at t D 1, recalling that in
this case we consider L > 1 goods, agent i (for i D 1; : : : ; I) is endowed at
the initial date t D 0 with a basket ei 2 RLS of contingent goods (in the case
of a farmer, the latter represents the crop obtained from the field at t D 1). For
l D 1; : : : ; L, row l of the basket of goods ei represents the quantity of good l
obtained in correspondence of the S possible states of the world .!1 ; : : : ; !S /, while
column s represents the quantities of the L goods obtained in correspondence of
the state of the world !s , for s D 1; : : : ; S. Given a basket of contingent goods
x 2 RLS , we denote by xls the amount of good l in state of the world !s and by
xs 2 RL the vector of L goods available in state of the world !s . When consumption
is allowed both at t D 0 and at t D 1 (so that L D 1), a basket of goods
is simply described by a vector in RSC1 , where the first component represents
1
Some elements of the economy are common knowledge among the agents if each agent knows
them, knows that the other agents know them, knows that the other agents know that he knows
and so on. In that context, agents cannot agree to disagree (see Fuydenberg & Tirole [746] for a
rigorous definition of common knowledge).
4 General Equilibrium Theory and No-Arbitrage 125
wealth/consumption at time t D 0 and the remaining S components represent

wealth/consumption at time t D 1 in correspondence of the S possible states of
the world.
This simple setting captures two essential features of financial markets: the
intertemporal dimension of agents’ decisions and their intrinsic riskiness. The
forward-looking feature of economic decisions is due to the fact that, at the initial
date t D 0, an agent faces different possible states of the world which will be
realized at the future date t D 1 and, hence, wealth obtained at t D 1 is risky.
Concerning the possible realizations of the state of the world at t D 1, an agent
describes his beliefs through a probability measure P on the finite probability space
represented by ˝ D .!1 ; : : : ; !S /. However, from the point of view of an agent,
a probability measure alone does not suffice in order to solve his decision making
problem. Indeed, if other agents are present in the economy, then every agent needs
to formulate forecasts on the behavior of the other agents. This happens exactly
when there are both future markets at t D 0 (i.e., markets open at t D 0 for assets
delivering wealth/goods at t D 1) and spot markets at t D 1. In this case, taking
their decisions at t D 0, agents have to forecast spot market prices at t D 1, which
in turn depend on the behavior of all the agents in the economy.
A crucial point of the analysis consists in understanding how agents form their
beliefs/expectations at t D 0 about future prices at t D 1. In our analysis, we
will make the rational expectations hypothesis, which amounts to the following: a)
the economic model is common knowledge among the agents (where by economic
model we generically refer to all those features that turn out to be relevant for the
solution to the decision problem); b) agents formulate their expectations by fully
exploiting all the available information. In addition, we suppose that these two
assumptions are common knowledge among all the agents themselves.
An assumption similar to the rational expectations hypothesis has been already
introduced in Chap. 1 to strengthen the capability of general equilibrium theory to
describe the economic world under certainty. More specifically, we assumed that
the economic model as well as agents’ rationality are common knowledge among
the agents themselves and that agents fully exploit this common knowledge when
taking their decisions. Note that, even without this assumption, general equilibrium
theory in a riskless environment is able to show that agents’ choices are compatible
in a perfectly competitive market when agents pursue their self-interest. However,
in a risky environment, the rational expectations hypothesis plays a much more
relevant role. Indeed, unlike in a riskless environment, in a risky economy the
rational expectations hypothesis is necessary to define the agents’ behavior and
the equilibrium. In particular, the rational expectations hypothesis is needed to
define the agents’ expectations about future prices, which are in turn needed to
determine an agent’s optimal choice. Without this hypothesis, or a similar one, we
would not be able to establish the compatibility of agents’ decisions in a perfectly
competitive market. The rational expectations hypothesis will represent the third
consistency requirement (informational consistency) and we will show that, in a
rational expectations equilibrium, forecasts are self-confirming (perfect forecasts).
We point out that one could replace the rational expectations assumption with a more
realistic assumption about the agents’ knowledge of the economy, but in that case
agents’ expectations would not be self-confirming (thus leading to biased forecasts),
meaning that there is space for learning. We will return to this topic in Chap. 9
and we refer the reader to Radner [1439] for a critical introduction to rational
expectations equilibrium theory.
Our analysis of agents’ interactions in the economy will be based on the notion
of equilibrium and we shall evaluate the efficiency of an allocation by relying on
the Pareto optimality criterion, as introduced in Sect. 1.3. In particular, we will
study the characteristics of future and spot markets allowing us to extend to a risky
environment the important relationship between equilibrium and Pareto optimality
established in Sect. 1.3 under certainty.
Extending the general equilibrium analysis to a risky environment, we shall also
address the valuation problem. Given the fundamentals of the economy (dividends’
distribution, agents’ preferences and endowments), the general valuation problem
consists in determining the prices of financial assets. We address this problem
from two different perspectives: equilibrium analysis and no-arbitrage pricing. In
the first case, equilibrium analysis allows to determine asset prices by assuming
that each agent maximizes his expected utility and markets clear (i.e., the total
demand equals the total supply). In particular, the equilibrium-based approach to
asset pricing allows to relate the price of an asset to the agents’ preferences and to
their endowments. On the other hand, adopting the no-arbitrage perspective, asset
prices will be determined by assuming that the market does not allow for arbitrage
opportunities (loosely speaking, the market does not allow for opportunities to make
money out of nothing without incurring in any risk). In this case, an asset’s price
is determined without any reference to the agents’ preferences and endowments.
We shall see that, according to the equilibrium-based valuation, the price of an
asset will be given by the expectation of the future dividend multiplied by the
marginal rate of substitution of an agent (pricing kernel), while, according to the
no-arbitrage paradigm, the price of an asset will be given as the expectation of
the future dividend with respect to a risk neutral probability measure, discounted
by the risk free rate. In particular, the risk neutral probability measure will be
different from the probability measure that describes the economy in the real world
(historical/statistical probability) and will be derived from the assumption that the
financial market does not allow for arbitrage opportunities. Moreover, we shall also
discuss the relation between these two valuation paradigms.
This chapter is structured as follows. In Sect. 4.1, we study the notion of
Pareto optimality in the presence of risk and explore its implications in terms
of risk sharing. In Sect. 4.2, considering different market settings, we introduce
the notion of rational expectations equilibrium. In Sect. 4.3, we deal with the
intertemporal consumption problem in a simple two-period setting and we study
the relations between equilibrium, market completeness, Pareto optimality and
aggregation. Section 4.4 introduces the notion of arbitrage opportunity and presents
the fundamental theorem of asset pricing and its implications for the valuation of
financial securities. At the end of the chapter, we provide a guide to further readings
as well as a series of exercises.
4.1 Pareto Optimality and Risk Sharing 127
4.1 Pareto Optimality and Risk Sharing
Let us consider a two-period economy (i.e., t 2 f0; 1g) with I agents (i D 1; : : : ; I),
L goods (l D 1; : : : ; L) and S elementary states of the world (s D 1; : : : ; S) realized
at date t D 1, with I 2, L 1 and S 2. As discussed in the introduction to the
present chapter, all agents share homogeneous beliefs, meaning that the probability
space .˝; / as well as the probability measure P are common knowledge among
the agents. We furthermore assume that each elementary event P !s occurs with a
strictly positive probability s > 0, for all s D 1; : : : ; S, with SsD1 s D 1. In
the present section, we shall consider two different situations: consumption only at
t D 1 and consumption at both dates t D 0 and t D 1. We start from the first case,
assuming that there are L > 1 goods.
Assuming that consumption is only allowed at the future date t D 1, we consider
I risk averse agents characterized by preference relations admitting the expected
utility representation with state independent utility functions defined on the basket
of consumption goods in t D 1, meaning that ui W RLC ! R, for all i D 1; : : : ; I.
Each agent i is characterized by the couple .ei ; ui /, where ui denotes his utility
function and ei 2 RC LS
represents his endowment in terms of quantities of the L
goods available in the S states of the world at t D 1. We shall always tacitly assume
that utility functions are differentiable, strictly increasing and strictly concave,
unless otherwise mentioned.
Since the state of the world realized at t D 1 is unknown at the initial date t D 0
when decisions have to be taken, we can formulate two different notions of Pareto
optimality. In the following, an allocation fxi 2 RC LS
I i D 1; : : : ; Ig is said to be
feasible
PI if the
PI feasibility constraint is satisfied in every possible state of the world:
x
iD1 s
i
e
iD1 s
i
for all s D 1; : : : ; S. Note also that, for each possible state of
the world, this is a set of L constraints (one for each good).
Definition 4.1 A feasible allocation fxi I i D 1; : : : ; Ig is ex-ante Pareto optimal if
0
there is no feasible allocation fxi I i D 1; : : : ; Ig such that
X
S
0
X
S
s ui .xis / s ui .xis /; for all i D 1; : : : ; I;
sD1 sD1
with at least one strict inequality, for some i 2 f1; : : : ; Ig.
Definition 4.2 A feasible allocation fxi I i D 1; : : : ; Ig is ex-post Pareto optimal if

0
there is no feasible allocation fxi I i D 1; : : : ; Ig such that
0
ui .xis / ui .xis /; for all i D 1; : : : ; I and s D 1; : : : ; S;
with at least one strict inequality, for some i 2 f1; : : : ; Ig and s 2 f1; : : : ; Sg.
The two above definitions of Pareto optimality evaluate the optimality of an
allocation fxi I i D 1; : : : ; Ig by referring to the different information available to
the agents in correspondence of the two dates t D 0 and t D 1. Indeed, at t D 0

(ex-ante) the state of the world is still unknown and, therefore, utilities associated to
consumption in the different states of the world are weighted by the corresponding
probabilities, while at t D 1 (ex-post) the state of the world is fully revealed.
In Sect. 1.3, we have characterized Pareto optimal allocations under certainty
in two distinct ways: a) maximizing the utility function of each single agent given
the utility levels of all other agents as well as the feasibility constraint imposed by
the available resources; b) maximizing a linear combination with positive weights
of the agents’ utility functions (social welfare function) subject to the feasibility
constraint. These two approaches can also be employed in the present risky setting.
A Pareto optimal allocation x 2 RC ILS
can be characterized similarly as in
Chap. 1 by requiring that the marginal rate of substitution between any couple of
goods is the same for all the agents of the economy. Note that, in the present risky
setting, we have LS different goods, since the L consumption goods are contingent
on the S possible states of the world, so that the total number of goods is L S
(contingent goods). Hence, considering an interior ex-ante Pareto optimal allocation
fxi 2 RCCLS
I i D 1; : : : ; Ig, the following condition must hold:
s uixls .xi
s / s uxjls .xj
s /
D ; for all i; j D 1; : : : ; I; s; r D 1; : : : ; S and k; l D 1; : : : ; L;
r uixkr .xi
r / j j
r uxkr .xr /
(4.1)
where uixls denotes the partial derivative of the utility function ui with respect to xls ,
for l D 1; : : : ; L, s D 1; : : : ; S and i D 1; : : : ; I. In the case of an ex-post Pareto
optimal allocation, it holds that
uixls .xi
s / uxjls .xj
s /
D ; for all i; j D 1; : : : ; I; s; D 1; : : : ; S; and k; l D 1; : : : ; L:
uixks .xi
s /
j j
uxks .xs /
(4.2)
It is easy to see that an ex-ante Pareto optimal allocation is also ex-post Pareto
optimal. Indeed, this follows directly from Definitions 4.1 and 4.2 and can also be
verified from conditions (4.1)–(4.2) (it suffices to take s D r in (4.1)).
The set of Pareto optimal allocations defines the contract curve of the economy,
as we are now going to illustrate by means of the Edgeworth box in the simple
case of two agents (I D 2), two possible states of the world (S D 2) and a single
consumption good (L D 1). We consider two agents (i D a; b), two states of the
world (s D 1; 2) with strictly positive probabilities .; 1 / and a unique good,
which we can think of as corn, recalling the example given in the introduction to this
chapter. Suppose that the field of farmer a produces the crop ea D .ea1 ; ea2 / 2 R2C
in the two states of the world f1; 2g, while the field of farmer b produces the crop
eb D .eb1 ; eb2 / 2 R2C . The expected utility of agent i 2 fa; bg is given by
ui .xi1 / C .1 /ui .xi2 /

and, in view of condition (4.1), an allocation x 2 R22

CC is ex-ante Pareto optimal if
the following condition holds:
0 0
ua .xa
1 / ub .xb
1 /
0 D : (4.3)
.1 /u .x2 /
a a
.1 /u 0 .xb
b
2 /
Condition (4.3) identifies the allocations belonging to the contract curve (i.e., ex-
ante Pareto optimal allocations) as those allocations in correspondence of which
the indifference curves of the two agents a and b are tangent. In order to develop
some intuition on the concept of (ex-ante) Pareto optimality, let us characterize the
contract curve in three particular cases:
a) ea1 C eb1 D ea2 C eb2 (no aggregate risk). In this case, the aggregate endowment
of the economy (i.e., the aggregate crop of the two farmers) is constant in the two
states of the world, so that the Edgeworth box is a square, see Fig. 4.1. In particular,
the certainty line of agent a coincides with that of agent b. By the analysis developed
in Sect. 3.3, in correspondence of the certainty line both agents have a marginal rate
of substitution for consumption in the two states of the world equal to the ratio
=.1 / of the probabilities of the two states of the world. Therefore, in view of
condition (4.3), the certainty line of the two farmers also coincides with the contract
x1 ub
x2
x2
ua x1
Fig. 4.1 Contract curve with no aggregate risk

curve. This implies that, if there is no aggregate risk in the economy, then Pareto
optimal allocations are characterized by full mutual insurance between the two
agents. Indeed, in correspondence of a Pareto optimal allocation, the consumption of
each agent at t D 1 is the same in both possible states of the world. By the analysis
developed in Sect. 3.3, the prices . p1 ; p2 / at t D 0 for the consumption good at
t D 1 contingent on the two states of the world (Arrow securities) implementing
a Pareto optimal allocation are actuarially fair, in the sense that the relative price
is equal to the ratio of the probabilities: p1 =p2 D =.1 /. In particular, it
can be easily verified that in equilibrium the consumption of both agents in both
states of the world is equal to the expectation of their respective endowments, i.e.,
1 D x2 D e1 C .1 /e2 , for i 2 fa; bg.
xi i i i
b) e1 Ce1 > e2 Ce2 (aggregate risk). In this case, the aggregate endowment of the
a b a b
economy in the state of the world !1 is larger than that in the state of the world !2 ,
meaning that there is aggregate risk. In the presence of aggregate risk, the Edgeworth
box is a rectangle, see Fig. 4.2. The certainty lines of the two agents a and b do not
coincide and the contract curve lies in the region between the certainty lines of
the two agents. As a consequence, the prices of the Arrow securities implementing
a Pareto optimal allocation are not actuarially fair and reflect the aggregate risk.
The ratio between the price of the consumption good contingent on !1 (the state
of the world corresponding to a rich harvest) and the price of the consumption
good contingent on !2 (the state of the world corresponding to a poor harvest)
implementing a Pareto optimal allocation is smaller than the actuarially fair relative
price, reflecting the relative availability/scarcity of the good: p1 =p2 < =.1 /.
As a matter of fact, if p1 =p2 =.1 /, then the tangency point between the
budget constraint and the indifference curve of agent a would be on the left of his
certainty line and similarly for agent b. This would contradict the existence of a
tangency point between the indifference curves of the two agents. On the other hand,
x1 ub
x2
x2
ua x1
Fig. 4.2 Contract curve with aggregate risk

if p1 =p2 < =.1 /, the tangency condition is verified inside the certainty lines of
the two agents (see Exercise 4.1).
c) One of the two agents is risk neutral. Suppose that agent a is risk neutral.
In this case, the indifference curves of agent a are straight lines with slope
=.1 /. The contract curve is identified by the certainty line of agent b and,
in correspondence of a Pareto optimal allocation, the risk neutral agent provides
complete insurance to the risk averse agent against the risk coming from the
uncertainty of the state of the world.
On the basis of the above observations, in the case of a simple economy with a
single good, two agents and two possible states of the world, we can conclude that
in correspondence of a Pareto optimal allocation all idiosyncratic risk is diversified
by mutual insurance among the agents (mutuality principle, see Wilson [1661]).
If there is no aggregate risk, then the agents will not bear any idiosyncratic risk
because the latter will be completely diversified. Agents will bear a risk only in
the case of aggregate risk. It should be noted, however, that this mutuality principle
does not necessarily hold if agents have heterogeneous beliefs. Moreover, the results
of this section rely on the assumption that the agents’ utility functions are state
independent.
The present analysis can be extended to an economy with I 2 agents and
S 2 possible states of the world. When there is a single good and agents have
homogeneous beliefs, condition (4.1) can be equivalently rewritten as follows (see
Exercise 4.2): for an interior ex-ante Pareto optimal allocation fxi I i D 1; : : : ; Ig
there exists a vector of strictly positive constants .1 ; : : : ; I / such that
0 0
i ui .xi
s / D j u .xs /;
j j
for all i; j D 1; : : : ; I and s D 1; : : : ; S: (4.4)
Condition (4.4) is often called the Borch condition, see Borch [269].
In the present setting, by relying on condition (4.4), we are able to give a general
and easy proof of the mutuality principle in the presence of a single good. Let
us first establish the co-monotonicity property of Pareto optimal allocations in the
presence of a single consumption good. We defineP the aggregate endowment in
correspondence of the state of the world !s as es WD IiD1 eis , for each s D 1; : : : ; S
(for simplicity, we assume that the aggregate endowment is strictly positive).
Theorem 4.3 Let .x1 ; : : : ; xI / 2 RIS CC be an ex-ante Pareto optimal allocation.
Then, for any s; r 2 f1; : : : ; Sg, the following properties are equivalent:
s xr for some i 2 f1; : : : ; Ig;

(i) xi i
(ii) xs xi
i
r for all i 2 f1; : : : ; Ig;
(iii) es er .
Proof .i/ ) .ii/: suppose that xi s xr for some i 20 f1; : : : ; Ig.0 Then, due to
i
the concavity of the utility function u , it holds that ui .xi

i
s / u .xr /. Due to
i i
j0 j j0 j
condition (4.4), this implies that u .xs / u .xr / for all j 2 f1; : : : ; Ig. By the
concavity of the utility functions u j , for all j 2 f1; : : : ; Ig, this implies property .ii/.
.ii/ ) .iii/: suppose that xi s xi r for all i 2 f1; : : : ; Ig. Then, since the
optimal allocation fx I i D 1;P
i
: : : ; Ig is feasible
PI and the utility functions are strictly
increasing, it holds that es D IiD1 xi s iD1 rx i
D er .
.iii/ ) .i/: suppose that xis < x i
r holds for all i 2 f1; : : : ; Ig. Then, similarly as in
the previous step, this would imply that es < er , thus yielding a contradiction. t
u
The above proposition illustrates the co-monotonicity property of Pareto optimal
allocations. Indeed, if a given agent i 2 f1; : : : ; Ig consumes more in the state of the
world !s than in the state of the world !r , then any other agent will also consume
more in !s than in !r . Furthermore, since the aggregate consumption equals the
aggregate endowment, this is also equivalent to es er . In the case of two possible
states of the world, two agents and a single good, the co-monotonicity property
corresponds in the Edgeworth box to the fact that the contract curve lies between
the certainty lines of the agents, see Fig. 4.2.
The result of Theorem 4.3 is of particular interest since it implies that agents do
not bear risk (i.e., they are fully insured, in the sense that their optimal consumption
plan is constant across all possible states of the world) if there is no aggregate risk
(i.e., if the aggregate endowment is independent of the state of the world). As a
matter of fact, as a consequence of Theorem 4.3, if the aggregate endowment is state
independent for a subset ˝N ˝ D f!1 ; : : : ; !S g of states of the world (i.e., there
is no aggregate risk on ˝) N then, in correspondence of an interior ex-ante Pareto
optimal allocation, the optimal consumption plan of each agent is constant over
all ! 2 ˝. N Summing up, in correspondence of a Pareto optimal allocation only
aggregate risk matters (mutuality principle).
Theorem 4.3 has another fundamental implication: in correspondence of a Pareto
optimal allocation, the optimal consumption of any agent i in the state of the world
!s does not depend on the agent’s individual endowment but only on the aggregate
endowment in the state !s , for all s D 1; : : : ; S. In other words, there is a one-to-one
correspondence between the aggregate endowment in state !s and the Pareto optimal
consumption plan of any agent in correspondence of !s . Since utility functions
are assumed to be state independent, this correspondence is also state independent.
Therefore, given a Pareto optimal allocation .x1 ; : : : ; xI / we can without loss of
generality assume that, for every i D 1; : : : ; I, there exists an increasing function
yi W RC ! RC such that xi s D y .es /, for all s D 1; : : : ; S. Such functions are called
i
sharing rules (see later in this section for more details and see also Gollier [800,
Proposition 79]).
Condition (4.4) confirms that, in correspondence of a Pareto optimal allocation, a
risk neutral agent will provide complete insurance to every risk averse agent. Indeed,
if agent i is risk neutral (meaning that the utility function ui is linear), it holds that
0
j uj .xj
s / D i ; for all j D 1; : : : ; I with j ¤ i and s D 1; : : : ; S;
thus implying that xj s is constant across all possible states of the world, for all
j D 1; : : : ; I with j ¤ i.
Let us now consider an economy with a single good (wealth) and consumption
at the two dates t D 0 and t D 1. Similarly as before, agent i (for i D 1; : : : ; I)
is represented by the couple .ei ; ui /, where the utility function ui W R2C ! R is
strictly increasing and concave and ei 2 RSC1 C represents the agent’s endowment
in terms of wealth at t D 0 and contingent wealth at t D 1. A consumption
plan of agent i is described by a non-negative .S C 1/-dimensional vector xi D
.xi0 ; xi1 ; : : : ; xiS /> , where xi0 denotes consumption at the initial date t D 0 and xis
denotes consumption at date t D 1 in correspondence of the state of the world !s ,
for s D 1; : : : ; S. In this context, in order to characterize Pareto optimal allocations,
we rely on Proposition 1.6, which can also be established in a risky setting. We
assume that agents have homogeneous beliefs. Given a vector of strictly positive
weights .a1 ; : : : ; aI ), the maximization problem determining an interior (ex-ante)
Pareto optimal allocation is given by
X
I X
S
max ai s ui .xi0 ; xis /;
SC1
.xi0 ;xi1 ;:::;xiS /2RC iD1 sD1
iD1;:::;I
subject to the feasibility constraint
X
I X
I
xis eis ; for all s D 0; 1; : : : ; S:
iD1 iD1
Note that, since we always assume that utility functions are strictly increasing, the
feasibility constraint is binding and can be expressed as an equality constraint.
The above problem can be dealt with by the Lagrange multiplier method
introducing a vector .0 ; 1 ; : : : ; S / 2 RSC1 . Necessary and sufficient conditions
(due to the strict concavity of the utility functions) for a strictly positive (ex-ante)
Pareto optimal allocation fxi 2 RSC1 CC I i D 1; : : : ; Ig are given by the complete
allocation of resources and by
X
S
ai s uix0 .xi
0 ; x s / D 0 ;
i
for all i D 1; : : : ; I; (4.5)
sD1
ai s uixs .xi
0 ; x s / D s ;
i
for all i D 1; : : : ; I and s D 1; : : : ; S: (4.6)
where ai > 0, for all i D 1; : : : ; I, and uix0 ./ and uixs ./ denote the derivatives of the
function ui ./ with respect to its first and second arguments, respectively. The above
conditions can be rewritten in terms of marginal rates of substitution as follows:
s uixs .xi
0 ; xs /
i
s
PS D ; for all s D 1; : : : ; S and i D 1; : : : ; I; (4.7)
rD1 r ux0 .x0 ; xr /
i i i 0
s uixs .xi
0 ; xs /
i
s
D ; for all s; r D 1; : : : ; S and i D 1; : : : ; I: (4.8)
r uixr .xi
0 ; xr /
i r
Condition (4.8) establishes that for any couple of states of the world the marginal
rate of substitution between wealth in correspondence of the two states is the same
for all the agents. Condition (4.7) establishes the same result for the marginal rate
of substitution between wealth in any state of the world in t D 1 and wealth in
t D 0. Observe that the marginal utility at t D 0 is affected by consumption at
t D 1. Condition (4.8) corresponds to the ex-ante Pareto optimality condition (4.1)
with L D 1. As a consequence, also in this setting with consumption at both dates
t 2 f0; 1g, we can prove the analogue of Theorem 4.3 showing that the mutuality
principle and the co-monotonicity property of a Pareto optimal allocation hold.
As a special case, if agents are characterized by time additive and state
independent utility functions of the form
ui .x0 ; xs / D ui0 .x0 / C ıi ui1 .xs /; for all i D 1; : : : ; I;
where 0 < ıi 1 represents the discount factor of agent i, then conditions (4.7)–
(4.8) characterizing an ex-ante Pareto optimal allocation reduce to
0
s ıi ui1 .xi
s / s
0 D ; for all s D 1; : : : ; S and i D 1; : : : ; I; (4.9)
ui0 .xi
0 /
0
0
s ui1 .xi
s / s
i0 i
D ; for all s; r D 1; : : : ; S and i D 1; : : : ; I: (4.10)
r u1 .xr / r
As discussed in case of consumption only at date t D 1, if agents have homogeneous

beliefs, time additive and state independent utility functions, then Theorem 4.3
implies that each Pareto optimal allocation is associated with a Pareto optimal
sharing rule, see Wilson [1661], Rubinstein [1485], Breeden & Litzenberger
[287], Constantinides [488]. A Pareto optimal sharing rule is a family of functions
fyi W RC ! RC I i D 1; : : : ; Ig, depending only on the aggregate endowment and
not on the initial allocation of resources nor on the state of the world, such that
s D y .es /, for all i D 1; : : : ; I and s D 0; 1; : : : ; S. In other words, a Pareto optimal
xi i
sharing rule describes a Pareto optimal allocation as a function of the aggregate

endowment. The next proposition provides a more precise result on Pareto optimal
sharing rules fyi W RC ! RC I i D 1; : : : ; Ig, showing in particular how marginal
increases in the aggregate endowment are shared among the agents.
Proposition 4.4 Consider an economy populated by I agents with homogeneous

beliefsPand time additive and state independent utility functions ui , i D 1; : : : ; I. Let
es D IiD1 eis be the aggregate endowment, for s D 0; 1; : : : ; S. The optimal sharing
rule fyi W RC ! RC I i D 1; : : : ; Ig associated to a Pareto optimal allocation
fxi I i D 1; : : : ; Ig satisfies

i0 dyi .es / tui yi .es /
y .es / WD D PI ; (4.11)
jD1 tu j y .es /
des j
for all i D 1; : : : ; I and s D 0; 1; : : : ; S, where tui denotes the risk tolerance

0 00
associated to the utility function ui , i.e., tui .x/ D ui .x/=ui .x/.
Proof Given a vector of strictly positive weights ai ; i D 1; : : : ; I, consider the utility
function of the representative agent defined as
!
X
I X
S
u.e0 ; e/ WD max ai ui0 .xi0 / C ıi s ui1 .xis / ;
.xi0 ;xi1 ;:::;xiS /2RSC1
C iD1 sD1
iD1;:::;I
P P
with e D .e1 ; : : : ; eS / and subject to the feasibility constraint IiD1 xis IiD1 eis ,
for all s D 0; 1; : : : ; S. The above objective function is additively separable with
respect to time and consumption in different states of the world and the feasibility
constraints are formulated separately with respect to each state of the world. Hence,
the representative agent’s utility function admits the representation
X
S
u.e0 ; e/ D u0 .e0 / C s u1 .es /;
sD1
where, for all s D 1; : : : ; S,
X
I X
I
u0 .e0 / WD max ai ui0 .xi0 / and u1 .es / WD max ai ıi ui1 .xis /:
.x10 ;:::;xI0 /2RIC iD1 .x1s ;:::;xIs /2RIC
iD1
PI i PI
iD1 x0 e0 iD1 xs es
i
By the envelope theorem, the first order conditions yield

0 0
ıi ai ui1 .xi
s / D s D u1 .es /; for all i D 1; : : : ; I and s D 1; : : : ; S; (4.12)
where s denotes the Lagrange multiplier associated to the feasibility constraint for
state s and the second equality amounts to the fact that s represents the marginal
increase of u1 as the endowment es is marginally increased. Differentiating with
respect to es we obtain
00 dxi
ıi ai ui1 .xi
s /
s
D u001 .es /; for all i D 1; : : : ; I and s D 1; : : : ; S: (4.13)
des
Due to the mutuality principle (see Theorem 4.3 and the following discussion), it
holds that xi s D y .es /, for some function y W RC ! RC , for all i D 1; : : : ; I and
i i
s D 1; : : : ; S. Hence, equation (4.13) can be rewritten as

00 0 00
ıi ai ui1 .xi
s /y .es / D u1 .es /;
i
for all i D 1; : : : ; I and s D 1; : : : ; S;
which in combination with (4.12) yields, for all i D 1; : : : ; I and s D 1; : : : ; S,

0
0 u001 .es / ui1 .xi
s /
yi .es / D : (4.14)
u01 .es / ui100 .xi
s /
PI PI 0
Recalling that iD1 yi .es / D es , so that iD1 yi .es / D 1, for all s D 1; : : : ; S, we
get
X
I XI 0
u001 .es / ui1 .xi u001 .es / X i
I

0 s /
1D yi .es / D 0 00 D 0 tui y .es / ;
iD1
u .e / u1 .xs /
iD1 1 s
i i u1 .es / iD1
so that equation (4.14) can be rewritten as

i0 tui yi .es /
y .es / D PI ;
jD1 tu j y .es /
j
0 00
for all i D 1; : : : ; I and s D 1; : : : ; S, where tui .x/ D ui .x/=ui .x/ is the risk
tolerance of agent i. A similar argument allows to prove the claim for consumption
at the initial date t D 0 (i.e., for s D 0). t
u
According to the above proposition, in correspondence of a Pareto optimal
allocation, any increment in the aggregate resources of the economy is shared among
all the agents in proportion to the individual risk tolerance. More specifically, the
marginal increase in the aggregate endowment of the economy is shared among
the agents in proportion to the ratio between their individual risk tolerance and the
average risk tolerance in the economy. Therefore, in correspondence of a Pareto
optimal allocation, a marginal increase of the aggregate endowment is allocated to
agents who are less risk averse and, therefore, to agents who are more willing to
bear risk. Note also that, due to the strict monotonicity and concavity of the utility
0
functions, the derivative yi is strictly positive. In turn, this directly implies that all
i
the functions y are increasing:
yi .es / yi .er / for all i D 1; : : : ; I ” es er ;

for all s; r 2 f1; : : : ; Sg, thus confirming once more the co-monotonicity property.
As we already mentioned, every agent consumes more in correspondence of state
!s than of state !r if and only if es er (compare with Theorem 4.3).
In general, Pareto optimal sharing rules are non-linear. However, if agents’
beliefs are homogeneous, the following proposition shows that the sharing rule is
linear if and only if the agents’ utility functions belong to the HARA class with a
common parameter b (to this effect, see also Exercise 4.5).
Proposition 4.5 Consider an economy populated by I agents with homogeneous
beliefs and time additive and state independent utility functions ui , i D 1; : : : ; I. A
sharing rule fyi W RC ! RC I i D 1; : : : ; Ig is linear in correspondence of every
Pareto optimal allocation if and only if all the agents have utility functions with
linear risk tolerance with a common slope b 0, i.e., if and only if
tui .x/ D ai C bx; for all i D 1; : : : ; I; (4.15)
for some .a1 ; : : : ; aI / 2 RIC .

Proof We prove the sufficiency part, referring the reader to Exercise 4.4 for the
necessity part (see also Wilson [1661] and Amershi & Stoeckenius [56]). In the
case of linear risk tolerance, condition (4.11) can be rewritten as
0 ai C byi .es /
yi .es / D PI ; for all i D 1; : : : ; I and s D 0; 1; : : : ; S;
jD1 a C bes
j
P
using the fact that IjD1 yj .es / D es . It can be checked that, for a given vector
1 P
y .0/; : : : ; yI .0/ such that IiD1 yi .0/ D 0, the solution to the above differential
equation is given by
ai C byi .0/
yi .es / D yi .0/ C PI es ; for all i D 1; : : : ; I and s D 0; 1; : : : ; S:
j
jD1 a
Indeed, if the sharing rule admits this representation, then

PI !
i0 ai C byi .0/ ai C byi .0/ jD1 aj b
y .es / D PI D PI PI C PI es
jD1 a
j
jD1 a
j
jD1 aj C bes jD1 aj C bes
ai C byi .0/ b ai C byi .0/ ai C byi .es /
D PI C PI P I
es D P I :
jD1 a C bes jD1 a C bes jD1 a C bes
j j j j
jD1 a
t
u
In view of the above proposition, a Pareto optimal sharing rule is linear if and
only if the agents’ utility functions belong to the HARA class with a common
cautiousness coefficient (b). Recall that the generalized power, logarithmic and
exponential utility functions all belong to the HARA class (see Sect. 2.2 and
Exercise 2.10). In particular, if the utility functions belong to the CARA class, so
that tui .x/ D ai , then the sharing rule is linear and is given by
ai
yi .es / D yi .0/ C PI es ; for all i D 1; : : : ; I and s D 0; 1; : : : ; S:
jD1 aj
Similarly, if the utility functions are of the CRRA type (i.e., tui .x/ D bx), then
0
Proposition 4.4 implies that yi .es / D yi .es /=es , so that
yi .es / D ki es ; for all i D 1; : : : ; I and s D 0; 1; : : : ; S;

P
where the vector .k1 ; : : : ; kI / satisfies IiD1 ki D 1.
The result of Proposition 4.5 is important. Indeed, in the case of agents with
utility functions characterized by linear risk tolerance with the same cautiousness
coefficient, any Pareto optimal allocation can be represented as an affine function
of the aggregate endowment. In particular, this feature implies that a Pareto optimal
allocation can be achieved by providing to the agents claims on a risk free asset and
on the aggregate endowment. In this context, a two mutual funds separation result
holds true: Pareto optimal allocations lie in the span of the risk free asset and of the
aggregate endowment (we will return on this point in Sect. 4.3).
4.2 Asset Markets and Equilibria
In this section, we shall be concerned with the relation between market equilibrium
and Pareto optimality, as discussed in Sect. 1.3 in the case of a riskless environment,
by making some assumptions on the structure of the economy and, in particular,
on the existence of spot and future markets for contingent goods. For the moment,
we consider a two-period economy, with L 1 consumption goods and where
consumption is only allowed at t D 1. Agents have the possibility of trading at the
initial date t D 0 in future markets (markets open at t D 0 for goods delivered at
the future date t D 1) and also at t D 1 in spot markets (markets open at t D 1
for immediate delivery). We can consider three different types of assets. We call
real asset an asset delivering at t D 1 a bundle of consumption goods, while we
call numéraire asset an asset delivering a single good which directly enters the
agents’ utility functions. Finally, we call nominal asset an asset delivering a good
which does not enter directly the agents’ utility function (in that sense, money could
be considered as a nominal asset). In the following, we shall mostly consider an
economy with real or numéraire assets.
In the presence of spot and future markets, as will become clear from the
following analysis, a crucial role is played by the notion of market completeness.
In a nutshell, we say that markets are complete if every contingent consumption
plan at t D 1 can be replicated (or spanned/reached) via a suitable portfolio
4.2 Asset Markets and Equilibria 139
composed of the available assets and a sufficient initial endowment, in the sense that
the payoff delivered by such a replicating portfolio equals the consumption plan in
correspondence of every possible state of the world .!1 ; : : : ; !S / at t D 1. As will
be shown below, a fundamental consequence of market completeness is that every
equilibrium allocation is Pareto optimal.
Arrow-Debreu Equilibrium
Let us introduce a first notion of equilibrium, which represents the most natural
extension of the concept of equilibrium introduced in Sect. 1.3 under certainty.
Following Debreu [535], general equilibrium theory can be extended to a risky
environment by assuming that at the initial date t D 0 there are L S financial
markets, namely one market for each of the L consumption goods in correspondence
of each of the S possible states of the world. This means that agents have the
possibility of investing in L S contingent goods, where each contingent good .l; s/
delivers one unit of the l-th consumption good in correspondence of the state of
the world !s and zero otherwise, for l D 1; : : : ; L and s D 1; : : : ; S. In particular,
these L S contingent goods are real assets, since each asset delivers a specific
consumption good contingent on the state of the world. We denote by q 2 RLS the
matrix of the prices of the L S contingent goods, where qls 2 R denotes the price
at date t D 0 of the contingent good delivering at date t D 1 one unit of the l-th
good only in correspondence of the state of the world !s . Note that the LS markets
can be considered future markets, since they are open at the initial date t D 0 for
delivery at the future date t D 1. It is important to observe that, in the presence of
L S markets, the economy is complete by construction. In fact, in order to replicate
an arbitrary consumption plan c 2 RLS , it suffices to invest at time t D 0 in a
suitable combination of the L S basic assets.
Formally, this economy with L S assets is not too different from the economy
considered in Chap. 1, since there is a financial market for every possible contingent
good .l; s/, for all l D 1; : : : ; L and s D 1; : : : ; S. Therefore, we can easily adapt to
the present setting the general equilibrium theory introduced in Sect. 1.3. Given the
matrix of prices q 2 RLS , the optimal consumption problem of agent i 2 f1; : : : ; Ig,
characterized by a utility function U i W RLS ! R and endowed with a basket of
contingent goods ei 2 RLS , becomes
max U i .x/; (PO1)

x2RLS
X
S X
L X
S X
L
qls xls qls eils : (4.16)
sD1 lD1 sD1 lD1
In the present setting, we can formulate the following definition of equilibrium.

Definition 4.6 The price-allocation couple .q I x1 ; : : : ; xI /, with q 2 RLS and
xi 2 RLS , for all i D 1; : : : ; I, constitutes an Arrow-Debreu equilibrium if
(i) for every i D 1; : : : ; I, given the matrix of prices q 2 RLS , the consumption
plan xi 2 RP
P
LS
is a solution of the optimum problem (PO1) of agent i;
I I
iD1 xls iD1 els , for all l D 1; : : : ; L and s D 1; : : : ; S.
i i
(ii)
Of course, if agents’ preferences are strictly increasing, then the inequality
in part (ii) of the above definition can be replaced by an equality. Concerning
the existence of an Arrow-Debreu equilibrium, Theorem 1.2 can be applied to
the present context. If the agents’ utility functions satisfy appropriate hypothe-
ses (strict monotonicity,
PS strict concavity
PI and continuity) and are of the form
U i .x/ D sD1 s u i
.x s / and iD1 e i
2 RCC
LS
, then the equilibrium prices

are strictly positive (q 2 RCC ) and the strictly positive equilibrium alloca-
LS
tion .x1 ; : : : ; xI / can be characterized by the following condition, for all i D
1; : : : ; I:
Uxi ls .xi / s uixl .xi

s / qls
D D ; for all s; r D 1; : : : ; S and k; l D 1; : : : ; L;
Uxi kr .xi / r uixk .xi
r / qkr
where uixl denotes the first derivative of the utility function ui with respect to its l-th
argument, for i D 1; : : : ; I and l D 1; : : : ; L.
In particular, in view of condition (4.1), this implies that an equilibrium
allocation (in the sense of Definition 4.6) is ex-ante Pareto optimal, thus
providing a version of the First Welfare Theorem in the context of Arrow-
Debreu equilibria. If the above condition is satisfied, we say that the
matrix of prices q 2 RCC LS
implements the Pareto optimal allocation
.x ; : : : ; x / starting from the initial allocation .e1 ; : : : ; eI /. Moreover, it can
1 I
be easily verified that also the Second Welfare Theorem (see Theorem 1.5)
can be extended to a risky setting in terms of Arrow-Debreu equilib-
ria.
The market structure adopted in the context of Arrow-Debreu equilibria allows
for trade only at the initial date t D 0, before the resolution of uncertainty, with no
trading taking place at t D 1. Note that, since ex-ante Pareto optimal allocations are
also ex-post Pareto optimal, if an Arrow-Debreu equilibrium allocation is reached
at the initial date t D 0 and the possibility of trading at the future date t D 1 is
then introduced, then there will be no incentive to trade at t D 1. Indeed, if there
were incentives to trade at t D 1, after the state of the world is revealed, then
this would contradict the existence of an Arrow-Debreu equilibrium at the initial
date t D 0 (see Exercise 4.6). Trading could occur at t D 1 only in the case of
multiple equilibria if agents do not coordinate among themselves on the reached
equilibrium.
Radner Equilibrium
As we have seen, the notion of Arrow-Debreu equilibrium allows for a straight-

forward application of classical equilibrium theory to a risky economy. However,
Pareto optimality is reached at equilibrium under the assumption of the existence of
L S future markets open at time t D 0, with all the trading activity taking place
simultaneously before the resolution of uncertainty. Clearly, this setting is hardly
realistic. However, it is possible to reach Pareto optimal allocations in equilibrium
by considering a smaller number of open markets, assuming that there are both
future markets open at t D 0 and spot markets open at t D 1. In particular, when
some of the L S contingent goods are not available for trade in future markets open
at t D 0, spot markets open at t D 1 play an important role, since trading after the
resolution of the uncertainty can allow the agents to reach Pareto optimality. This
idea goes back to Arrow [76].
Let us consider the case of N future markets open at t D 0 for real assets. Asset
n (for n D 1; : : : ; N) delivers at t D 1 a basket of contingent goods aQ n described by
2 3
an11 an12 ::: an1S
6 an an22 ::: an2S 7
6 21 7
aQ n D 6 : :: :: :: 7 2 R :
LS
4 :: : : : 5
anL1 : : : : : : anLS
For n D 1; : : : ; N, l D 1; : : : ; L and s D 1; : : : ; S, the quantity anls represents the

amount of good l delivered by asset n at time t D 1 in correspondence of the state
of the world !s . Equivalently, we can think of aQ n as an RL -valued random variable
with S possible realizations. We denote by p 2 RN the vector of the prices of the
N real assets. For the sake of convenience, we assume that the aggregate supply of
all N assets is normalized to zero. Besides the N future markets open at time t D 0
for real assets, we assume that at t D 1 (i.e., after the state of the world is revealed)
there are L spot markets, one for each of the L goods. Summing up, in the economy
there are N C L markets. We also denote by q 2 RLS the matrix collecting the
prices of the L goods in the spot markets open at t D 1, so that the column qs of the
matrix q represents the prices of the L goods in correspondence of the state !s .
Similarly as above, equilibrium analysis proceeds by first determining the
individual behavior of the agents, taking prices as given, and then by aggregating
the agents’ decisions, requiring that equilibrium prices make agents’ decisions
compatible among themselves. In the present setting, since trading is allowed to
take place at both dates t D 0 and t D 1, agents have to take decisions at the initial
date t D 0 without knowing the spot prices prevailing at the future date t D 1,
since the latter will depend on the state of the world (which will only be revealed at
t D 1). As a consequence, in order to solve their decision problem, agents have to
formulate expectations/forecasts about future spot prices.
Agents’ forecasts are crucial to determine first the agents’ individual behavior
and, in a second step, the market equilibrium. Moreover, an agent cannot make
forecasts on future prices without making some hypotheses on the information
available to the other agents, on their behavior and on their forecasts as well (the
so-called forecast the forecasts of the others phenomenon). We assume that all
agents formulate rational expectations and that this fact is common knowledge
among all the agents. Therefore, every agent knows the economic model, fully
exploits all the available information and knows that all the other agents will do
the same (this corresponds to the informational consistency requirement mentioned
in the introduction to the present chapter). In this context, unlike in the economy
under certainty described in Chap. 1, this hypothesis turns out to be necessary
to characterize agents’ decisions and to establish their compatibility in a risky
environment under perfect competition.
If the agents’ preferences and (homogeneous) beliefs, the initial endowments, the
probability space and the matrix aQ n , for all n D 1; : : : ; N, are common knowledge
among the agents, then, under the rational expectations hypothesis and thanks to
the consistency requirements, the definition of spot market equilibrium prices in
correspondence of each state of the world .!1 ; : : : ; !S / reduces to a deterministic
problem. From the point of view of time t D 0, spot market equilibrium prices are
represented by random variables. This type of equilibrium is known as the perfect
foresight, rational expectations equilibrium (or Radner equilibrium, see Radner
[1437]). In correspondence of an equilibrium, agents have common expectations
and correctly anticipate future prices: market uncertainty completely vanishes and is
reduced to uncertainty about the state of the world realized at t D 1. In this context,
the rational expectations hypothesis allows us to determine the agents’ behavior and
the equilibrium prices and, therefore, the assumptions on the agents’ rationality play
a more relevant role than under certainty. Without such hypotheses, one would not
be able to define an agent’s optimal behavior and to establish the compatibility of
agents’ decisions through a price system. Any other assumption on agents’ forecasts
does not lead to perfect foresight, meaning that forecasts will be biased, inducing
agents to learn over time. In this sense, a perfect foresight equilibrium represents the
conclusion of a learning process: agents have nothing more to learn about the world.
Coherently with the interpretation of an equilibrium status provided in Chap. 1, a
rational expectations equilibrium can be thought of as a rest point in the agents’
learning process.
On the basis of the above discussion, assuming rational expectations, the optimal
consumption problem of agent i, for i D 1; : : : ; I, can be written as
max U i .x/; (PO2)

x2RLS
subject to the two following budget constraints (at time t D 0 and at time t D 1,
respectively), for some z 2 RN :
!
X
N
>
p z0 and q>
s xs q>
s ans zn C eis ; for all s D 1; : : : ; S;
nD1
(4.17)
where the vector p 2 RN denotes the prices of the N assets available for trade at the
initial date t D 0, z 2 RN denotes the vector of trades in the N future markets, ans
denotes the s-th column of the matrix aQ n , qs is the s-th column of the matrix q and
represents the prices of the L goods in state !s forecasted at t D 0 and, as usual,
eis 2 RL denotes the endowment of agent i in state !s . In the above problem, it is
implicitly assumed that the initial endowment in the N future assets is null.
Problem (PO2) actually consists of two problems: a consumption problem and
an investment problem. The consumption problem consists in choosing, for each
possible state of the world .!1 ; : : : ; !S /, the consumption plan at t D 1 given the
spot prices forecasted at t D 0, while the investment problem consists in choosing
the vector z of trades in the N future assets available at time t D 0. In particular,
the investment problem amounts to ensure that at time t D 1 the agent is endowed
with the wealth needed to finance his consumption plan in each state of the world.
Note also that the vector z is allowed to take values in RN , meaning in particular that
short sales are allowed (i.e., the components of z can take negative values).
We are now in a position to formulate the following definition.
Definition 4.7 The tuple f p ; q I .z1 ; x1 /; : : : ; .zI ; xI /g, with p 2 RN , q 2
RLS , xi 2 RLS and zi 2 RN , for all i D 1; : : : ; I, constitutes a Radner
equilibrium if
(i) for every i D 1; : : : ; I, given . p ; q / 2 RN RLS , the trade vector -
consumption plan .zi ; xi / is a solution of Problem (PO2) for agent i;
P I
(ii)
PiD1 n 0,
zi
PI
for all n D 1; : : : ; N;
I
iD1 xls iD1 els , for all l D 1; : : : ; L and s D 1; : : : ; S.
i i
(iii)
Note that, in the above definition, the agents’ expectations made at the initial date
t D 0 on the spot prices prevailing at date t D 1 are self-fulfilling, in the sense that
they will clear all the spot markets for every possible state of the world once the
state of the world is revealed at date t D 1.
The Case of S Arrow Securities (Complete Market)
Let us now specialize the structure of the economy. We assume that, at the initial
date t D 0, there exist S future markets for S assets (i.e., N D S), with the s-th asset
delivering one unit of the first good in correspondence of the state of the world !s
and zero otherwise, for each s D 1; : : : ; S. In other words, in the economy there are
S numéraire Arrow securities available for trade at date t D 0 (see also Arrow [76]).
In this context, one can prove the existence of a Radner equilibrium and also the
Pareto optimality of the equilibrium allocation.
If S future markets are open at time t D 0 for the S Arrow securities, then it
can be easily verified that every consumption plan in t D 1 can be obtained by a
suitable portfolio of the S Arrow securities. Indeed, given the forecasted spot prices
q, a bundle of goods c 2 RLS can be simply obtained by holding a sufficient
amount of the first good in each state of the world .!1 ; : : : ; !S / and then trading
in the spot markets open at time t D 1. The ex-ante Pareto optimality of a Radner
equilibrium is proved in the following proposition, which also establishes a one-to-
one correspondence between Arrow-Debreu equilibria and Radner equilibria in an
economy with S markets for Arrow securities open at t D 0 (compare also with
Mas-Colell et al. [1310, Proposition 19.D.1]).
Proposition 4.8 Given an economy .ei ; U i /, with ei 2 RC
LS
and U i W RC
LS
! R,
for all i D 1; : : : ; I, the following hold:
(i) if the tuple .q I x1 ; : : : ; xI /, with q 2 RCC LS
, is an Arrow-Debreu equilibrium,
then there exist a vector of prices of the Arrow securities p 2 RSCC and a set of
trade vectors .z1 ; : : : ; zI / 2 RSI such that f p ; q I .z1 ; x1 /; : : : ; .zI ; xI /g
is a Radner equilibrium;
(ii) if the tuple f p ; q I .z1 ; x1 /; : : : ; .zI ; xI /g, with p 2 RSCC and q 2 RCC LS
,
is a Radner equilibrium, then there exists a vector 2 RCC such that S
.1 q1 ; : : : ; S qS I x1 ; : : : ; xI / is an Arrow-Debreu equilibrium.

Proof .i/: let ps WD q1s , for all s D 1; : : : ; S. For each i D 1; : : : ; I, if xi 2 RLS
satisfies the budget constraint (4.16), then there exists a vector zi 2 RN such that the
budget constraint (4.17) is satisfied (recall that N D S). Indeed, let
1 > i
zis WD q .xs eis /; for all s D 1; : : : ; S:
q1s s
By (4.16), it holds that
X
S X
S
ps zis D q>
s .xs es / 0;
i i
sD1 sD1
so that the first requirement of the budget constraint (4.17) is met. Furthermore, we
also have q> i i i
s .xs es / D q1s zs , for all s D 1; : : : ; S and, therefore, x also satisfies
i
the second requirement of the budget constraint (4.17), recalling that at time t D 0
the available assets are the S Arrow securities for the first good. Conversely, for all
i D 1; : : : ; I, let xi 2 RLS and .zi1 ; : : : ; ziS / be such that the budget constraint (4.17)
P
is satisfied, so that SsD1 ps zis 0 and q> i
s .xs es / ps zs , recalling that ps D q1s
i i
(for all s D 1; : : : ; S). By summing the constraints over s we obtain
X
S X
S
q>
s .xs es /
i i
ps zis 0:
sD1 sD1
Therefore, xi satisfies the budget constraint (4.16), for all i D 1; : : : ; I. We have

thus shown that, in correspondence of prices p D .q11 ; : : : ; q1S /, the budget sets
associated to an Arrow-Debreu equilibrium and to a Radner equilibrium coincide.
This implies that the allocation .x1 ; : : : ; xI / of an Arrow-Debreu equilibrium
implemented by prices q 2 RCC LS
is also an allocation of a Radner equilibrium
in correspondence of Arrow security prices p 2 RSCC , spot prices q 2 RCC LS
and of the vector of trades on future markets z D .z1 ; : : : ; zS / 2 R defined

i i i S
1 > i
s WD q qs .xs es /, for i D 1; : : : ; I. Indeed, .x ; z / 2 R
by zi RS
i i i LS
1s
solves
PIproblem (PO2)
PI for agent i given the price vectors . p ; q / and it holds
1 > i
that iD1 zs D iD1 q qs .xs eis / 0, for all s D 1; : : : ; S, while the third
i
1s
requirement in Definition 4.7 comes directly from Definition 4.6.
.ii/: set s > 0 so that s q1s D ps , for all s D 1; : : : ; S. By relying on arguments
analogous to those used in the first part of the proof, it can be easily verified that
the elements xi 2 RLS satisfying the budget constraint (4.17) for a vector of
trades z 2 RS also satisfy (4.16) with respect to prices .1 q1 ; : : : ; S qS / 2 RCC LS
.
Therefore, x 2 R i LS
solves problem (PO1) for agent i in correspondence of prices
.1 q1 ; : : : ; S qS /, for every i D 1; : : : ; I, and contingent goods markets clear, as
a consequence of part (iii) of Definition 4.7. By Definition 4.6, this implies that
.1 q1 ; : : : ; S qS I x1 ; : : : ; xI / is an Arrow-Debreu equilibrium. t
u
As can be seen from the proof of the above proposition, the quantity s D ps =q1s
represents the price at date t D 0 of one unit of wealth obtained at time t D 1
in correspondence of the state of the world !s , for all s D 1; : : : ; S. The strict
monotonicity of the agents’ utility functions and the fact that all the states of the
world have a strictly positive probability of occurrence imply that the equilibrium
prices of the S Arrow securities are strictly positive.
Proposition 4.8 shows that the set of Arrow-Debreu equilibria of an economy
with L S goods, as considered at the beginning of the present section, coincides
with the set of Radner equilibria obtained with S Arrow securities available for
trade at time t D 0. This result has two important consequences: first, the Radner
equilibrium allocation is ex-ante Pareto optimal (and, hence, also ex-post Pareto
optimal); second, in order to establish the existence of a Radner equilibrium with S
Arrow securities, we can rely on classical equilibrium analysis to prove the existence
of an Arrow-Debreu equilibrium.
A remarkable feature of the above result is represented by the fact that the
presence of spot markets open at t D 1 allows for a reduction in the number of
markets needed to achieve Pareto optimality in equilibrium. Indeed, when S Arrow
securities are available for trade at time t D 0, only L C S markets are needed to
reach a Pareto optimal allocation through a Radner equilibrium instead of L S
markets as in the case of an Arrow-Debreu equilibrium. More specifically, since
agents are allowed to trade goods in spot markets open at t D 1, at the initial date
t D 0 agents are only interested in transferring wealth among different states of the
world and this can be done through S assets each delivering one unit of a reference
good in correspondence of a specific state of the world (Arrow securities). At time
t D 1, in correspondence of the specific state of the world which will be realized,
agents will trade in the L spot markets. We want to remark that, in order to analyse
the economy, we had to introduce a rather strong hypothesis on the agents’ ability
to forecast future spot prices: agents’ expectations are rational and, therefore, self-
confirming in equilibrium.
The Case of Complex Securities
The results obtained so far on the existence and the properties of Radner equilibria
require the existence of a number of Arrow securities equal to the number S of
possible states of the world. This assumption is hardly realistic, because Arrow
securities represent theoretical assets which are not traded in real financial markets.
Indeed, assets typically traded in financial markets pay dividends in correspondence
of more than one elementary state of the world, i.e., they are complex securities.
In the remaining part of this section, we aim at extending the equilibrium analysis
presented so far to this more realistic situation. To simplify the presentation, we
restrict our attention to N numéraire assets delivering dividends expressed in terms
of the first good,2 so that the spot price of the first good at t D 1 is equal to one
across all possible states of the world. The prices of the N future assets are denoted
by p 2 RN , the dividend of the n-th asset is denoted by dQ n 2 RS , for all n D 1; : : : ; N,
and we let D 2 RSN be the matrix collecting the asset dividends:
2 3
d11 d12 ::: d1N
6d21 d22 ::: d2N 7
6 7
DD6 : :: :: :: 7 :
4 :: : : : 5
dS1 : : : : : : dSN
The matrix D encodes significant information on the structure of the financial

market. Indeed, the matrix D is associated with the linear map z 7! Dz from RN
onto RS which describes contingent consumption plans (i.e., vectors c 2 RS ) as a
result of the investment decisions taken at the initial date t D 0 in the N assets
(i.e., z 2 RN ). In other words, the map z 7! Dz describes the consumption plan
c D Dz 2 RS (at time t D 1) generated by an investment strategy z 2 RN .
It is important to observe that the set of consumption plans attainable by trading
in the N (future) markets open at time t D 0 is not necessarily the whole space RS .
The linear subspace of RS containing all consumption plans attainable by trading
in the N markets is the image (or range) I.D/ of the linear map z 7! Dz and the
dimension of the linear subspace I.D/ RS is less or equal than S. A consumption
plan (or contingent wealth at t D 1) c 2 RS is said to be reachable (or attainable) by
trading in the N assets if c 2 I.D/, i.e., if it is possible to reach (or attain/replicate)
2
Using the notation anls introduced before, this corresponds to ans D .dsn ; 0; : : : ; 0/ 2 RL , where
dsn is the dividend paid by asset n in correspondence of the state of world !s , for all n D 1; : : : ; N
and s D 1; : : : ; S.
the consumption plan c by means of a portfolio zc 2 RN of the N assets available

for trade at time t D 0, starting from a suitable endowment, so that c D Dzc . The
portfolio zc 2 RN which replicates the consumption plan c is called replicating
portfolio.
The dividend matrix D allows for an easy characterization of market complete-
ness. As already mentioned, markets are complete if every contingent consumption
plan c 2 RS at t D 1 can be reached/replicated by a suitable portfolio zc 2 RN of the
N assets available for trade at the starting date t D 0. Hence, in the present setting,
it is immediate to see that markets are complete if and only if I.D/ D RS , i.e., the
rank of the matrix D is equal to S (i.e., rank.D/ D S). In other words, markets are
complete if and only if the N traded assets allow to span the whole space RS . On
the other hand, if markets are incomplete, then agents can only reach a subset of all
possible consumption plans in RS by investing in the N available assets.
The N assets available for trade at time t D 0 are said to be non-redundant if their
dividends are linearly independent, i.e., if it is not possible to express the dividend
of any asset as a linear combination of the dividends of all the other assets. This
property is satisfied if N S and rank.D/ D N (i.e., the columns of the matrix
D are linearly independent vectors). In particular, when N D S, meaning that the
number of traded assets is equal to the number of possible states of the world, this is
equivalent to the non-singularity of the square matrix D (i.e., det.D/ ¤ 0). We also
say that the N assets enjoy the uniqueness of representation property if, for every
c 2 I.D/, there exists a unique vector zc 2 RN such that Dzc D c. The uniqueness
of representation property holds if and only if the dimension of ker.D/ is zero and,
therefore, if and only if the N assets are non-redundant.
Without loss of generality, limiting our attention to a full rank matrix D, the
following situations may arise:
• N > S: in this case, I.D/ D RS and, 8c 2 RS , dimfz 2 RN W Dz D cg D N S;
• N D S: in this case, I.D/ D RS and, 8c 2 RS , dimfz 2 RN W Dz D cg D 0;
• N < S: in this case, I.D/ RS and, 8c 2 I.D/, dimfz 2 RN W Dz D cg D 0.
In the last two cases, the N assets enjoy the uniqueness of representation property,
while in the first two cases markets are complete. A necessary condition for market
completeness is that the number of traded assets is greater or equal than the number
of possible states of the world, i.e., N S. Loosely speaking, markets are complete
when market participants can trade in a set of assets which span all possible risky
scenarios of the economy. For an attainable contingent consumption plan c 2 I.D/,
we can define its market value as follows.
Definition 4.9 Let c 2 RS be a contingent consumption plan belonging to I.D/.
The market value of c at t D 0, denoted by V.c/, is defined as the value at t D 0 of
the replicating portfolio for c, i.e.,
V.c/ WD f p> zc W zc 2 RN and Dzc D cg:

Note that, if the N assets do not enjoy the uniqueness of representation property,
0 0
then there may exist two different portfolios zc ; zc 2 RN satisfying Dzc D c D Dzc .
0
In principle, the market values associated to the two portfolios zc and zc can be
different. However, in correspondence of an equilibrium allocation, the market value
of any consumption plan c 2 I.D/ is uniquely defined (this corresponds to the so-
called Law of One Price, see Proposition 4.18 and Exercise 4.22). This is closely
connected to the absence of arbitrage opportunities. In Sect. 4.4 we shall investigate
the relation between the absence of arbitrage opportunities and the existence of an
equilibrium. In particular, this implies that Definition 4.9 is well posed.
In order to study the existence and the properties of equilibrium allocations, the
notion of market completeness plays a particularly important role. Indeed, if markets
are complete, then agents are unrestricted in their capacity to transfer wealth across
different states of the world by forming suitable portfolios of the assets available
for trade. In particular, this implies that agents can reach Arrow-Debreu equilibria,
as shown in the following proposition. Note also that, as a consequence of the First
Welfare Theorem, equilibrium allocations are Pareto optimal.
Proposition 4.10 Given an economy .ei ; U i /, with ei 2 RC LS
, and U i W RC
LS
! R,
for all i D 1; : : : ; I, with N assets traded at t D 0 such that rank.D/ D S, then the
following hold:
(i) if the tuple .q I x1 ; : : : ; xI /, with q 2 RCC LS
, constitutes an Arrow-Debreu
equilibrium, then there exist a vector of prices p 2 RNCC and a set of
trade vectors .z1 ; : : : ; zI / 2 RNI such that f p ; q I .z1 ; x1 /; : : : ; .zI ; xI /g
constitutes a Radner equilibrium;
(ii) if the tuple f p ; q I .z1 ; x1 /; : : : ; .zI ; xI g, with p 2 RNCC and q 2 RCC LS
,
constitutes a Radner equilibrium, then there exists a vector 2 RCC such that S
.1 q1 ; : : : ; S qS I x1 ; : : : ; xI / constitutes an Arrow-Debreu equilibrium.

The proof of the above proposition is similar to that of Proposition 4.8 (see Dana
& Jeanblanc [514] and Mas-Colell et al. [1310, Proposition 19.E.2]). As in the case
of Proposition 4.8, this result has two important implications: in the case of complete
markets, the Radner equilibrium allocation is ex-ante Pareto optimal and, moreover,
the existence of a Radner equilibrium can be established by relying on classical
existence results for Arrow-Debreu equilibria.
When markets are incomplete, in the sense that not every contingent consumption
plan can be spanned by the available assets (i.e., I.D/ RS ), it is not possible to
establish a general connection between Radner and Arrow-Debreu equilibria. While
the existence of a Radner equilibrium can be established under suitable conditions,
the equilibrium allocation will not necessarily be Pareto optimal. Indeed, unlike in
the situation of complete markets, in the presence of incomplete market agents have
a limited possibility of transferring wealth across different states of the world by
trading on the available assets. As a consequence, the possibilities for risk sharing
are limited.
It is worth pointing out that the allocation .x1 ; : : : ; xI / obtained in correspon-
dence of a Radner equilibrium with financial assets delivering dividends D 2 RSN
4.3 Equilibrium with Intertemporal Consumption 149
is still an equilibrium allocation if the dividends’ structure is modified in a way

0
such that the new dividend matrix D0 2 RSN satisfies I.D/ D I.D0 / (see Mas-
Colell et al. [1310, Proposition 19.E.3]). In other words, the equilibrium allocation
only depends on the linear subspace spanned by the financial assets and not on the
specific structure of the financial markets open at t D 0. As a consequence of this
property, equilibrium allocations are unaffected by the addition (or the deletion) of
redundant assets.
Note that, for L D 1, the trade vector z (together with the initial endowment
and the dividend matrix D) completely determines
PN the amount of consumption good
obtained at time t D 1, since xs D nD1 d sn zn C es , as a consequence of the
constraint (4.17) together with the standing assumption of strictly increasing utility
functions. In the case of incomplete markets, the following definition of constrained
ex-ante Pareto optimality can be formulated, see Diamond [571] and Mas-Colell
et al. [1310, Definition 19.F.1].
1
PI .z ;i : : : ; z / 2 R
I NI
Definition 4.11 The asset allocation is constrained ex-ante
Pareto optimal if it is feasible (i.e., iD1 z 0) and if there does not exist a feasible
0 0
asset allocation .z1 ; : : : ; zI / 2 RNI such that the expected utility associated with
the latter allocation is greater or equal than the expected utility associated with the
allocation fz1 ; : : : ; zI g, for all the agents, and strictly greater for at least one agent
i 2 f1; : : : ; Ig.
According to the above definition, an asset allocation is constrained ex-ante
Pareto optimal if there is no asset redistribution which induces a Pareto improve-
ment. Recalling that for L D 1 the asset allocation directly determines the
consumption plan, it can be proved that any Radner equilibrium in incomplete
markets is constrained ex-ante Pareto optimal in the sense of the above definition
(see Mas-Colell et al. [1310, Proposition 19.F.1]). Note, however, that the situation
L D 1 is rather particular, since with a single consumption good at t D 1 there are no
possibilities for trade once the state of the world is revealed, so that the consumption
plan is fully determined by the initial asset allocation. When L 2, it is not possible
to establish an analogous result.
4.3 Equilibrium with Intertemporal Consumption
In this section, we address the intertemporal consumption problem, by allowing

utility functions to depend on consumption at both dates t D 0 and t D 1.
In particular, we shall focus our attention on the relation between equilibrium
allocations and Pareto optimality and on the possibility of characterizing the
equilibrium allocation by means of a representative agent. We first consider the
case of complete financial markets and then the more complex case of incomplete
markets.
Complete Markets
Let us consider an economy with a single good (which we generically interpret as

wealth) and S possible states of the world .!1 ; : : : ; !S /. In line with the previous
section, we assume that at the initial date t D 0 agents can trade in S future
markets, each of them corresponding to one of the S basic Arrow securities. The
utility function of agent i (for each i D 1; : : : ; I) is defined with respect to wealth
consumed at the initial date t D 0, denoted by x0 , and wealth consumed at the
future date t D 1, denoted by xs , in correspondence of the s-th state of the world
!s , for each s D 1; : : : ; S. As before, the function .x0 ; xs / 7! u.x0 ; xs / is assumed to
be strictly increasing and concave. Moreover, we always assume that consumption
is non-negative. For s D 1; : : : ; S, we denote by ps the price of the s-th Arrow
security paying one unit of wealth in correspondence of the state of the world !s
and nothing in correspondence of all the other states. For i D 1; : : : ; I, we denote
by .ei0 ; ei1 ; : : : ; eiS / the endowment of agent i, with ei0 denoting the initial wealth at
time t D 0 and eis the endowment in terms of wealth at time t D 1 contingent on the
state of the world !s , for s D 1; : : : ; S. We assume that agents share homogeneous
beliefs. The optimal consumption problem of agent i can be formulated as follows:
X
S
max s ui .x0 ; xs /; (PO3)
SC1
.x0 ;x1 ;:::;xS /2RC sD1
X
S X
S
x0 C ps xs ei0 C ps eis : (4.18)
sD1 sD1
In the above formulation of the budget constraint, the price of wealth consumed at
the initial date t D 0 has been conventionally set equal to one. As usual, since the
utility functions are assumed to be strictly increasing, the budget constraint (4.18)
can be formulated as an equality.
Problem (PO3) can be solved by means of the associated Lagrangian. Letting
i be the Lagrange multiplier associated with the optimal consumption problem of
agent i, necessary and sufficient conditions (due to the strict concavity of the utility
function) for a strictly positive optimal consumption plan .xi0 ; x1 ; : : : ; xS / become
i i
X
S
s uix0 .xi
0 ; xs / D ;
i i
(4.19)
sD1
s uixs .xi
0 ; xs / D p s ;
i i
for all s D 1; : : : ; S; (4.20)
where, as before, uix0 and uixs denote the derivatives of the utility function ui with
respect to its first and second argument, respectively. Since i > 0, it holds that
s uixs .xi
0 ; xs /
i
PS D ps ; for all s D 1; : : : ; S and i D 1; : : : ; I; (4.21)
rD1 r ux0 .x0 ; xr /
i i i
in correspondence of the equilibrium price vector . p1 ; : : : ; pS / 2 RSCC . In

particular, condition (4.21) means that, for all s D 1; : : : ; S, the equilibrium price
ps of the s-th Arrow security corresponds to the ratio between the marginal utility
of consumption at time t D 1 in the state of the world !s weighted by s and
the (expected) marginal utility of consumption at the initial date t D 0. As a
consequence, the marginal rates of substitution between consumption at the initial
date t D 0 and consumption at t D 1 in correspondence of the state of the world
!s , for s D 1; : : : ; S, are equal among all the agents. By the analysis developed in
Sect. 4.1, this implies the (ex-ante) Pareto optimality of the equilibrium allocation.
This can also be seen by letting ai WD 1=i , for all i D 1; : : : ; I, and 0 WD 1 and
s WD ps , for all s D 1; : : : ; S, and comparing (4.19)–(4.20) with (4.5)–(4.6). We
have thus shown that, if at date t D 0 agents can trade in future markets where all S
basic Arrow securities are traded, then the equilibrium allocation is Pareto optimal.
In other words, the possibility of spanning all possible future states of the world
by trading in future markets open at t D 0 allows agents to reach a Pareto optimal
allocation in equilibrium.
As a special case, when the agents’ utility functions are time additive and state
independent, i.e., ui .x0 ; xs / D ui0 .xs / C ı i ui1 .xs /, for two strictly increasing and
concave utility functions ui0 and ui1 and for some discount factor ı i 2 .0; 1/,
condition (4.21) becomes
0
ı i s ui1 .xi
s /
0 D ps ; for all s D 1; : : : ; S and i D 1; : : : ; I: (4.22)
ui0 .xi
0 /
As pointed out in the previous section, Arrow securities are not traded in real
financial markets: traded securities are in general complex and deliver a dividend
described by a random variable, taking different values depending on the realization
of the state of the world. Hence, let us now assume that N complex securities are
traded at date t D 0, with security n delivering a random dividend dQ n with possible
realizations .d1n ; : : : ; dSn / 2 RS , where dsn denotes the amount of wealth delivered
by security n in the state of the world !s , for n D 1; : : : ; N and s D 1; : : : ; S.
In the presence of financial markets open at t D 0 with N complex securities, we
assume that the initial endowment of agent i (for all i D 1; : : : ; I) is given in terms
of wealth and units of the N securities. More specifically, the endowment of agent
i is represented by a vector .ei0 ; ei1 ; : : : ; eiN /, where ein , for n D 1; : : : ; N, denotes
the amount of security n initially held by agent i at t D 0 while ei0 denotes the
endowment of wealth at t D 0. The optimal consumption problem of agent i can
then be formulated as follows:
X
S X N
max s u x0 ;
i
zn dsn ; (PO3’)
x0 2RC ;.z1 ;:::;zN /2RN
sD1 nD1
X
N X
N
x0 C pn zn ei0 C pn ein ; (4.23)
nD1 nD1
where pn 2 RC denotes the price of security n at time t D 0, for n D 1; : : : ; N, and

zn is the number of units of security n traded at t D 0 by agent i, for n D 1; : : : ; N.
In other words, the vector z D .z1 ; : : : ; zN / 2 RN denotes the investment strategy of
agent i in the N available securities. Note that allowing z to belong to RN means in
particular that short sales are allowed.
As can be easily checked, the conditions characterizing the optimal consumption
choice in equilibrium are given by
X
S
s uixs .xi
0 ; xs /
i
PS dsn D pn ; for all n D 1; : : : ; N and i D 1; : : : ; I;
sD1 rD1 r u i .xi ; xi /
x0 0 r
(4.24)
where
X
N X
N
0 D e0
xi i
pn .zi
n en /
i
and s D
xi n dsn ;
zi for all s D 1; : : : ; S;
nD1 nD1
and p 2 RN denotes the vector of equilibrium prices. It is important to observe

that condition (4.24) does not have the same implications of condition (4.21),
because it does not imply that marginal rates of substitution between consumption
at t D 1 in correspondence of the different states of the world and consumption
at the initial date t D 0 are equal among all the agents. As a consequence, the
equilibrium allocation associated to Problem (PO3’) is not necessarily ex-ante
Pareto optimal. However, in the special and important case of complete financial
markets, we can prove the Pareto optimality of an equilibrium allocation, as shown
in the following proposition (in line with the result of Proposition 4.10, see also
Huang & Litzenberger [971, Section 5.7]).
Proposition 4.12 If markets are complete (i.e., if rank.D/ D S, using the notation
introduced in Sect. 4.2), then the equilibrium allocation corresponding to Prob-
lem (PO3’) is ex-ante Pareto optimal.
Proof If markets are complete, ex-ante Pareto optimality can be obtained as a

consequence of Proposition 4.10 together with the Pareto optimality of Arrow-
Debreu equilibria (First Welfare Theorem). However, we shall prove the claim
in a more direct and constructive way. The optimality conditions (4.24) defining
an equilibrium allocation in correspondence of an equilibrium can be rewritten in
matrix notation as follows, for all i D 1; : : : ; I:
2 3 2 1 uix .xi ;xi
1 /
3 2 3
d11 d21 ::: dS1 PS 1 0 p1
6 d12 6 r ux0 .x0 ;xr / 7
i i
dS2 7
i
6 d22 ::: 7 6
rD1
7 6 p2 7
6 : :: 7 6
:: 7D6 7
:
4 :: :: :: 5 6 : 7 4 :: 7
6 (4.25)
: : : 4 i i 5
S uixS .x0 ;xS /
:5
d1N : : : : : : dSN PS
rD1 r uix0 .xi
0 ;xr /
i pN
The first matrix appearing in the left-hand side of (4.25) is D> . If the matrix D has
full rank and N S, then we can build a submatrix D 2 RSS corresponding to
S linearly independent securities (i.e., S linearly independent rows of D> ). We can
then consider the following equation, where the matrix appearing on the left hand
side is now D> :
2 3 2 1 uix .xi ;xi
1 /
3 2 3
d11 d21 ::: dS1 PS 1 0 p1
6d12 7 6 rD1 r uix0 .xi
0 ;xr / 7
i
6 d22 ::: dS2 7 6 6

7 6p2 7
6 : 7 6 :: 7D6 : 7 :
4 :: :: :: ::
5 6 : 7 4:7
: : : 4 S uixS .x0 ;xS /
i i 5 :5
d1S : : : : : : dSS PS i i
rD1 r ux0 .x0 ;xr /
i pS
Multiplying by the inverse of the matrix D> (which exists since rank.D/ D S), we
obtain
2 3 2 31 2 3 2 3
1 uix1 .xi0 ;x1 /
i
PS d11 d21 : : : dS1 p1 m1
6 rD1 r ux0 .x0 ;xr / 7 6
i i i
7 6 7 6 7
6 7 d d : : : d p m
6 :: 7D6
6
12 22 S2 7
7
6 27
6 7 DW
6 27
6 : 7:
6 : 7 4 :: : : : : :: 5 :
4 :: 5 4 :: 5
4 S uixS .x0 ;xS /
i i 5 : : : :
PS
rD1 r ux0 .x0 ;xr /
i i i d1S : : : : : : dSS pS mS
In particular, observe that the right hand side of the last relation does not depend on
i, so that
s uixs .xi
0 ; xs /
i
PS D ms ; for all s D 1; : : : ; S and i D 1; : : : ; I: (4.26)
u
rD1 r x0 0
i .xi ; xi /
r
We have thus shown that, in correspondence of an equilibrium allocation, marginal

rates of substitution between consumption at t D 1 in the different states of the
world and consumption at the initial date t D 0 are equal among all agents. In
view of condition (4.7), this implies the ex-ante Pareto optimality of the equilibrium
allocation. t
u
The strict monotonicity of the utility function ui implies that the quantities
ms introduced in the proof of Proposition 4.12 are strictly positive. In particular,
the vector m 2 RSCC is the unique solution to the equation (omitting possibly
redundant assets)
D> m D p : (4.27)
In Proposition 4.12, the equilibrium of the economy has been defined starting
from the N securities paying the dividends represented by the matrix D. From the
equilibrium prices of the original N securities we can derive the implicit equilibrium
prices of other assets by means of replication arguments. As a consequence, under
the assumption of market completeness, Problems (PO3) and (PO3’) are equivalent,
up to a suitable identification of the budget constraints.
Corollary 4.13 If markets are complete (i.e., if rank.D/ D S), then the quantity ms
defined in (4.26) is the implicit equilibrium price (or market value) of the s-th Arrow
security yielding 1 at time t D 1 in correspondence of the state of the world !s and
0 otherwise, for every s D 1; : : : ; S.
Proof Since markets are complete, every contingent consumption plan c 2 RS
can be replicated by investing in the N available securities. As in the proof of
Proposition 4.12, let us define by D the submatrix composed of the linearly
independent columns of D. Let us fix an arbitrary s 2 f1; : : : ; Sg and consider the
vector 1s WD .0; : : : ; 0; 1; 0; : : : ; 0/ 2 RS , where the 1 element is in the s-th position,
representing the payoff of the s-th Arrow security. The payoff 1s can be replicated by
investing in a portfolio z.s/ 2 RS such that D z.s/ D 1s . Multiplying by the inverse
matrix D1 , we get z.s/ D D1 1s . Hence, the implicit equilibrium price of the s-th
Arrow security is given by the market value (see Definition 4.9) of the replicating
portfolio z.s/, thus yielding p> z.s/ D p> D1 1s D ms . t
u
In view of the above corollary, the quantity ms associated with the state of the
world !s is also called state price. Note that, if markets are complete and an asset
paying c is introduced, then its equilibrium price is m> c, as a consequence of
the above corollary. The following proposition presents two important properties of
asset prices in correspondence of an equilibrium allocation, namely that the market
value V.c/ defined in Definition 4.9 is linear. Moreover, Proposition 4.14 implies
that identical contingent consumption plans have identical market values at t D 0.
Proposition 4.14 Suppose that markets are complete (i.e., rank.D/ D S). Then, in
correspondence of an equilibrium allocation, the following hold:
(i) if c 2 RS represents a contingent consumption plan, then its market value V.c/
at t D 0 (in the sense of Definition 4.9) is equal to c> m ;
(ii) if c; cQ 2 RS represent two contingent consumption plans and cO WD c C cQ , then it
holds that
V.Oc/ D V.c/ C V.Qc/:

Proof (i): market completeness implies the existence of a portfolio zc of S linearly

independent securities such that D zc D c, i.e., zc D D1 c. Similarly as in the proof
of Corollary 4.13, the market value at time t D 0 of such a portfolio is given by
m> c.
(ii): the claim follows from part (i), since m> cO D m> .cC cQ / D m> cCm> cQ : u
t
Summing up, if N S securities are traded in the market at time t D 0 and
rank.D/ D S, then the equilibrium prices of the N assets implicitly and uniquely
define the prices of the S Arrow securities (state prices). Moreover, as shown
in (4.26), the vector of the state prices coincides with the vector of the agents’
marginal rates of substitution, thus implying the (ex-ante) Pareto optimality of the
equilibrium allocation.
Representative Agent Analysis
In the context of a complete markets economy, equilibrium prices can also be

characterized by means of a representative agent economy, in analogy to the riskless
setting considered in Proposition 1.6. In particular, equilibrium prices can be deter-
mined in terms of the no-trade equilibrium of a single representative agent endowed
with the resources of the whole economy, see Constantinides [488]. We denote by
e D .eP0 ; e1 ; : : : ; eS / the aggregate endowment of an economy with I agents, where
es WD IiD1 eis , for all s D 0; 1; : : : ; S. In an economy with a single agent endowed
with the aggregate resources of the original economy, the representative agent will
not trade in correspondence of an equilibrium and his optimal consumption plan
will be given by the aggregate endowment .e0 ; e1 ; : : : ; eS /. We will develop the
representative agent analysis in an intertemporal consumption setting, noting that
the results presented below can also be established in an economy with consumption
only at t D 1 and a single good.
Let us consider an economy populated by I agents with homogeneous beliefs,
time additive and state independent utility functions ui .x0 ; xs / D ui0 .x0 / C ıui1 .xs /,
with discount factor ı 2 .0; 1/, and with initial endowment .ei0 ; ei1 ; : : : ; eiS /, for all
i D 1; : : : ; I. Note that the following analysis is also valid when agents exhibit
different time preferences, i.e., when agents’ preferences are characterized by
different discount factors. In the present context, we construct the utility function of
a representative agent, following the construction adopted in a riskless economy (see
Proposition 1.6). Since markets are assumed to be complete, there is a unique vector
of the prices of the Arrow securities p D . p1 ; : : : ; pS / implicit in the equilibrium
prices of the original N securities (see Corollary 4.13). As in Proposition 1.6,
let ai WD 1=i , where i denotes the Lagrange multiplier associated with the
optimal consumption plan .xi 0 ; x1 ; : : : ; xS / of agent i given as the solution to
i i
Problem (PO3) in correspondence of the equilibrium price vector p of the Arrow

securities, for i D 1; : : : ; I. Similarly as in the proof of Proposition 4.4, we then
define the utility function of the representative agent as
X
S
u.e0 ; e1 ; : : : ; eS / WD u0 .e0 / C ı s u1 .es /; (4.28)
sD1
where, for all s D 1; : : : ; S,
X
I X
I
u0 .e0 / WD max ai ui0 .xi / and u1 .es / WD max ai ui1 .xi /:
.x1 ;:::;xI /2RIC .x1 ;:::;xI /2RIC
PI iD1 PI iD1
iD1 x e0 iD1 x es
i i
(4.29)
In other words, the representative agent’s utility function is constructed as a

weighted sum of the individual agents’ utility functions, with the weights being
given by the inverse of the Lagrange multipliers associated to each individual agent’s
optimal consumption problem, subject to the available resources constraints. Note
that, since utility functions are assumed to be strictly increasing, the inequality con-
straints appearing in (4.29) can be equivalently formulated as equality constraints.
Similarly as in the proof of Proposition 4.4, the first order conditions yield, for
all i D 1; : : : ; I and s D 1; : : : ; S,
X
I
0 dxi X
I
dxi
u00 .e0 / D ai ui0 .xi
0 /
0
D 0
D 1; (4.30)
iD1
de 0 iD1
de0
X
I
0 dxi p X dxi
I
p
u01 .es / D ai ui1 .xi
s /
s
D s s
D s ; (4.31)
iD1
des ıs iD1 des ıs
where we have used the optimality conditions (4.19)–(4.20) of the individual agents’
optimal consumption problems together with the fact that ai D 1=i , for all i D
1; : : : ; I. Conditions (4.30)–(4.31) together imply that
ıs u01 .es /

D ps ; for all s D 1; : : : ; S: (4.32)
u00 .e0 /
Condition (4.32) shows that the equilibrium price vector p D . p1 ; : : : ; pS / of the
original economy corresponds to the vector of the equilibrium prices of the Arrow
securities in an economy with a representative agent endowed with the resources of
the whole economy and with the utility function u defined in (4.28)–(4.29). Indeed,
by imposing that the optimal choice for the representative agent is obtained in
correspondence of the aggregate endowment .e0 ; e1 ; : : : ; eS / (no-trade equilibrium)
and by the optimality conditions (4.30)–(4.31), the marginal rate of substitution of
the representative agent between consumption in the state of the world !s at time
t D 1 and consumption at time t D 0 is equal to the Arrow security equilibrium
price ps , for all s D 1; : : : ; S. Condition (4.32) is of the same form of the optimality
condition (4.22) obtained in the original economy with I agents with time additive
and state independent preferences. Due to the assumption of complete markets, a no-
trade equilibrium of a representative agent economy can equivalently be obtained by
allowing the representative agent to trade in the N original assets instead of the S
Arrow securities.
The representative agent’s utility function defined in (4.28)–(4.29) depends on
the weights .a1 ; : : : ; aI /. Since such weights are defined in terms of the Lagrange
multipliers of Problem (PO3) of each individual agent i 2 f1; : : : ; Ig, they depend
on the initial endowments of the I agents and, hence, on the initial allocation of
resources. This means that the equilibrium price vector p in general depends on
the distribution of resources among the agents. Therefore, the equivalence between
the original economy with I agents and the representative agent’s economy holds
only in correspondence of equilibrium prices. In this sense, the representative agent
is only locally representative of the original economy.
However, the results discussed at the end of Sect. 1.3 on the existence of a
representative agent in a strong sense can be extended to the present context. Indeed,
if agents have homogeneous beliefs, the same discount factor and time additive
state independent utility functions with linear risk tolerance and with identical
cautiousness (compare with Proposition 4.5), then equilibrium prices will only
depend on the aggregate endowment, regardless of the distribution of resources
among the I agents ( aggregation property). In other words, equilibrium prices will
1
not depend on the weights .a1 ; : : : ; aI /. The generalized power ui .x/ D b1 .i C
b1
bx/ , with b … f0; 1g, exponential u .x/ D i exp.x=i / and logarithmic
b i
ui .x/ D log.i C bx/ utility functions, with the same coefficient b for all the agents,
satisfy the above conditions and, hence, allow for aggregation in a strong sense, see
Rubinstein [1485], Milne [1344]. More precisely, in the case of generalized power
utility functions, we have the following result (compare with Huang & Litzenberger
[971, Section 5.25]).
Proposition 4.15 Suppose that markets are complete and let
1 b1
ui0 .x/ D ui1 .x/ D .i C bx/ b ;
b1
with b … f0; 1g, for all i D 1; : : : ; I. Then the implicit equilibrium prices p D
. p1 ; : : : ; pS / of the S Arrow securities are given by
P 1=b
I
ıs iD1 i C bes
ps D 1=b ; for all s D 1; : : : ; S:
PI

iD1 i C be 0
Proof Under the present assumptions, we can explicitly compute the functions u0
and u1 introduced in (4.29). Indeed, the generalized power utility form and the first
order conditions involved in the constrained maximization problems of (4.29) give
1 X
I
b
ai .i C bxi
s / D s and s D es ;
xi
iD1
for all s D 0; 1; : : : ; S and i D 1; : : : ; I, where s is the Lagrange multiplier

associated to the s-th state of the world. In particular, this implies that, for all
i D 1; : : : ; I and s D 0; 1; : : : ; S,
b
s
D i C bxi
s :
ai
PI
Summing over all i, recalling that i
iD1 xs D es , for all s D 0; 1; : : : ; S, and solving
for s , we get
! 1b ! 1b
X
I X
I
s D i C bes .a /
i b
:
iD1 iD1
In turn, from the above first order condition, this implies that
! 1b ! 1b
X
I X
I
1b
ai .i C bxi
s / D i C bes .ai /b ;
iD1 iD1
so that
X
I
ai b1
u0 .e0 / D .i C bxi
0 /
b
iD1
b1
! 1b ! 1b !
1 XI X
I X
I
D i C be0 .ai /b i C be0
b 1 iD1 iD1 iD1
! b1 ! 1b
1 XI b
X
I
D i C be0 .ai /b ;
b 1 iD1 iD1
and similarly for all u1 .es /, s D 1; : : : ; S. The equilibrium prices of the S Arrow
securities are then given by condition (4.32), which in the present case reduces to
P 1=b
I
ıs iD1 i C be s
ps D 1=b ;
PI
iD1 i C be 0
for all s D 1; : : : ; S, thus proving the claim. t

u
In particular, note that the equilibrium prices of the Arrow securities computed in
Proposition 4.15 do not depend on the weights .a1 ; : : : ; aI / but only on the aggregate
endowment .e0 ; e1 ; : : : ; eS /. As shown in Exercise 4.7, an analogous result holds true
in the case of exponential utility functions.
Summing up, if the aggregation property holds in a strong sense, then the
equilibrium prices of the original economy with I agents can be characterized
in terms of the no-trade equilibrium for a representative agent endowed with the
aggregate resources of the original economy, regardless of the initial resource
distribution.
Incomplete Markets
In reality, financial markets are typically incomplete, in the sense that not every
contingent consumption plan can be replicated by some portfolio of the securities
available for trade at the initial date t D 0. This means that S > N, i.e., the number
of (linearly independent) available securities is smaller than the number of possible
states of the world. In other words, the available securities do not completely span
the randomness of the economy.
In some cases, it is possible to make financial markets complete by introducing
financial derivatives. In general terms, a financial derivative is a security whose
dividend is defined in terms of the dividend of another security traded in the market
(underlying security). As a canonical example, consider the case of a (European)
Call option. A Call option is a financial derivative which gives to the holder the
right, but not the obligation, to buy at the future date t D 1 the underlying security
for some pre-specified price K (strike price), fixed at the initial date t D 0 of the
writing of the option. Following Ross [1465] and Breeden & Litzenberger [287], we
now show how the introduction of Call options can help completing the market.
Let us assume that there exists a portfolio (state index portfolio) zsi 2 RN such
that its dividend dsi delivered at t D 1 takes distinct values in correspondence of
different states of the world, i.e., the random variable dsi W ˝ ! R is such that
dsi .!r / ¤ dsi .!s / if r ¤ s, for all r; s D 1; : : : ; S. Without loss of generality, we
suppose that d si .!1 / < : : : < d si .!S /. Let dssi WD dsi .!s /, for s D 1; : : : ; S. Let
us also assume that agents have the possibility of investing in S 1 Call options,
having the state index portfolio as underlying security, with strike prices Ks D dssi ,
for s D 1; : : : ; S 1. Since the holder of a Call option is not obliged to exercise the
option at t D 1 (i.e., to buy the state index portfolio at the pre-specified price Ks ), he
will exercise the option only if the state index portfolio has at date t D 1 a dividend
dsi greater than the exercise price. Hence, the payoff at time t D 1 of the Call option
with strike price Ks is given by maxfdsi Ks I 0g, so that the s-th Call option will
be exercised only in correspondence of the states of the world .!sC1 ; : : : ; !S /. At
the initial date t D 0, there are S open financial markets, one market for the state
index portfolio and S 1 markets for the S 1 Call options. The dividend matrix D
associated to this economy can be represented as
2 si 3
d1 d2si d3si ::: dSsi
6 0 dsi dsi dsi dsi : : : dSsi d1si 7
6 2 1 3 1 7
60 0 d3si d2si : : : dSsi d2si 7
6 7:
6 :: :: 7
40 ::: 0 : : 5
0 0 ::: 0 dS dS1
si si
Since we assumed that d1si < : : : < dSsi , it holds that rank.D/ D S and, therefore,
markets are complete. As a consequence, if we allow agents to trade a sufficient
number of Call options and the latter allow to span all possible states of the world,
as it is the case under the present assumptions, then equilibrium allocations will
be Pareto optimal. Note that the existence of a state index portfolio is equivalent
to the existence of a security which allows to perfectly identify at time t D 1 the
realization of the state of the world, so that any other security can be viewed as
a derivative having the state index portfolio as underlying security. As such, the
existence of a state index portfolio represents a rather strong assumption.
When markets are incomplete (i.e., when rank.D/ < S), the available securities
do not span all possible states of the world, meaning that agents have limited
possibilities of transferring wealth across different states of the world by investing
in the market at the initial date t D 0. As a consequence, in correspondence of
an equilibrium allocation, agents will not have in general the same marginal rates
of substitution, meaning that there are Pareto improving reallocations of wealth
that cannot be implemented due to market incompleteness. This implies that, in
the presence of market incompleteness, equilibrium allocations are not necessarily
Pareto optimal. Moreover, when rank.D/ < S, there exist infinitely many solutions
to equation (4.27), so that the implicit prices of the Arrow securities are not uniquely
defined by the equilibrium prices of the N traded securities. Indeed, all possible
prices of the Arrow securities belong to a vector subspace of dimension S rank.D/.
In a two-period economy with a single good (which we generically interpret
as wealth) and not necessarily complete markets, it can be shown that every
allocation associated with a Radner equilibrium (see Sect. 4.2) is constrained ex-
ante Pareto optimal, in the sense of Definition 4.11, see Diamond [571], Stiglitz
[1568], Geanakoplos & Polemarchakis [765]. We present this result in a rather
general framework assuming that agents are endowed with traded assets as well
as with exogenous state-contingent wealth. More specifically, we assume that, for

every i D 1; : : : ; I, agent i is endowed with ei0 units of wealth at the initial date
t D 0 as well as with ein units of the n-th security, for all n D 1; : : : ; N. Moreover,
for every i D 1; : : : ; I, agent i is endowed with an additional state-contingent wealth
at time t D 1 (which can be thought of as labor income or as an exogenous source
of wealth), represented by the vector .v1i ; : : : ; vSi / 2 RS . In the present context, an
allocation can be described by a tuple .x10 ; : : : ; xI0 I z1 ; : : : ; zI / 2 RI RNI , where xi0
denotes agent i’s consumption at time t D 0 and zin denotes the number of units of the
n-th asset held in the portfolio by the i-th agent, for i D 1; : : : ; I and n D 1; : : : ; N.
Agent i’s consumption at P time t D 1 in correspondence of the state of the world
!s is given by xis D vsi C NnD1 zin dsn , for s D 1; : : : ; S. An allocation is said to be
feasible if
X
I X
I X
I X
I
xi0 ei0 and zin ein ; for all n D 1; : : : ; N:
iD1 iD1 iD1 iD1
Proposition 4.16 In the above setting, suppose that the agents’ utility functions are
strictly increasing and let .x1 I 1
0 ; : : : ; x0 I z ; : : : ; z / 2 R R
I I NI
be an equilibrium

allocation, with corresponding equilibrium prices p 2 RCC . Then the allocation
N
.x1 I 1
0 ; : : : ; x0 I z ; : : : ; z / is constrained ex-ante Pareto optimal.
I
Proof Arguing by contradiction, suppose that .x1 I 1

0 ; : : : ; x0 I z ; : : : ; z / is not
I
constrained ex-ante Pareto optimal. In this case, there exists a feasible allocation
.Ox10 ; : : : ; xO I0 I zO1 ; : : : ; zOI / such that
X
S X
S
s ui .Oxi0 ; xO is / s ui .xi
0 ; xs /;
i
for all i D 1; : : : ; I; (4.33)
sD1 sD1
P
with xO is D vsi C NnD1 zOin dsn , for all i D 1; : : : ; I and s D 1; : : : ; S, where
the inequality in (4.33) is strict for at least one i 2 f1; : : : ; Ig. Since the utility
function ui is strictly increasing, the budget constraint is satisfied as an equality in
correspondence of the optimal choice .xi 0 I z /, for all i D 1; : : : ; I, so that
i
X
N X
N
0 C
xi pn zi
n D e0 C
i
pn ein ; for all i D 1; : : : ; I: (4.34)
nD1 nD1
Hence, (4.33) and (4.34) together imply that
X
N X
N
xO i0 C pn zOin ei0 C pn ein ; for all i D 1; : : : ; I;
nD1 nD1
with strict inequality for at least one i 2 f1; : : : ; Ig. Summing over all i D 1; : : : ; I,
we get
I
X X
N XI X
N
xO i0 C pn zOin > ei0 C pn ein :
iD1 nD1 iD1 nD1
Since p 2 RNCC , this contradicts the feasibility of the allocation .Ox10 ; : : : ; xO I0 I zO1 ;
: : : ; zOI /, thus proving the claim. t
u
As shown in Sect. 4.2, market completeness suffices to ensure the (ex-ante)
Pareto optimality of equilibrium allocations. However, equilibrium allocations can
be Pareto optimal even in incomplete markets. More specifically, we say that
markets are effectively complete if every Pareto optimal allocation can be attained
through security markets. It can be shown that, if the consumption sets of the agents
are bounded from below and closed and markets are effectively complete, then
every equilibrium allocation is Pareto optimal, thus providing a version of the First
Welfare Theorem in the context of incomplete markets (see LeRoy & Werner [1191,
Chapter 16]).
When agents have homogeneous beliefs and time additive state independent
utility functions, a Pareto optimal allocation can be described in terms of a sharing
rule fyi W RC ! RC I i D 1; : : : ; Ig such that xi s D y .es /, for all s D 0; 1; : : : ; S
i
and i D 1; : : : ; I (see Proposition 4.4). In particular, the existence of a sharing rule

implies that Pareto optimal allocations only depend on the aggregate endowment
.e1 ; : : : ; eS / at time t D 1 in correspondence of the different states of the world
.!1 ; : : : ; !S / as well as on e0 . As a consequence, the financial market is effectively
complete as long as it is complete with respect to the aggregate endowment states
.e1 ; : : : ; eS /, meaning that, for all s D 0; 1; : : : ; S, there exists an Arrow security
paying one unit of wealth if the aggregate endowment equals es and zero otherwise.
If this is the case, in order to attain the Pareto optimal allocation, then every agent i
will choose to hold yi .es / units of such Arrow security, for each s D 0; 1; : : : ; S. Note
also that, as shown in Theorem 4.3, only the realization of the aggregate endowment
matters and not the specific state of the world associated to the realization. In
other words, in order to achieve effective completeness, it is not necessary to span
different states of the world associated with identical realizations of the aggregate
endowment. If markets are not complete with respect to the aggregate endowment
states, then, similarly as above, Call options written on the aggregate endowment
can be introduced in order to make the market effectively complete.
If the agents’ endowments are given in terms of units of the N available securities,
then the aggregate P P endowment portfolio (i.e., the portfolio delivering the aggregate
endowment IiD1 NnD1 ein dsn in correspondence of the state of the world !s , for
each s D 1; : : : ; S) is always traded in the market. We call it the market portfolio.
In this case, markets are effectively complete if and only if the agents’ consumption
plans in correspondence of every Pareto optimal allocation belong to the asset span
(see LeRoy & Werner [1191, Section 16.3]). When the agents’ endowments are
not only defined in terms of the traded assets (for instance, agent i’s endowment
is composed of traded assets as well as of an exogenous vector .v1i ; : : : ; vSi /, as
considered before Proposition
P P4.16), markets Pare effectively complete only if the
aggregate endowment IiD1 NnD1 ein dsn C IiD1 vsi , for s D 1; : : : ; S, lies in the
asset span.
As a simple consequence of the above observations, in the absence of aggregate
risk every equilibrium allocation will be Pareto optimal regardless of the complete-
ness of the market, provided that the agents’ endowments at t D 1 belong to the asset
span. Indeed, suppose that the agents’ endowments are defined in terms of units of
the traded assets and the aggregate endowment does not depend on the state of the
world (i.e., there is no aggregate risk). Then, as a consequence of Theorem 4.3, the
consumption plan of every agent in correspondence of a Pareto optimal allocation
is riskless and can be attained by simply holding a fraction of the market portfolio,
whose dividend is constant across all possible states of the world. Hence, in the
present situation, markets are effectively complete and every equilibrium allocation
is Pareto optimal (see also Exercise 4.8 and compare with LeRoy & Werner [1191,
Section 16.5] and Munk [1363, Section 7.4.2]).
In a similar way, Proposition 4.5 shows that when the agents’ utility functions
exhibit linear risk tolerance with the same cautiousness coefficient, then the Pareto
optimal sharing rule is linear with respect to the aggregate endowment. This result
implies that every equilibrium allocation will be Pareto optimal provided that a risk
free asset is traded in the market and the agents’ endowments lie in the asset span
(for instance, agents’ endowments are given in terms of units of the traded assets,
as considered above). In this case, to reach a Pareto optimal allocation, every agent
needs only to trade the risk free asset and the market portfolio, thus confirming
that the agents’ consumption plans in correspondence of Pareto optimal allocations
exhibit a two funds separation property.
Concerning the possibility of characterizing equilibrium prices by means of the
no-trade equilibrium of a single agent economy, an analysis based on a represen-
tative agent is possible whenever equilibrium allocations are Pareto optimal, as
it is the case in complete markets (see Sect. 4.2). However, if the agents’ utility
functions are of the form considered in Proposition 4.5, then the construction of
a representative agent in a strong sense and the aggregation property can also be
established in incomplete markets provided that a risk free asset is traded and the
aggregate endowment is spanned by the traded assets (or the agents’ endowments
are directly given in terms of units of the traded assets), see Milne [1344] and
Detemple & Gottardi [563]. In this case, the representative agent’s utility function
will belong to the same class of the agents’ utility functions (compare with the proof
of Proposition 4.15). Equilibrium asset prices will be characterized as in (4.32) in
terms of the marginal utility of the representative agent evaluated in correspondence
of the aggregate endowment of the economy. Summing up, three properties are
strictly related: effective completeness, demand aggregation and fund separation.
In this regard, the HARA class of utility functions plays a crucial role allowing
to obtain these three key properties, under suitable conditions on the economy’s
structure. We refer to Chap. 9 for a further analysis of the implications of market
incompleteness.
4.4 The Fundamental Theorem of Asset Pricing
In this section, we study the relations between three cornerstones of modern finance
theory, namely
(a) the absence of arbitrage opportunities,
(b) the existence of a positive linear pricing functional,
(c) the existence of an equilibrium of the economy.
Loosely speaking, the Fundamental Theorem of Asset Pricing asserts the equiva-
lence of the above three properties, providing on the one hand a precise character-
ization of viable economies (i.e., economies for which there exists an equilibrium)
and, on the other hand, opening the door to the risk neutral valuation of financial
derivatives. Results on the fundamental theorem of asset pricing can be traced back
to Ross [1464], Harrison & Kreps [904] and Harrison & Pliska [905] (see also the
notes at the end of the chapter).
In the first part of this section, we introduce the notion of arbitrage opportunity
and provide the precise connection between properties (a) and (b), while, in the
second part, we shall be concerned with the relation with property (c). Finally, we
illustrate the above concepts in the context of the classical binomial model, with a
special focus on the valuation and hedging of financial derivatives, and we present
the classical result of the Modigliani-Miller theorem.
As in Sect. 4.3, we always assume that agents live in a two-period economy
(i.e., t 2 f0; 1g) with a single consumption good and S possible states of the world,
characterized by probabilities .1 ; : : : ; S /, with s > 0 for all s D 1; : : : ; S. As
before, we assume that agents have homogeneous beliefs. However, we want to
point out that the results presented below also hold in the case of heterogeneous
beliefs .1i ; : : : ; Si /, for i D 1; : : : ; I, provided that all the agents agree on null
sets, i.e., si > 0 for all i D 1; : : : ; I and s D 1; : : : ; S. In the economy, there
are N securities available for trade at the initial date t D 0, with prices p 2 RN .
We denote by D 2 RSN the matrix representing the dividends of the N assets in
correspondence of the S possible states of the world. In this context, as considered
in Sect. 4.3, a portfolio of the N securities is represented by a vector z 2 RN , where
zn denotes the number of units of security n held in the portfolio, for n D 1; : : : ; N.
For a portfolio z 2 RN , its price at date t D 0 is given by p> z and the corresponding
random dividend (or payoff ) at t D 1 is represented by the vector Dz 2 RS .
No-Arbitrage and Risk Neutral Probability Measures
In real financial markets, there should be no possibility of creating wealth out of

nothing without risk. This intuitive concept is formalized by the notion of arbitrage
opportunity, as presented in the following definition. Recall that, for a vector x 2 RS ,
4.4 The Fundamental Theorem of Asset Pricing 165
the inequality x 0 means that xs 0 for all s D 1; : : : ; S, while x > 0 means that
xs 0 for all s D 1; : : : ; S and xs > 0 for at least one s 2 f1; : : : ; Sg.
Definition 4.17 For a price-dividend couple . p; D/ 2 RN RSN , a portfolio z 2
RN is said to be
(i) an arbitrage opportunity of the first kind if p> z 0 and Dz > 0;
(ii) an arbitrage opportunity of the second kind if p> z < 0 and Dz 0.
A portfolio z 2 RN is simply said to be an arbitrage opportunity if it is an arbitrage
opportunity of the first kind and/or an arbitrage opportunity of the second kind. We
say that the market is arbitrage free if . p; D/ does not admit arbitrage opportunities.
In view of the above definition, an arbitrage opportunity of the first kind
represents an investment opportunity with zero (or even negative) cost at time t D 0
yielding a non-negative and non-zero payoff at time t D 1. Similarly, an arbitrage
opportunity of the second kind represents an investment opportunity with a negative
cost at time t D 0 yielding a non-negative (but possibly null) payoff at time t D 1.
In general, the two concepts are distinct (see Exercise 4.15). Defining the matrix

>
p
D WD 2 R.SC1/N ;
D
a portfolio z 2 RN is an arbitrage opportunity if and only if Dz > 0. The following

proposition presents some simple but important consequences of the absence of
arbitrage opportunities.
Proposition 4.18 If there are no arbitrage opportunities, then the following hold:
(i) the Law of One Price holds, i.e., if z; z0 2 RN are two portfolios which satisfy
Dz D Dz0 , then p> z D p> z0 ;
(ii) if z 2 RN is a portfolio such that Dz D 0, then p> z D 0;
(iii) there is no riskless arbitragePopportunity, i.e., there does not exist a portfolio
z 2 RN with p> z 0 and NnD1 dsn zn D c, for all s D 1; : : : ; S, for some
c > 0.
Proof (i): Arguing by contradiction, let z; z0 2 RN satisfy Dz D Dz0 and suppose
that p> z z0 . Define the portfolio zN WD z z0 . Then p> zN < 0 and DNz D 0, thus
showing that zN is an arbitrage opportunity of the second kind. This contradicts the
assumption of the absence of arbitrage opportunities (the case p> z > p> z0 can be
treated in an analogous way).
(ii): It suffices to apply part (i) with z0 D 0 2 RN .
(iii): This follows directly by Definition 4.17, since a riskless arbitrage opportunity
is an arbitrage opportunity of the first kind. t
u
As can be seen by inspecting the proof of the above proposition, statements (i)-
(ii) are actually equivalent, in the sense that the Law of One Price holds if and only
if Dz D 0 implies that p> z D 0. In particular, the Law of One Price implies that
portfolios with identical payoffs must also have identical prices, otherwise it would
be possible to realize an arbitrage opportunity. However, the implication does not
hold in the converse direction, meaning that it is possible that the Law of One Price
holds but, nevertheless, arbitrage opportunities (of the first kind) are present, as
shown in Exercise 4.16.
Having introduced the notion of arbitrage opportunity, we now aim at providing
a general characterization of an arbitrage free market. To this end, we have to define
the notion of pricing functional. As a preliminary, let c 2 RS be a payoff-contingent
consumption plan which can be attained by trading in the securities available in the
market, i.e., c 2 I.D/, with I.D/ denoting the image of the dividend matrix D. This
means that there exists a portfolio zc 2 RN satisfying Dzc D c (we say that zc is the
replicating portfolio for the payoff c). As explained in Sect. 4.2, the market value
of the payoff c, defined in Definition 4.9 and denoted by V.c/, is given by the value
at time t D 0 of the replicating portfolio zc , so that V.c/ D p> zc . In particular, if
there are no arbitrage opportunities, Proposition 4.18 implies that, if there exist two
distinct portfolios z and z0 such that c D Dz D Dz0 , then it holds that p> z D p> z0
(Law of One Price), so that V W I.D/ ! R is indeed a function (meaning that, for all
c 2 I.D/, the market value V.c/ is uniquely defined). Of course, if markets are not
complete (i.e., if rank.D/ < S), then not every contingent consumption plan c 2 RS
can be perfectly replicated. We introduce the following definition.
Definition 4.19 A mapping Q W RS ! R is said to be a pricing functional on RS if
Q.c/ D V.c/, for every c 2 I.D/.
In other words, a pricing functional Q extends the market valuation functional
V W I.D/ ! R to contingent consumption plans which do not necessarily lie in the
span of the traded securities. A pricing functional Q is said to be linear if, for every
c; c0 2 RS and ˛; ˇ 2 R, it holds that Q.˛c C ˇc0 / D ˛Q.c/ C ˇQ.c0 /. A pricing
functional Q is said to be positive (strictly positive, resp.) if, for every c 0 (c > 0,
resp.), it holds that Q.c/ 0 (Q.c/ > 0, resp.).
The following theorem contains the essence of the Fundamental Theorem
of Asset Pricing and, in particular, shows the equivalence between the absence
of arbitrage opportunities and the existence of a strictly positive linear pricing
functional.
Theorem 4.20 For a price-dividend couple . p; D/, the following are equivalent:
(i) there are no arbitrage opportunities;
(ii) there exists a strictly positive linear pricing functional Q W RS ! R;
(iii) there exists a solution m 2 RSCC to the system p D D> m.
Proof .i/ ) .ii/: let us define the linear space M WD fDz W z 2 RN g and the cone
H WD RSC1C . Both M and H are closed and convex subsets of R
SC1
. The assumption
that there are no arbitrage opportunities can be equivalently formulated as
H \ M D f0g:
The separating hyperplane theorem3 gives the existence of a non-zero linear

functional F W RSC1 ! R such that
F.x/ < F. y/; for all x 2 M and y 2 H n f0g:
Since M is a linear space, this implies that F.x/ D 0, for all x 2 M, and, hence,
F. y/ > 0, for all y 2 H n f0g. In turn, since F is a linear functional on RSC1 ,
>
there exists a vector u D .u0 ; u1 ; : : : ; uS / 2 RSC1
CC such that F. y/ D u y, for all
y 2 R . Define then the functional Q W R ! R by Q.c/ WD .u1 ; : : : ; uS /> c=u0 ,
SC1 S
for all c 2 RS . Clearly, Q is a strictly positive linear functional on RS . Moreover, if

c 2 I.D/, so that there exists zc 2 RN satisfying c D Dzc , it holds that
1 1 1 c
Q.c/ D .u1 ; : : : ; uS /> c D .u1 ; : : : ; uS /> Dzc D F Dz C u0 p> zc
u0 u0 u0
D p> zc D V.c/;
thus showing that Q.c/ D V.c/, for all c 2 I.D/. In view of Definition 4.19, we
have thus proved that Q is a strictly positive linear pricing functional.
.ii/ ) .iii/: using the notation introduced in the previous step of the proof, define
m WD .u1 ; : : : ; uS /=u0 2 RSCC . Since u> x D 0, for all x 2 M (equivalently, u> Dz D
0, for all z 2 RN ), we obtain, dividing by u0 and recalling that u0 > 0,
p> z C m> Dz D 0; for all z 2 RN :
This implies that m 2 RSCC solves the system p D D> m.

.iii/ ) .i/: arguing by contradiction, suppose that z 2 RN is an arbitrage
opportunity, so that Dz > 0. Then, since m 2 RSCC and p D D> m, it holds that
0 < .1; m1 ; : : : ; mS /Dz D p> z C m> Dz D 0;
thus yielding a contradiction. t

u
As can be seen be inspecting the proof of the above theorem, the pricing
functional Q is completely characterized by the vector m, in the sense that Q.c/ D
m> c, for all c 2 RS . Moreover, the quantities .m1 ; : : : ; mS / admit the interpretation
of state prices, i.e., ms represents the market value at time t D 0 of one unit of wealth
contingent on the realization of the state of the world !s , for s 2 f1; : : : ; Sg. Indeed,
let 1s denote the vector .0; : : : ; 0; 1; 0; : : : ; 0/ 2 RS , with the element 1 being in the
s-th position. Then, recalling that Q.c/ D m> c, for all c 2 RS , and applying the
3
In the present context, the separating hyperplane can be stated as follows (see Duffie [593]). Let
M and H be closed convex cones in RSC1 such that M \ H D f0g. Then, if H does not contain a
linear subspace other than f0g, than there is a non-zero linear functional F W RSC1 ! R such that
F.x/ < F. y/, for all x 2 M and y 2 H n f0g.
pricing functional Q to the payoff represented by the vector 1s , we get Q.1s / D ms ,

for all s D 1; : : : ; S. In this sense, the state price vector m describes the arbitrage
free prices of the Arrow securities implicit in the price-dividend couple . p; D/.
However, the state price vector m is not necessarily uniquely specified (equivalently,
Theorem 4.20 does not assert the uniqueness of the pricing functional Q). As will
be shown below in Proposition 4.26, m is unique as long as markets are complete.
Even though a risk free asset is not necessarily traded in the market, a pricing
functional Q determines the risk free rate implicit in the price-dividend couple
. p; D/. Indeed, let us consider the riskless payoff PS1 WD .1; : : : ; 1/ 2 R . Then,
S
>
in view of Theorem 4.20, we get Q.1/ D m 1 D sD1 ms . Hence, the risk free rate
P
rf implicitly determined by the pricing functional Q is given by 1=rf D SsD1 ms .
Note that the absence of arbitrage opportunities implies that there cannot exist two
portfolios with riskless payoffs and different rate of returns (see Exercise 4.19 and
compare also with Proposition 4.18).
The absence of arbitrage opportunities admits an alternative characterization in
terms of the existence of a risk neutral probability measure, as defined below.
PS
Definition 4.21 A vector D .1 ; : : : ; S / 2 RSCC with
sD1 s D 1
represents a risk neutral probability measure for the price dividend couple . p; D/ if
1 X
S
pn D dsn ; for all n D 1; : : : ; N;
rf sD1 s
where rf denotes the risk free rate of return.

We have the following result, which relates the absence of arbitrage opportunities
to the existence of a risk neutral probability measure.
Proposition 4.22 For a price-dividend couple . p; D/, the following are equiva-
lent:
(ii) there exists a risk neutral probability measure D .1 ; : : : ; S /.
Proof .i/ ) .ii/: if there are no arbitrage opportunities, then Theorem 4.20 gives
the existence of a state price vector m 2 RSCC . Let
X
S
ms
m0 WD ms and s WD ; for all s D 1; : : : ; S:
sD1
m0
Then, recalling that rf D 1=m0 > 0 (see above) and that p D D> m, we get
X
S X
S
ms 1 X
S
pn D dsn ms D m0 dsn D dsn ; for all n D 1; : : : ; N;
sD1 sD1
m0 rf sD1 s
thus showing that .1 ; : : : ; S / D .m1 =m0 ; : : : ; mS =m0 / is a risk neutral probability
measure.
.ii/ ) .i/: let .1 ; : : : ; S / be a risk neutral probability measure and let z 2 RN
P
be an arbitrage opportunity. Then, since pn D r1f SsD1 s dsn , for all n D 1; : : : ; N,
and Dz 0, it holds that (recalling that 2 RSCC )
1 XX
N S
>
p zD dsn zn 0;
rf nD1 sD1 s
thus showing that z cannot be an arbitrage opportunity. t

u

Denoting by E Œ the expectation with respect to a risk neutral probability
measure , it then follows directly from Definition 4.21 that
1 Q
pn D E Œdn ; for all n D 1; : : : ; N; (4.35)
rf
where dQ n denotes the random variable taking values .d1n ; : : : ; dSn /.

The valuation formula (4.35) can also be extended to payoffs not traded in
the market. Indeed, recalling the relation between a pricing functional Q, the
corresponding state price vector m and the associated risk neutral probability
measure (see Proposition 4.22), it holds that4
1 X
S
1
Q.c/ D m> c D s cs D E ŒQc; for all c 2 RS : (4.36)
rf sD1 rf
Formulae (4.35)–(4.36) explain why a probability measure D .1 ; : : : ; S / as

in Definition 4.21 is called risk neutral. Indeed, the arbitrage free price at time t D 0
of a random payoff is simply given by its expectation (computed with respect to the
probability measure ), discounted by the risk free rate of return. Equivalently,
letting rQn WD dQ n =pn denote the random return of security n (assuming that pn > 0),
for n D 1; : : : ; N, it holds that
1 Q
E ŒQrn D E Œdn D rf ; for all n D 1; : : : ; N:
pn
This shows that, under a risk neutral probability measure , the expected rate of
return of every traded security coincides with the risk free rate of return. Moreover,
for all s D 1; : : : ; S, the arbitrage free price of the s-th Arrow security 1s paying one
4
In the following, with some abuse of notation, we equivalently denote by
xQ or x D .x1 ; : : : ; xS / 2 RS the random variable Qx taking values .x1 ; : : : ; xS / in the S states
of the world.
unit of wealth in correspondence of the state of the world !s and zero otherwise is
given by
Q.1s / D ms D m0 s D s =rf :
A risk neutral probability measure D .1 ; : : : ; S / is related to the original

(also called physical/statistical/historical) probability measure .1 ; : : : ; S / by the
likelihood ratio ` D .`1 ; : : : ; `S / WD .1 =1 ; : : : ; S =S /. In particular, it holds
that `s > 0, for all s D 1; : : : ; S. This shows that the two probability measures
are equivalent, in the sense that they both assign a strictly positive probability
to all states of the world .!1 ; : : : ; !S /. Moreover, we can derive an alternative
representation of the risk neutral valuation formula (4.36) in terms of the original
probability measure and of the likelihood ratio `:
1 Q 1
Q cQ / ;
Q.c/ D EŒ`Qc D EŒQc C Cov.`; (4.37)
rf rf
where we have used the identity Cov.`; Q cQ / D EŒ`Q

Qc EŒ`Q EŒQc, together with the
Q
fact that EŒ` D 1. Formula (4.37) shows that the arbitrage free price of a random
payoff cQ represented by a vector c 2 RS can be decomposed into two terms: a first
term which is simply given by the expected payoff discounted by the risk free rate
and a second term which depends on the covariance between the random payoff and
the likelihood ratio.
As shown in Theorem 4.20, the absence of arbitrage opportunities implies the
existence of a pricing functional Q, or, equivalently (see Proposition 4.22), the
existence of a risk neutral probability measure . For a random payoff represented
by a vector c D .c1 ; : : : ; cS / 2 RS , the value Q.c/ represents an arbitrage free price
at time t D 0 in the sense that, if the original financial market composed of the
N traded securities is extended to include the payoff c at the price Q.c/, then no
arbitrage opportunity will be present in the extended market. To prove this claim,
we shall first consider the case of a complete market and then the more general case
of an incomplete market.
If markets are complete then I.D/ D RS , so that all random payoffs c 2 RS can
be attained by trading in the N available securities. As a consequence, the unique
arbitrage free price of the payoff c is given by Q.c/ D V.c/ D p> zc , where zc 2 RN
is a portfolio satisfying Dzc D c. It is easy to check that any other price for the payoff
c necessarily introduces arbitrage opportunities in the extended market composed of
the N original securities together with the payoff c (see Exercise 4.20). This proves
that, if markets are complete, then the arbitrage free price of any payoff-contingent
consumption plan is uniquely determined by the pricing functional Q.
When markets are incomplete, things are not so simple and the arbitrage free
price of a payoff-contingent consumption plan c 2 RS is not necessarily uniquely
defined. Nevertheless, one can determine an interval to which the price of a given
payoff c 2 RS must belong in order to exclude arbitrage opportunities in the
extended market where the payoff c is traded together with the original N securities.
As a preliminary, let us give the following definition.5
Definition 4.23 Let c 2 RS represent a random payoff. The super-replication price
of c, denoted by qu .c/, is defined as
qu .c/ WD minf p> z W z 2 RN and Dz cg:
Similarly, the sub-replication price of c, denoted by ql .c/, is defined as
ql .c/ WD maxf p>z W z 2 RN and Dz cg:
The super-replication price qu .c/ of a payoff c represents the smallest amount

of wealth needed to form a portfolio having a payoff always greater or equal than
the payoff c, for every possible state of the world. An analogous interpretation can
be given for the sub-replication price ql .c/. The following result holds on qu .c/ and
ql .c/ (the proof is given in Exercise 4.21).
Proposition 4.24 If there are no arbitrage opportunities, then the following hold:
(i) for every c 2 RS , it holds that qu .c/ ql .c/;
(ii) if c 2 I.D/, then qu .c/ D ql .c/ D V.c/;
(iii) if c … I.D/, then qu .c/ > ql .c/.
In particular, statement (ii) of the above proposition shows that, if a payoff c is
attainable, then its arbitrage free price is uniquely defined and is equal to the value
of its replicating portfolio zc . Equivalently, if a payoff c belongs to I.D/, we can
say that c is redundant, in the sense that adding the payoff c to the original market
does not alter the space of contingent consumption plans that can be spanned by the
available securities. On the other hand, if c … I.D/, we say that c is non-redundant.
If the payoff c is not attainable (if c … I.D/), then there exists a non-trivial interval
of possible arbitrage free prices. This is illustrated in Fig. 4.3 in a case with two
states and assets that span
a line. For a payoff c which cannot be attained, any price
outside of the interval ql .c/; qu .c/ leads to arbitrage opportunities in the extended
financial market composed of the N original securities together with the payoff c
(see Exercise 4.23).
As remarked after Theorem 4.20, the absence of arbitrage opportunities does not
ensure the uniqueness of the pricing functional Q or, equivalently, of the risk neutral
probability measure D .1 ; : : : ; S /. In the following proposition, we present
a dual characterization of the super-replication and sub-replication prices in terms
of risk neutral probability measures. We denote by ˘ the set of all risk neutral
5
In the absence of arbitrage opportunities, it can be shown that inff p> z W z 2 RN and Dz cg D
minf p> z W z 2 RN and Dz cg and, similarly, supf p> z W z 2 RN and Dz cg D maxf p> z W
z 2 RN and Dz cg, meaning that the infimum and the supremum are actually attained by some
portfolios, see Föllmer & Schied [721, Theorem 1.32].
x1
I(D)
q (c) qu (c)
x2
Fig. 4.3 Super and sub-replication in an incomplete market
probability measures associated to a given price-dividend couple . p; D/, i.e.,

X
S X
S

˘ D 2 RSCC W s D 1 and pn D s dsn =rf for all n D 1; : : : ; N ;
sD1 sD1
with respect to a risk free return rf .

Proposition 4.25 Let c 2 RS . If there are no arbitrage opportunities, then
1 X
S
qu .c/ D sup s cs DW hsup .c/; (4.38)
2˘ rf sD1
1 X
S
ql .c/ D inf cs DW hinf .c/: (4.39)
2˘ rf sD1 s
Proof We only present the proof of (4.38), the proof of (4.39) being analogous. Let
z 2 RN be a portfolio such that Dz c. Then, for any 2 ˘ , it holds that
1 X 1 XX X
S S N N
s cs s dsn zn D pn zn D p> z:
rf sD1 rf sD1 nD1 nD1
Taking the infimum over all portfolios z 2 RN such that Dz c and the supremum
over all risk neutral probability measures 2 ˘ , we get hsup .c/ qu .c/. It
remains to prove the converse inequality. Let p" WD hsup .c/ C ", for some " > 0, and
consider the extended financial market given by the N original securities together
with the payoff c traded at price p" . Since p" > hsup .c/, there exists no risk neutral
probability measure for the extended market and, hence, the extended market is
not arbitrage free (see Proposition 4.22). This means that there exist z 2 RN and
z0 2 R n f0g such that
Dz C cz0 0 and p> z C p" z0 0; (4.40)
with at least one the two inequalities being strict. By taking the expectation with
respect to any risk neutral probability measure 2 ˘ (and discounting by the
risk free rate rf ) and then taking the supremum over all 2 ˘ , the first inequality
in (4.40) gives p> z C hsup .c/z0 0. Together with the second inequality in (4.40),
this implies that z0 < 0. But then, again the first inequality of (4.40) implies that
Dz=z0 c. In view of Definition 4.23, this means that
hsup .c/ C " D p" p> z=z0 qu .c/:
Since " is arbitrary, this proves the desired inequality. t

u
In turn, Proposition 4.25 allows to derive a general criterion in order to establish
when, in a general incomplete market, a given payoff can be attained by a portfolio
of the N available securities. Indeed, summarizing the main results presented so
far, we have that, if there are no arbitrage opportunities, then the following are
equivalent, for any c 2 RS :
(i) c 2 I.D/, i.e., there exists a portfolio zc 2 RN such that Dzc D c;
(ii) the map ˘ 3 7! r1f E ŒQc is constant, i.e., the discounted risk neutral
expectation of cQ does not depend on the choice of the risk neutral probability
measure;
(iii) qu .c/ D ql .c/.
In incomplete markets, there exist infinitely many risk neutral probability measures.
However, as shown above, the risk neutral expectation of any attainable payoff does
not depend on the specific risk neutral probability measure chosen.
In general, financial markets are incomplete. It is therefore of interest to provide
a precise characterization of market completeness. This is the content of the
following theorem (which is usually called the second Fundamental Theorem of

Asset Pricing), showing that, in the absence of arbitrage opportunities, the risk
neutral probability measure is unique if and only if the market is complete.
Theorem 4.26 If there are no arbitrage opportunities, then the following are
equivalent:
(i) the market is complete, i.e., I.D/ D RS ;
(ii) there exists a unique solution m 2 RSCC to the system p D D> m;
(iii) there exists a unique risk neutral probability measure.
Proof In view of the proof of Proposition 4.22, it is clear that (ii) and (iii) are
equivalent. The implication .iii/ ) .i/ follows from the above arguments. In order
to show that (i) implies (ii), suppose that there exist two vectors m; m0 2 RSCC such
that p D D> m D D> m0 , so that D> .m m0 / D 0. Since I.D/ D RS is equivalent to
rank.D/ D S, this implies that m D m0 , thus proving the claim. t
u
We want to remark that, in general, there is no implication between the absence
of arbitrage opportunities and the notion of market completeness, in the sense that
markets can be complete even in the presence of arbitrage opportunities (an explicit
example is given in Exercise 4.20).
No-Arbitrage and Equilibrium
Having analysed the relation between the absence of arbitrage opportunities and the
existence of a strictly positive linear pricing functional (or, equivalently, of a risk
neutral probability measure), we are now in the position to investigate the relation
between the absence of arbitrage opportunities and the solvability of portfolio
optimization problems and, in turn, the existence of an equilibrium.
We consider an agent’s optimal portfolio problem in a general (complete or
incomplete) market, with intertemporal consumption and S possible states of the
world, as considered in Sect. 4.3. The general economy is populated by I agents,
with state independent utility functions ui W R2 ! R defined with respect to
consumption at time t D 0 and consumption at time t D 1, for i D 1; : : : ; I. As
before, we suppose that there are N securities traded at time t D 0 with dividend
matrix D 2 RSN and price vector p 2 RN . For all i D 1; : : : ; I, the endowment of
agent i will be denoted by .ei0 ; ei1 ; : : : ; eiS / 2 RSC1 , where ei0 represents the wealth
of agent i in t D 0 and eis the wealth of agent i in t D 1 contingent on state !s ,
for s D 1; : : : ; S. The endowment .ei1 ; : : : ; eiS / can represent exogenous contingent
wealth as well as the contingent wealth generated by an initial endowment in the
N available securities. We generically denote by z 2 RN a trade vector (portfolio),
expressed in terms of units bought/sold of the N securities. In this context, denoting
by .xi0 ; xi1 ; : : : ; xiS / 2 RSC1 the consumption plan of agent i in t D 0 and in t D 1 in
correspondence of the S possible states of the world, the optimal portfolio problem
can be written as
X
S
max s ui .xi0 ; xis /; (PO4)
sD1
where the consumption plan .xi0 ; xi1 ; : : : ; xiS / 2 RSC1 is subject to the budget
constraint
X
N
xi0 ei0 p> z and xis eis C dsn zn ; for all s D 1; : : : ; S; (4.41)
nD1
for some trade vector z 2 RN . We denote by zi 2 RN the optimal trade vector
of agent i. Recall that, if the utility function ui is strictly increasing in both of its
arguments, then the budget constraint (4.41) can be written as an equality. In this
case, given the endowment .ei0 ; ei1 ; : : : ; eiS /, the optimal consumption Problem (PO4)
reduces to determining the optimal portfolio zi 2 RN .
Proposition 4.27 For any i 2 f1; : : : ; Ig, suppose that the utility function
ui W R2 ! R is strictly increasing in both of its arguments and that there exists
an optimal portfolio zi 2 RN in correspondence of the price-dividend couple
. p; D/. Then the are no arbitrage opportunities.
Proof Arguing by contradiction, let zi 2 RN denote the optimal portfolio
associated to Problem (PO4) and suppose that zN 2 RN is an arbitrage opportunity.
Consider then the following problem
X
S X
N
>
max s ui xi
0 p N
z ; x i
s C d sn N
z n ; (4.42)
2RC
sD1 nD1
0 and xs denote the optimal consumption of agent i at date t D 0 and

where xi i
t D 1, respectively, given as the solution to Problem (PO4) subject to the budget

constraint (4.41). Since the optimal portfolio zi satisfies the budget constraint,
then the portfolio zi C Nz also satisfies the budget constraint, for any 2 RC ,
because zN is an arbitrage opportunity. Since at least one of the two inequalities
p> zN 0 and DNz 0 is strict and the utility function ui is strictly increasing in
both of its arguments, the optimization problem (4.42) does not admit a solution,
thus contradicting the existence of an optimal portfolio zi . t
u
Proposition 4.27 shows that, if agents’ preferences are characterized by utility
functions which are strictly increasing both in consumption at time t D 0 and in
consumption at time t D 1, then the presence of arbitrage opportunities is incom-
patible with the existence of an optimal portfolio. Since arbitrage opportunities
represent portfolios which “create wealth out of nothing”, such a result is hardly
surprising. In particular, Proposition 4.27 shows that the existence of an optimal
portfolio zi excludes arbitrage opportunities of the first as well as of the second
kind. If the utility function ui W R2 ! R is only assumed to be strictly increasing
in the first argument and non-decreasing in the second argument, then the result of
Proposition 4.27 holds true with respect to arbitrage opportunities of the second kind
(see Exercise 4.30 and compare also with LeRoy & Werner [1191, Section 3.6]).
Proposition 4.27 concerns the optimal portfolio problem of individual agents,
but can be easily extended to a Radner equilibrium of the economy, as shown in the
following corollary, the proof of which is an immediate consequence of the previous
proposition. Indeed, in view of Definition 4.7, every agent’s equilibrium portfolio
has to be an optimal portfolio in the sense of Problem (PO4).
Corollary 4.28 Suppose that ui W R2 ! R is strictly increasing in both of its
arguments, for some i D 1; : : : ; I, and let . p; D/ be the price-dividend couple
associated to a Radner equilibrium of the economy. Then there are no arbitrage
opportunities.
We now aim at studying the converse implication. More precisely, starting from a
price-dividend couple . p; D/ which does not admit arbitrage opportunities, we want
to address the existence of an optimal portfolio for an individual agent. As a second
step, we want to study the existence of a Radner equilibrium of an economy such
that p corresponds to the price vector implementing the equilibrium allocation. We
say that the price-dividend couple . p; D/ represents a viable financial market if there
exists an economy supporting the price vector p in correspondence of an equilibrium
allocation. The following proposition provides a first and simple answer to such a
question.
Proposition 4.29 Let . p; D/ be a price-dividend couple which does not admit
arbitrage opportunities. Then there exists an economy populated by I risk neutral
agents with arbitrary endowments .ei0 ; ei1 ; : : : ; eiS / 2 RSC1 , for all i D 1; : : : ; I, with
beliefs given by risk neutral probabilities and a risk free discount factor 1=rf , such
that the price vector p is associated to a Radner equilibrium of this economy.
Proof If . p; D/ does not admit arbitrage opportunities then, in view of Proposi-
tion 4.22, there exists a risk neutral probability measure D .1 ; : : : ; S / (note
that, if the market is incomplete, there exist infinitely many risk neutral probability
measures: for the purposes of this proof, it suffices to choose an arbitrary element
2 ˘ ). For all i D 1; : : : ; I, let .ei0 ; ei1 ; : : : ; eiS / be an arbitrary vector in RSC1
and consider the expected utility function
1 X i
S
U i .xi0 ; xi1 ; : : : ; xiS / D xi0 C x;
rf sD1 s s
where 1=rf is given by the sum of the state prices associated to (compare with the
discussion before Definition 4.21). Then, the portfolio optimization problem (PO4)
subject to the budget constraint (4.41) can be rewritten as

1 X i X
S N
max ei0 p> zi C s es C dsn zin :
zi 2RN rf sD1 nD1
P
Since is a risk neutral probability measure, the equality pn C r1f SsD1 s dsn D
0 is always satisfied, so that any portfolio which satisfies the budget constraint (4.41)
as an equality is optimal. In particular, the portfolio zi D 0 2 RN is an optimal
choice, for all i D 1; : : : ; I (i.e., each agent consumes the wealth generated by his
initial endowment). In view of Definition 4.7, this proves the existence of a Radner
equilibrium associated to the price vector p 2 RN . t
u
The above result shows that, if all the agents of the economy are risk neutral and
their beliefs correspond to a risk neutral probability measure, then there exists a no-
trade equilibrium associated with the arbitrage free price-dividend couple . p; D/.
In that sense, the couple . p; D/ represents a viable financial market. As can be
seen by inspecting the proof of Proposition 4.29, in the presence of an incomplete
market, the I agents are also allowed to exhibit heterogeneous beliefs, as long as all
the agents’ beliefs correspond to risk neutral probabilities. Proposition 4.29 also
explains why the probabilities are called risk neutral. Indeed, a probability
measure 2 ˘ corresponds to the beliefs of a risk neutral agent supporting
the price vector p in correspondence of a no-trade equilibrium. Note also that, in
the setting of Proposition 4.29, nothing prevents us from choosing I D 1, so that
the arbitrage free price-dividend couple . p; D/ also corresponds to the no-trade
equilibrium of a single risk neutral representative agent whose beliefs correspond to
a risk neutral probability measure (in this regard, compare also with the discussion
in Lengwiler [1182, Section 5.3.2]).
As we have seen in the previous chapters, economic agents are typically risk
averse. We now aim at understanding the implications of the absence of arbitrage
opportunities on the existence of an optimal portfolio for a risk averse agent and,
consequently, on the existence of a Radner equilibrium for an economy populated
by risk averse agents.
Let us consider the optimal consumption Problem (PO4) subject to the additional
constraint that the optimal consumption plan .xi 0 ; x1 ; : : : ; xS / is non-negative, i.e.,
i i
X
S
max s ui .xi0 ; xis /; (PO4’)
sD1
X
N
0 xi0 ei0 p> z and 0 xis eis C dsn zn ; for all s D 1; : : : ; S;
nD1
(4.43)
for some portfolio z 2 RN . The following proposition shows that the existence of
an optimal portfolio for the above problem is actually equivalent to the absence of
arbitrage opportunities, under the additional assumption that the utility function ui
is continuous.
Proposition 4.30 For any i 2 f1; : : : ; Ig, suppose that ui W R2 ! R is continuous
and strictly increasing in both of its arguments. Then, there exists an optimal
portfolio zi 2 RN for Problem (PO4’) in correspondence of the price-dividend
couple . p; D/ if and only if . p; D/ does not admit arbitrage opportunities.
Proof As in Proposition 4.27, the existence of an optimal portfolio for Prob-
lem (PO4’) subject to the budget constraint (4.43) implies that there are no
arbitrage opportunities. Conversely, if the couple . p; D/ does not admit arbitrage
opportunities, it can be shown that agent i’s budget set (identified by the budget
constraint (4.43)) is closed and bounded, i.e., it is a compact set (see LeRoy &
Werner [1191], Theorem 3.6.5). Since every continuous function has a maximum
over a compact set (Weierstrass theorem), this implies the existence of an optimal
portfolio zi 2 RN . t
u
If consumption is not restricted to be non-negative, then the absence of arbitrage
opportunities does not suffice in general to ensure the existence of an optimal
portfolio (see Exercise 4.33).
So far, in the case of concave utility functions, we have focused our attention on
the relation between the absence of arbitrage opportunities and the existence of an
optimal portfolio for an individual agent. Now, we aim at establishing a relation
between the arbitrage free property of a price-dividend couple . p; D/ and the
existence of an economy populated by I risk averse agents such that p corresponds to
the price vector implementing an equilibrium allocation. In other words, assuming
that the couple . p; D/ is arbitrage free, we want to prove the existence of an
equilibrium of an economy such that p D p , where p denotes the equilibrium
price vector of the economy, following the notation of Sect. 4.2. To this end, we rely
on the representative agent technique, as introduced in Sect. 1.3 under certainty and
in the previous section under risk.
Suppose that the price-dividend couple . p; D/ does not admit arbitrage opportu-
nities and consider an economy populated by I agents with strictly increasing and
concave utility functions ui W R2 ! R, for i D 1; : : : ; I, separable with respect to
time and with discount factor ı, i.e.,
ui .x0 ; xs / D ui0 .x0 / C ıui1 .xs /; for i D 1; : : : ; I:
We assume that all the I agents have the same discount factor ı and believe
in the same probability distribution. Furthermore, we assume that the agents’
endowments lie in the asset span, i.e., ei 2 I.D/, for every i D 1; : : : ; I. For a
given vector of weights .a1 ; : : : ; aI / 2 RICC , we define the candidate representative
agent’s utility function as in (4.28)–(4.29). Then, the utility functions u0 and u1
are strictly increasing and concave and we can consider the optimal consumption
Problem (PO4’) for the representative agent’s utility function u. Under a continuity
assumption and in the absence of arbitrage opportunities, Proposition 4.30 can be
applied to the representative agent’s optimal consumption problem and we denote
by .x0 ; x1 ; : : : ; xS / the optimal consumption plan in correspondence of the aggregate
P
endowment .e0 ; e1 ; : : : ; eS / of the economy, with es D IiD1 eis , for s D 0; 1; : : : ; S.
Hence, if the representative agent is endowed with the optimal consumption plan
.x0 ; x1 ; : : : ; xS /, then the price vector p corresponds to the no-trade equilibrium
of the single representative agent economy. The allocation f.xi 0 ; x1 ; : : : ; xS /I i D
i i
1; : : : ; Ig which defines the representative agent’s utility function (4.28) will be a

Pareto optimal allocation for the economy populated by the I agents. In turn, by the
Second Welfare Theorem, the allocation f.xi 0 ; x1 ; : : : ; xS /I i D 1; : : : ; Ig together
i i

with the price vector p D p represents a Radner equilibrium of the economy with
I agents endowed with f.xi 0 ; x1 ; : : : ; xS /I i D 1; : : : ; Ig as initial allocation.
i i
In the above argument, the utility functions ui of the I agents were not assumed
to be of a specific form. Indeed, the above construction of an economy yielding p as
equilibrium price vector is based on taking the agents’ utility functions as given and
endowing the agents with a suitable allocation. As discussed in Sect. 4.3, in general
equilibrium prices depend on the allocation of resources. However, if the agents’
utility functions are of the generalized power, logarithmic or exponential form, then
the aggregation property holds in a strong form and equilibrium prices only depend
on the aggregate resources and not on the distribution of resources among the I
agents. In this case, it suffices to endow the I agents with any feasible allocation such
that .e0 ; e1 ; : : : ; eS / D .x0 ; x1 ; : : : ; xS /, with .x0 ; x1 ; : : : ; xS / denoting the optimal
consumption plan in the representative agent economy supporting the price vector
p. Note also that this result does not need market completeness, but only effective
market completeness, as discussed at the end of Sect. 4.3.
Proposition 4.30 implies that there exists an economy with a single risk averse
agent such that the price vector p is associated to a no-trade equilibrium, thus linking
market viability to the absence of arbitrage opportunities. Moreover, there is a direct
relation between the single agent’s marginal utility of optimal consumption and a
risk neutral probability measure. This is the content of the following proposition.
Proposition 4.31 For any i 2 f1; : : : ; Ig, let ui W R2 ! R be concave and strictly
increasing in both of its arguments and denote by .xi 0 ; x 1 ; : : : ; x S / 2 RC
i i SC1
the
optimal consumption plan of Problem (PO4’) subject to the budget constraint (4.43)
in correspondence of the price-dividend couple . p; D/. Then the following hold:
(i) the price vector p corresponds to the no-trade equilibrium of the economy
populated by the single agent i endowed with .xi 0 ; x1 ; : : : ; xS /;
i i
(ii) if .x0 ; x1 ; : : : ; xS / 2 RCC , then a risk neutral probability measure can be

i i i SC1
defined by letting
s uixs .xi
0 ; xs /
i
s WD PS ; for all s D 1; : : : ; S; (4.44)
rD1 r uxr .x0 ; xr /
i i i
where uixs denotes the partial derivative of the function ui with respect to the
second argument evaluated in correspondence of state !s , for s D 1; : : : ; S and
i D 1; : : : ; I.
Proof (i): this follows directly from the optimality of .xi 0 ; x1 ; : : : ; xS /.
i i
(ii): since .x0 ; x1 ; : : : ; xS / 2 RCC is an interior solution to Problem (PO4’) and ui

i i i SC1
is strictly increasing (so that the budget constraint (4.43) is satisfied as an equality),
the first order optimality conditions imply that
X
S X
S
pn r uix0 .xi
0 ; xr / C
i
r uixr .xi
0 ; xr /drn D 0;
i
for all n D 1; : : : ; N:
rD1 rD1
By the assumptions on the function ui , it holds that uix0 .xi

0 ; xr / > 0, for all r D
i
1; : : : ; S, so that the last condition can be rewritten as
X
S
uixs .xi
0 ; xs /
i
pn D s PS dsn ; for all n D 1; : : : ; N: (4.45)
sD1 rD1 r uix0 .xi
0 ; xr /
i
Similarly to part (iii) of Theorem 4.20, equation (4.45) implies Pthat we can define a
S
state price vector m 2 RSCC by letting ms WD s uixs .xi 0 ; x i
s /=. rD1 r ux0 .x0 ; xr //,
i i i
for all s D 1; : : : ; S. Hence, as in the proof of Proposition 4.22, we can define a risk
neutral probability measure D .1 ; : : : ; S / by letting
ms s uixs .xi
0 ; xs /
i
s WD PS D PS ; for all s D 1; : : : ; S:
rD1 r uxr .x0 ; xr /
i i i
sD1 ms
t
u
In the setting of Proposition 4.31, the implicit risk free rate rf is given by
PS
1 XS
sD1 s uxs .x0 ; xs /
i i i
D ms D PS : (4.46)
sD1 s ux .x0 ; xs /
rf i i i
sD1 0
According to formula (4.44), the risk neutral probability s associated to the state of
the world !s is equal to the ratio between the marginal utility of consumption at time
t D 1 in correspondence of the state of the world !s weighted by s and the expected
value of the marginal utility of consumption. Similarly, in view of equation (4.46),
the risk free rate is given by the ratio between the expected marginal utility of
consumption at time t D 0 and the expected marginal utility of consumption at
time t D 1. Using the expectation operator EŒ, equation (4.45) can be equivalently
rewritten as6

E uixQ .xi
0 ;xQ i /dQ n
pn D ; for all n D 1; : : : ; N; (4.47)
E uix0 .xi0 ;x Q i /
where xQ i denotes the random variable taking .xi 1 ; : : : ; xS / as possible values and
i
Qdn denotes the random dividend of the n-th security (i.e., the n-th column of the
dividend matrix D). The term uixQ .xi
0 ;x
Q i /=EŒuix0 .xi
0 ;x Q i / is often called stochastic
discount factor (or pricing kernel), since it represents the random variable which, if
taken as a discount factor, relates the random payoff dQ n at time t D 1 of a security
with its price pn at time t D 0. Using the definition of covariance together with
equation (4.46), we can equivalently write

EŒdQ n Cov uixQ .xi Q i /; dQ n
0 ;x
pn D C ; for all n D 1; : : : ; N: (4.48)
rf E uix0 .xi
0 ;x
Q i /
The last formula shows that the price at time t D 0 of any traded security can
be decomposed into two terms: a first term representing the expected discounted
dividend (under the statistical probability measure) and a second term represented
by the covariance of the dividend with the stochastic discount factor.
Similarly to the risk neutral valuation formula, also the valuation rule (4.47) can
be applied to payoffs not traded in the market. Indeed, any stochastic discount factor
gives rise to a strictly positive linear pricing functional, which we denote by Qu , in
order to make explicit the dependence on the utility function u. This leads to the
marginal utility pricing rule (or stochastic discount factor pricing rule):

E uixQ .xi
0 ;x
Q i /Qc EŒQc Cov uixQ .xi
0 ;xQ i /; cQ
Q .c/ WD i i i D
u
C ; (4.49)
E ux0 .x0 ; xQ / rf E uix0 .xi0 ;x
Q i /
where we denote equivalently by cQ or by c D .c1 ; : : : ; cS / the random variable

having .c1 ; : : : ; cS / as possible values, representing a generic payoff/contingent
consumption plan. Clearly, Qu is a pricing functional, in the sense of Definition 4.19.
As in the case of the risk neutral valuation formula (4.35), it can be shown that
Qu .c/ is an arbitrage free price. This means that, if the original financial market
with N securities is in equilibrium and is extended to include the payoff c at the
price Qu .c/, then the extended financial market comprising N C 1 securities will be
free of arbitrage opportunities.
6
With some abuse of notation, we denote by uiQx the derivative of the function ui with respect to its
second argument, for i D 1; : : : ; I.
In particular, if the utility function ui W R2 ! R is time separable (with a discount

factor 0 < ı 1), i.e., ui .x0 ; xs / D ui0 .x0 / C ıui1 .xs /, formula (4.44) reduces to
0
s ui1 .xi
s /
s D PS i0 i
; for all s D 1; : : : ; S;
rD1 r u1 .xr /
0
where ui1 denotes the first derivative of the utility function ui1 . In view of equa-
tion (4.46), the risk free rate rf satisfies
1 XS 0
ui .xi /
Dı s 1i0 si ; (4.50)
rf sD1
u0 .x0 /
0
where ui0 denotes the first derivative of the utility function ui0 . Similarly, equa-
tion (4.47) can be rewritten as

i0 i
u .Qx /
pn D ı E 1i0 i dQ n ; for all n D 1; : : : ; N; (4.51)
u0 .x0 /
and relation (4.48) can be consequently rewritten as

i0 i
EŒdQ n u .Qx /
pn D C ı Cov 1i0 i ; dQ n ; for all n D 1; : : : ; N:
rf u0 .x0 /
Analogous representations hold true for the valuation rule (4.49).

Formula (4.44) implies that to the optimal consumption plan .xi 0 ; x1 ; : : : ; xS /
i i
one can associate a likelihood ratio ` D .`1 ; : : : ; `S /, where
uixs .xi ; xi /

`s D i 0i s ; for s D 1; : : : ; S:
E uxQ .x0 ; xQ i /
This representation of the likelihood ratio implies that, for s D 1; : : : ; S, the risk
neutral probability s is greater than the original probability s if and only if
the marginal utility of the optimal consumption in correspondence
of the state of
the world !s is larger than the expectation E uixQ .xi
0 ; Q
x i
/ . In other words, the risk
neutral probability introduced in part (ii) of Proposition 4.31 embeds risk aversion,
in the sense that it assigns a probability greater than the statistical probability to the
states of the world characterized by a level of consumption lower than the average
and vice-versa.
Proposition 4.31 and the following results have been established in equilibrium
considering the individual agents i D 1; : : : ; I. However, when markets are (effec-
tively) complete, the result of Proposition 4.31 and formulae (4.47)–(4.51) can also
be applied to the representative agent economy. Indeed, if markets are (effectively)
complete, then every equilibrium allocation of the economy with I agents is Pareto
optimal and, hence, can be characterized in terms of a representative agent economy,

where the representative agent’s utility function is of the form (4.28)–(4.29). In
particular, in the case of utility functions which are separable with respect to
time and with a discount factor ı, the valuation formula (4.51) with respect to
the representative agent’s no-trade equilibrium in correspondence of the aggregate
endowment .e0 ; e1 ; : : : ; eS / becomes

XS
u01 .Qe/ Q u0 .es /
pn D ı E 0 d n D ı s 01 dsn ; for all n D 1; : : : ; N; (4.52)
u0 .e0 / sD1
u0 .e0 /
where eQ denotes the random variable taking values .e1 ; : : : ; eS /, and the correspond-
ing stochastic discount factor is represented by the random variable taking values

ıu01 .e1 / ıu01 .eS /
;:::; 0 : (4.53)
u00 .e0 / u0 .e0 /
As in the case of the stochastic discount factor pricing rule (4.49), also for-
mula (4.52) can be extended to the valuation of payoffs not traded in the original
financial market. Note that the representative agent’s stochastic discount factor (and,
hence, the associated risk neutral probability measure) is greater in correspondence
of the states of the world where the aggregate endowment is lower and vice-versa,
thus taking into account the agents’ risk aversion. Of course, all the considerations
made after Proposition 4.31 also apply to a representative agent economy.
The valuation formula (4.52) has interesting applications to asset pricing. In
particular, as shown in Exercise 4.36, if the agents’ utility functions are assumed to
be of the generalized power form (with identical discount factors and cautiousness
coefficients) and if the aggregate endowment is lognormally distributed, then one
can derive a version of the celebrated Black-Scholes pricing formula for Call
options.
The Binomial Model
In this section, we present the classical single-period binomial model, which

provides a simple example of an economy with two possible states of the world
and two securities. The binomial model allows to illustrate in a clear way most of
the concepts introduced in the present section and is particularly interesting in view
of its practical applications to derivative pricing. At the same time, the binomial
model already contains most of the key features of more complex stochastic models
widely employed in quantitative finance.
In the binomial model, there are two securities available for trade at the initial
date t D 0: a risk free security, generically called bond, and a risky security,
generically called stock. Let Bt denote the price of the bond and St the price of
the stock at time t, for t 2 f0; 1g. Since the bond represents a risk free asset, the
value B1 is constant across the two possible states of the world. More specifically,
we suppose that B0 is conventionally set equal to 1 and B1 WD rf , where rf > 0
denotes the risk free rate. On the contrary, the stock represents a risky asset, so that
its price S1 at time t D 1 depends on the realization of the state of the world. More
specifically, for some given initial price S0 > 0, we assume that S1 D S0 zQ, where zQ
is a random variable taking values fu; dg, with u > d > 0, with probabilities and
1 , respectively (with > 0). Summing up, the price processes .Bt /tD0;1 and
.St /tD0;1 can be represented as follows:
tD0 tD1

B0 D 1 B1 D rf
S1 D S0 u (4.54)
%
S0
1 &
S1 D S0 d
Equivalently, in terms of the notation previously introduced, we have

1 rf S0 u
pD and DD : (4.55)
S0 rf S0 d
In the binomial model, the financial market is assumed to satisfy the underlying
hypotheses of this section, namely, there are no transaction costs, short sales are
allowed, the risk free rate is the same for lending as well as for borrowing and the
two securities are perfectly liquid in the market. In this context, a portfolio is simply
represented by a vector z 2 R2 , where the first component z1 denotes the number of
units of the bond and z2 the number of units of the stock.
As a preliminary, we show that the market is complete and provide a necessary
and sufficient condition for the absence of arbitrage opportunities.
Proposition 4.32 In the single-period binomial model (4.55) the market is com-
plete. Moreover, there are no arbitrage opportunities if and only if d < rf < u.
Proof Since u > d and rf > 0, it is easy to see that the matrix D is invertible, so
that I.D/ D R2 , meaning that the market is complete. Due to Theorem 4.20, there
are no arbitrage opportunities if and only if there exists a vector m 2 R2CC such that
p D D> m. Since the matrix D is invertible it holds that m D .D> /1 p, i.e.,
rf d
!
1 S0 d rf 1 rf .ud/
mD D urf :
rf S0 .u d/ S0 u rf S0 rf .ud/
Under the standing assumptions that rf > 0 and u > d, it is immediate to see that
m 2 R2CC if and only if d < rf < u. t
u
In view of Proposition 4.22, there exists a (unique) risk neutral probability
measure if and only if d < rf < u, as shown in the following corollary. Note
that, since the market is complete, if a risk neutral probability measure exists, then
it is necessarily unique (see Proposition 4.26).
Corollary 4.33 There exists a risk neutral probability measure . ; 1 / if and
only if d < rf < u, in which case it holds that
rf d u rf
D and 1 D :
ud ud
Proof As in the proof of Proposition 4.22, it suffices to let WD m1 =.m1 C m2 /

and rely on the proof of Proposition 4.32. t
u
The no-arbitrage condition d < rf < u admits an intuitive interpretation. Indeed,
the failure of one of these two inequalities means that one of the two securities
dominates the other, in the sense that it has a greater return in each possible state
of the world. In this case, it would be possible to create an arbitrage opportunity
by investing in the more profitable security and selling short the other. For instance,
if d rf , then it would be possible to realize an arbitrage opportunity with the
portfolio z D .S0 ; 1/. At time t D 0, the value of such a portfolio is zero, while
at time t D 1 the dividend will be either S0 .u rf / > 0 or S0 .d rf / 0 in
correspondence of the two possible states of the world, thus yielding an arbitrage
opportunity. The case u rf can be treated in a similar way.
Let us now consider the problem of valuing a derivative contract in the context
of the binomial model. A first and simple approach to the valuation of financial
derivatives consists in the application of the risk neutral valuation formula (4.36). In
particular, due to Corollary 4.33, if the payoff of the derivative is represented by a
vector c D .cu ; cd / 2 R2 , then the price at time t D 0 of the derivative, denoted by
Q.c/, can be computed as

1 1 rf d u rf
Q.c/ D E Œc D cu C cd : (4.56)
rf rf ud ud
In the binomial model, since markets are complete, the valuation formula (4.56)
can also be obtained by means of two alternative procedures, namely the value
at time t D 0 of the portfolio replicating the derivative and the Delta-hedging
strategy. Let us first consider the replicating portfolio approach. As we have seen in
Proposition 4.32, the market is complete and, hence, for any contingent consumption
plan c 2 R2 there exists a replicating portfolio zc 2 R2 such that c D Dzc . More
specifically, we have the following proposition.
Proposition 4.34 Let c D .cu ; cd / 2 R2 represent the payoff of a financial contract.

Then there exists a unique portfolio zc 2 R2 such that Dzc D c, explicitly given by
!
1 ucd dcu
z D
c rf ud
1 cu cd :
S0 ud
Moreover, the value at time t D 0 of the portfolio zc 2 R2 is given by (4.56).

Proof Since markets are complete, it holds that
!
1 ucd dcu
1 1 S0 d S0 u cu
z DD cD
c
D rf ud
1 cu cd :
rf S0 .u d/ rf rf cd S0 ud
It is then immediate to verify that p> zc coincides with the right-hand side of (4.56).
t
u
According to the above proposition, the unique arbitrage free price of a payoff
c 2 R2 is given by the market value p> zc of the replicating portfolio. As discussed
before, any other price for the payoff c would introduce arbitrage opportunities in
the extended financial market composed of the two original securities (i.e., the bond
and the stock) together with the additional payoff c.
The value of a payoff c 2 R2 can also be obtained in terms of the so-called
Delta hedging strategy. The Delta hedging technique consists in determining the
portfolio, composed of one unit of the contract delivering the payoff c and units
of the underlying stock, such that the portfolio’s dividend at time t D 1 is riskless,
in the sense that it does not depend on which of the two states of the world will be
realized at time t D 1. In that sense, the position in the payoff c is hedged by the
position in the underlying security. Let us denote by Vt . / the value of such a
portfolio at time t, for t 2 f0; 1g.
Proposition 4.35 Let c D .cu ; cd / 2 R2 represent the payoff of a financial contract
and let Vt . /, for t 2 f0; 1g, be defined as above. Then V1 . / does not depend on
the state of the world if and only if
cu cd
D : (4.57)
S0 .u d/
Proof By definition, the quantity is such that the value of the portfolio V1 . /
does not depend on the state of the world, i.e.,
cu S0 u D cd S0 d:
This directly implies (4.57), recalling that u > d and S0 > 0. t

u
Formula (4.57) also explains the reason of the terminology Delta hedging.
Indeed, the strategy , which represents the number of units of the underlying asset
needed to compensate the riskiness of the random payoff c, is given by the ratio
between the difference of the derivative’s payoff and the difference of the dividend
of the underlying stock in correspondence of the two possible states of the world.
Since the portfolio constructed in Proposition 4.35 is riskless, if there are no
arbitrage opportunities (in the market composed by the bond, the stock and the
derivative), then its rate of return has to be equal to the risk free rate of return rf ,
i.e.,

Q.c/ C S0 rf D cu C S0 u;
where Q.c/ denotes the arbitrage free price at time t D 0 of the financial contract
having payoff c. The above formula implies that
cu S0 u 1 ucd dcu cu cd
Q.c/ D C S0 D C :
rf rf u d ud
As can be easily seen, this last expression coincides with the right-hand side
of (4.56).
The Modigliani-Miller Theorem
By Theorem 4.20, the absence of arbitrage opportunities is equivalent to the

existence of a (possibly non-unique) strictly positive linear pricing functional. In
particular, the linearity of the pricing functional implies that, in the absence of
market frictions, the value of a firm is independent of its capital structure. This
result is known as the Modigliani-Miller theorem and goes back to Modigliani &
Miller [1348, 1349].
In order to state the main result of Modigliani & Miller [1348], let us consider a
two-period arbitrage free market and a firm starting its activities at the initial date
t D 0 and terminating at t D 1. We denote by V1 the revenues generated by the firm
at date t D 1. We shall consider two capital structures:
(a) an unleveraged firm, financing its activities only by issuing equity;
(b) a leveraged firm, financing its activities by issuing equity and debt.
We assume that the equity of the firm is composed of shares which are traded in the
market and, similarly, that the debt is composed of traded bonds, with a total nominal
value equal to K > 0. We assume that there are no transaction costs, constraints on
trading, taxes or any other market friction.
In case (a), we denote by V0un the total value of the firm at the initial date 0 and by
St the market value of the firm stock at date t, for t 2 f0; 1g. In case (b), we denote
un
by V0lv the value of the leveraged firm at t D 0 and by Bt and Stlv the value of the firm
debt and stock, respectively, at date t, for t 2 f0; 1g. As in the preceding sections of
this chapter, we assume that there are S possible states of the world .!1 ; : : : ; !S /.
Let us first consider case (a). In this case, the revenues of the firm at date t D 1
are fully distributed among the equity holders, so that the total market value of the
firm stock is simply equal to the total firm value, i.e.,
S1un .!s / D V1un .!s /; for all s D 1; : : : ; S:
Since the firm stock is traded in the market and the market is assumed to be arbitrage
free, Theorem 4.20 implies that there exists a state price vector m 2 RSCC such that
X
S X
S
V0un D S0un D ms S1un .!s / D ms V1 .!s /: (4.58)
sD1 sD1
Let us now consider case (b). In the case of a leveraged firm, with a debt of
nominal value K, we assume that equity holders have limited liability. This means
that, at date t D 1, the amount K has to be repaid to the debt holders if the firm
revenues exceed K, while, if the firm revenues do not suffice to repay the liability
K, then the debt holders only receive the liquidation value of the firm, due to the
limited liability of the equity holders. In other words, it holds that
B1 .!s / D minfV1 .!s /I Kg and S1 .!s / D maxfV1 .!s / KI 0g;
for all s D 1; : : : ; S. The value of the firm debt at the initial date t D 0 is then given
by
X
S X
S X
S

B0 D ms B1 .!s / D ms minfV1 .!s /I Kg D ms K maxfK V1 .!s /I 0g
sD1 sD1 sD1
(4.59)
and the value of the stock of the leveraged firm at t D 0 is given by
X
S X
S
S0lv D ms S1lv .!s / D ms maxfV1 .!s / KI 0g: (4.60)
sD1 sD1
In view of (4.60), the market value of the stock of a leveraged firm can be
represented as a Call option written on the firm assets, with the exercise price being
the nominal value of the debt. In an analogous way, by (4.59), the market value of
the debt can be expressed as the difference between its nominal value and the value
of a Put option written on the firm assets, with strike price K.
The Modigliani-Miller theorem states that the financial structure of a firm does
not affect the total firm value. In view of above relations (4.58), (4.59) and (4.60),
we can easily verify this claim. Indeed, it holds that
X
S X
S
V0lv D S0lv C B0 D ms maxfV1 .!s / KI 0g C ms minfV1 .!s /I Kg
sD1 sD1
(4.61)
X
S
D ms V1 .!s / D V0un :
sD1
The Modigliani-Miller relation (4.61) does not rely on market completeness, but
only on the absence of arbitrage opportunities. Indeed, under the assumption that
both the equity and the debt are traded, their market values at the initial date t D 0
do not depend on the specific choice of the state price vector m 2 RSCC . However,
if a firm modifies its financial structure by issuing non-traded securities which are
not spanned by the available assets, then the result of Modigliani-Miller does not
necessarily hold. As shown in Exercise 4.37, the presence of taxation leads to a
modification of the Modigliani-Miller result and it turns out that the value of a firm
is increasing with respect to the value of the debt.
As a consequence of relation (4.61), it holds that the cost of capital of a leveraged
firm is equal to the cost of capital of an unleveraged firm plus a spread which
depends on the debt-equity ratio of the leveraged firm (see Modigliani & Miller
[1348, Proposition II]). Indeed, let us define the following quantities:
rd WD EŒB1 =B0 ; rlv WD EŒS1lv =S0lv and run WD EŒS1un =S0un D EŒV1 =V0un :
(4.62)
In particular, run and rlv represent the expected rates of return on the equity for an
unleveraged firm and for a leveraged firm, respectively. By relying on the identity
S0lv D V0lv B0 D V0un B0 ;
which is a consequence of (4.61), we can write
EŒV1 B1 run V0un rd B0

rlv D D
S0lv S0lv
(4.63)
run .S0lv C B0 / rd B0 B0
D lv
D run C .run rd / lv :
S0 S0
This shows that the cost of capital for a leveraged firm can be decomposed into the
sum of the cost of capital for an unleveraged firm and a second component given
by the debt-to-equity ratio of the leveraged firm multiplied by the spread run rd .
A connection between the Modigliani-Miller result and the CAPM model will be
presented in Exercise 5.11 in the following chapter.
For more detailed expositions of general equilibrium theory under risk we refer the
reader to Mas-Colell et al. [1310], Dothan [581], Magill & Shafer [1288], Dana &
Jeanblanc [514], Magill & Quinzii [1286]. Radner [1439, p.932] clearly defines the
concept of Radner equilibrium: “agents have common expectations if they associate
the same (future) prices to the same events . . . . I shall say that the plans of the
agents are consistent if, for each commodity, each date, and each event at that
date, the planned supply of that commodity at that date in that event equals the
planned demand, and if a corresponding condition holds for the stock markets.
An equilibrium of plans, prices, and price expectations is a set of prices on the
current market, a set of common expectations for the future, and a consistent set
of individual plans, one for each agent, such that, given the current prices and
price-expectations, each individual agent’s plan is optimal for him, subject to an
appropriate sequence of budget constraints.” It is pointed out that “Traders need not
agree on the probabilities of future environmental events, and therefore they need
not agree on the probability distribution of future prices, but they must agree on
which future prices are associated with which events”, see Radner [1439, p.940].
On the mutuality principle and risk sharing, the seminal paper is Wilson [1661].
Townsend [1599] performed an empirical analysis of the mutuality principle using
micro data from a village in India. He showed that only a small part of income
variation is due to aggregate risk and income fluctuations are largely idiosyncratic.
Although financial markets are not developed, there is evidence of mutual insurance.
The analysis presented in this chapter concerns economies with a finite number
of states of the world. The analysis with infinitely many states requires more refined
tools and we refer to the special issue of the Journal of Mathematical Economics
(1996) on this topic. On the role of options in achieving market completeness in
an economy with infinitely many states see Green & Jarrow [822], Brown & Ross
[311], Nachman [1365].
It is possible to extend the analysis developed in Sect. 4.2 to assets delivering
dividends defined in terms of L > 1 goods (real assets), see Magill & Shafer [1288]
and Geanakoplos [763]. In this case, we have that D 2 RLSN . Let Ds be the
L N matrix obtained by specifying D in the state of the world !s , for each s D
1; : : : ; S. If the economy is regular (i.e., for each state of the world !s we can extract
a row ds from Ds so that the vectors .d1 ; : : : ; dS / are linearly independent), then
the existence of a Radner equilibrium can be proved and the equivalence between
Radner equilibrium allocations and Arrow-Debreu equilibrium allocations can be
established for an open set of economies of full measure (generic existence result),
see Magill & Shafer [1287]. To this end, N S. When this equivalence holds true,
an equilibrium allocation is Pareto optimal. Also the invariance result with respect
to the financial structure is a generic result. If the economy is not regular, then
the existence of the Radner equilibrium and its Pareto non-optimality are generic
results, see Duffie & Shafer [598], Geanakoplos [763], Magill & Shafer [1288].
A non-existence example with assets paying bundles of goods has been provided
in Hart [906]. In proving the existence of the equilibrium, problems arise because
the budget correspondences are not always upper semicontinuous. To overcome this
problem, constraints on agents’ trades in asset markets can be introduced (e.g., no
short sales), see Radner [1437].
If nominal assets are traded in the market, then the existence of a Radner
equilibrium can be established in a complete or incomplete market without short sale
constraints under standard regularity conditions, see Cass [375] and Werner [1653].
In Balasko & Cass [114] it is shown that if markets are incomplete and nominal
returns are exogenous, then there are multiple equilibria with real indeterminacy
of dimension S N (with N being the number of linearly independent assets). In
Geanakoplos & Mas-Colell [764] it is shown that the degree of indeterminacy is
S 1 if the current prices of the assets are endogenous to the model, whatever
the number of assets. If markets are complete, then the equilibrium allocation is
Pareto optimal and the equilibrium is determinate. If markets are incomplete, then
generically the equilibrium allocation is not Pareto optimal. An example presented
in Hart [906] shows that in incomplete markets with many goods the allocation
associated with a Radner equilibrium is not necessarily constrained ex-ante Pareto
optimal and equilibria can be Pareto ordered (see Geanakoplos & Polemarchakis
[765] and Magill & Shafer [1288] for constrained generic inefficiency results). On
these topics see also Magill & Shafer [1288].
For a generic existence result of a Radner equilibrium in a financial market with
financial derivatives see Krasa [1130] and Krasa & Werner [1131]. In our analysis,
the financial structure is assumed to be exogenous. However, this represents a quite
strong assumption in the context of an incomplete market. Market incompleteness
means that there is an insurance demand which is not satisfied by the market and
that a Pareto improvement can be obtained by introducing enough securities. In this
setting, does an incentive emerge to complete the market through financial inno-
vation? The enormous financial innovation observed in the last decades suggests a
positive answer. However, this is not always the case. An enlightening example has
been provided in Hart [906], showing that the effect of opening new markets is to
make everybody worse off. A detailed analysis of this topic lies outside the scope of
this book and we refer the reader to Allen & Gale [46] and Duffie & Rahi [597]. The
analysis of financial innovation requires in the first place to identify the proposer of
the new contract: in the literature it is assumed that the new contract is proposed
by a firm issuing new securities or by an agent aiming to gain by trading the new
asset. Note that the results provided in the literature are extremely sensitive to the
hypotheses of the model. In many cases, financial innovation does not complete
the market and does not induce a Pareto improvement. Under some conditions, the
equilibrium allocation turns out to be constrained Pareto optimal. Note also that an
equilibrium may fail to exist in a general financial market where financial derivatives
are traded, see for instance Polemarchakis & Ku [1424]. In Hart [906], the author
provides an example of an incomplete market economy with multiple equilibria
ranked according to the Pareto order.
On the aggregation property and mutual funds separation in incomplete markets

see Milne [1346], Detemple & Gottardi [563], Hens & Pilgrim [937]. Note that, in
the special case of exponential utility functions, aggregation in a strong sense can
also be established with heterogeneous beliefs and heterogeneous time preferences,
as long as all agents agree on the set of events with zero probability, see Huang &
Litzenberger [971, Section 5.26].
An exhaustive analysis of the Fundamental Theorem of Asset Pricing is provided
in Connor [476], Dothan [581], Dybvig & Ross [611], Duffie [593] and for an
overview of the history and the role of this result in the context of mathematical
finance we refer to Schachermayer [1501]. In the context of continuous time
models, a general formulation of the fundamental theorem of asset pricing has been
obtained in Delbaen & Schachermayer [544, 545]. We also mention that, always in
a continuous time setting, an asset pricing theory has been developed in Platen &
Heath [1420] by relying on a weaker notion of arbitrage (see also Fernholz [683]
for a descriptive theory of financial markets based on a weaker notion of arbitrage).
See also Fontana [724] for a unifying analysis of several no-arbitrage conditions.
The theorem has been extended to an economy with states characterized by null
probability in Willard & Dybvig [1660]. An arbitrage opportunity is strictly related
to the existence of a couple of consumption plans ordered according to the first
order stochastic dominance criterion. In particular, an arbitrage opportunity implies
the existence of a couple of consumption plans ordered according to the criterion,
but not vice-versa, see Jarrow [1018]. Note that one can also derive a version of
the Fundamental Theorem of Asset Pricing by considering the absence of arbitrage
opportunities of the second kind. In that case, one obtains the existence of a positive
(but not necessarily strictly positive) linear pricing functional or, equivalently, of a
solution m 2 RSC to the system p D D> m (see Theorem 5.2.2 of LeRoy & Werner
[1191] and compare also with Exercise 4.17). The capital structure discussed at the
end of Sect. 4.4 in the context of the Modigliani-Miller theorem is also at the basis
of the seminal model considered in Merton [1331] for the pricing of corporate debt.
4.6 Exercises
Exercise 4.1 Let us consider an economy with two agents a; b, two possible states
of the world with strictly positive probabilities .; 1 / and a single consumption
good. Suppose that the endowments of agents a and b satisfy ea1 C eb1 > ea2 C eb2 , so
that there is aggregate risk. Show that the prices p1 ; p2 of the contingent goods in
correspondence of an interior Pareto optimal allocation satisfy p1 =p2 < =.1 /.
Exercise 4.2 Consider the setting of Sect. 4.1 in the presence of a single good
(i.e., L D 1). Show that condition (4.1) characterizing an ex-ante Pareto optimal
allocation is equivalent to the Borch condition (4.4).
4.6 Exercises 193
Exercise 4.3 Let fyi W RC ! RC I i D 1; : : : ; Ig be a Pareto optimal sharing rule.

Show that yi is linear if and only if

tu0 i yi .e/ D tu0 k yk .e/ ; for all i; k D 1; : : : ; I;
where tu0 i . yi .e// denotes the first derivative of the risk tolerance of agent i computed
in correspondence of the Pareto optimal consumption allocation yi .e/, where e
denotes an arbitrary realization of the aggregate endowment.
Exercise 4.4 Prove the necessity part of Proposition 4.5.
Exercise 4.5 Consider utility functions of time 1 consumption of the generalized
power utility form
1 b1
ui .x/ D .i C bx/ b ; with b … f0; 1g; for all i D 1; : : : ; I;
b1
where agent i has discount factor ıi , for i D 1; : : : ; I. By relying on condition (4.4),

show that the Pareto optimal sharing rule is linear with respect to the aggregate
endowment. In an analogous way, prove the same result in the case of exponential
utility functions ui .x/ D i exp.x=i /, for all i D 1; : : : ; I.
Exercise 4.6 Consider an economy with L S future markets open at time t D 0,
for every contingent good .l; s/, for all l D 1; : : : ; L and s D 1; : : : ; S, as considered
at the beginning of Sect. 4.2. Prove that, in correspondence of an Arrow-Debreu
equilibrium allocation, there cannot be incentives to trade at date t D 1 after the
state of the world has been revealed.
Exercise 4.7 Suppose that markets are complete and let
X
S
ui .x0 ; x1 ; : : : ; xS / D i exp.x0 =i / i ı s exp.xs =i /;
sD1
for all i D 1; : : : ; I. Show that the equilibrium prices q D .q1 ; : : : ; qS / of the S
Arrow securities are given by
P
ıs exp.es = IiD1 i /
qs D P ; for all s D 1; : : : ; S:
exp.e0 = IiD1 i /
Exercise 4.8 Consider an economy with two traded assets and tree possible states
of the world, where the dividend matrix D is given by
2 3
1 0
D D 40 15 :
0 1
Suppose that there are two agents (i.e., I D 2), whose utility functions are differen-
tiable, strictly increasing and strictly concave and only depend on consumption at
time t D 1. The probabilities associated to the three states of the world are given by
.1=4; 1=4; 1=2/. The endowment of the first agent is given by one unit of the first
asset, while that of the second agent is given by one unit of the second asset. Show
that the Pareto optimal allocation is given by the consumption plan .1=4; 1=4; 1=4/
for the first agent and .3=4; 3=4; 3=4/ for the second agent (compare also with
LeRoy & Werner [1191], Example 16.5.1).
Exercise 4.9 Consider an economy with two possible states of the world and two
agents a and b, with homogeneous beliefs .; p 1 / about the realization of the
state of the world and utility functions u.x/ D x defined on consumption at time
t D 1. Let p1 and p2 denote the prices of the two Arrow securities paying one unit of
the consumption good in correspondence of each state of the world and let .ei1 ; ei2 /
denote the endowment of agent i, for i 2 fa; bg, in terms of the two contingent
consumption goods.
(i) Determine the equilibrium allocation when the agents’ endowments are given
by .ea1 ; ea2 / D .40; 60/ and .eb1 ; eb2 / D .60; 40/.
(ii) Determine the equilibrium allocation when the agents’ endowments are given
by .ea1 ; ea2 / D .40; 60/ and .eb1 ; eb2 / D .50; 50/.
(iii) How does the equilibrium allocation change if the agents have heterogeneous
beliefs such that a > b ?
(iv) Determine the equilibrium allocation in the case of a logarithmic utility
function u.x/ D log.x/.
Exercise 4.10 Consider an economy with two possible states of the world and
two agents a and b. Assume that both agents are characterized by a logarithmic
utility function u.x/ D log.x/ defined on consumption at time t D 1. Let p1 and
p2 denote the equilibrium prices of the two Arrow securities and suppose that the
initial endowments of the agents are given by .ei1 ; ei2 / D ˛ i .e1 ; e2 /, for i 2 fa; bg,
for some couple .˛ a ; ˛ b / satisfying ˛ a C ˛ b D 1 and where .e1 ; e2 / denotes the
economy’s aggregate endowment. Show that, if the two agents have homogeneous
beliefs .; 1 / about the possible realization of the state of the world, then they
will not trade in equilibrium.
Exercise 4.11 Consider an economy with two possible states of the world, with
associated probabilities of occurrence .1=2; 1=2/, and a representative agent with
expected logarithmic utility function of the form
1 1
u.x0 ; x1 ; x2 / D log.x0 / C log.x1 / C log.x2 /:
2 2
In the economy it is possible to trade the two Arrow securities and a risk free asset
with constant unitary payoff. Determine the equilibrium prices .q1 ; q2 / of the two
Arrow securities and the return rf of the risk free asset in the representative agent
economy when the aggregate endowment is given by .e0 ; e1 ; e2 / D .1; 3; 1/.
4.6 Exercises 195
Exercise 4.12 Consider the same representative agent economy described in Exer-
cise 4.11, but with the expected utility function
1 1
u.x0 ; x1 ; x2 / D x0 C x1 C x2 ;
2 2
with 0 < < 1. Determine the equilibrium prices .q1 ; q2 / of the two Arrow
securities as well as the return rf of the risk free asset with constant payoff 1.
Exercise 4.13 Consider the optimal consumption Problem (PO3) for an agent
characterized by a time additive state independent utility function of the generalized
1 b1
power form u.x/ D b1 . C bx/ b , with b … f0; 1g, and discount factor ı. Letting
PS
eN WD e0 C sD1 ps es (i.e., the present value of the endowment, where p1 ; : : : ; pS
denote the prices of the S Arrow securities) and x0 the optimal consumption at t D 0,
the quantity eN x0 represents intertemporal saving. Show that intertemporal saving
is an affine function of eN (compare also with Lengwiler [1182, Section 5.4.1]).
Exercise 4.14 Consider an economy with two possible states of the world, with
associated probabilities of occurrence .1=2; 1=2/, and a representative agent with
logarithmic utility function. In the economy there are two traded assets: the first
asset, with price p1 D 1, delivers a risk free payoff of 1 in correspondence of both
states of the world, while the second asset, with price p2 , delivers the random payoff
.1=2; 2/. Determine the equilibrium price p2 of the second asset when the economy’s
aggregate endowment is given by .e1 ; e2 /.
Exercise 4.15
(i) Consider an economy with two possible states of the world and two securities
(i.e., S D N D 2), with

3 2 1
DD and pD :
3 2 1
Show that there exists an arbitrage opportunity of the second kind but there are
no arbitrage opportunities of the first kind.
(ii) Consider an economy with two possible states of the world and two securities
(i.e., S D N D 2), with

32 1
DD and pD :
33 1
Show that there exists an arbitrage opportunity of the first kind but there are no
arbitrage opportunities of the second kind.
(iii) Show that, if there exists a portfolio zN such that DNz > 0 with p> zN > 0, then the
existence of an arbitrage opportunity of the second kind implies the existence
of an arbitrage opportunity of the first kind.
Exercise 4.16 Consider the same economy described in part (ii) of Exercise 4.15.
Show that the Law of One Price holds (but, as shown in Exercise 4.15, there are
arbitrage opportunities of the first kind).
Exercise 4.17 For a given price-dividend couple . p; D/, prove that, if there exists a
positive (but not necessarily strictly positive) linear pricing functional Q, then there
are no arbitrage opportunities of the second kind.
Exercise 4.18 Show that, if the Law of One Price fails to hold, then every payoff
c 2 I.D/ can be replicated with an arbitrarily small initial wealth.
Exercise 4.19 Let z; z0 2 RN be two portfolios with riskless payoffs, i.e.,
X
N X
N
dsn zn D c and dsn z0n D c0 ;
nD1 nD1
for some c; c0 > 0, for all s D; 1 : : : ; S. Show that, if there are no arbitrage
0
opportunities, then p>c z D p>c z0 (i.e., the two portfolios z and z0 have the same rate
of return).
Exercise 4.20 Let c D .c1 ; : : : ; cS / 2 RS represent a random payoff and suppose
that there exists a portfolio zc 2 RN such that Dzc D c. Suppose that it is possible
to trade the payoff c for a price pNC1 ¤ p> zc . Show that there exists an arbitrage
opportunity in the extended market represented by
0 1
p1 2 3
B :: C d11 : : : d1N c1
B C 6 :: 7 :
p0 D B : C D0 D 4 ::: : : : ::
: :5
@ pN A
dS1 : : : dSN cS
pNC1
Exercise 4.21 Prove Proposition 4.24.

Exercise 4.22 Suppose that there are no arbitrage opportunities. Prove that Defini-
tion 4.9 is well-posed, in the sense that, for any c 2 I.D/, the market value V.c/ is
uniquely defined.
Exercise 4.23 Let c D .c1 ; : : : ; cS / 2 RS represent a random payoff which is not
i.e., c … I.D/.
attainable in the market, Suppose that it is possible to trade the payoff
c for a price pNC1 … ql .c/; qu .c/ . Show that there exists an arbitrage opportunity
in the extended market represented by
0 1
p1 2 3
B :: C d11 : : : d1N c1
B C 6 :: : : :: :: 7 :
p0 D B : C 0
D D4 : : : :5
@ pN A
dS1 : : : dSN cS
pNC1
4.6 Exercises 197
Exercise 4.24 Consider an economy with two possible states of the world and two
securities (i.e., S D N D 2), as in Exercise 4.15, with

32 1
DD and pD :
33 1
Show that there are arbitrage opportunities but, nevertheless, the market is complete.
Exercise 4.25 Consider an economy with S D N D 3, with the following dividend
matrix:
2 3
143
D D 46 2 45 :
235
(i) Show that the market is complete.

(ii) Given the price vector p> D .2:15; 2:7; 3:35/, determine the state price vector.
(iii) Determine the risk free rate of return rf implicit in the couple . p; D/ and the
portfolio which attains the riskless payoff .1; 1; 1/. Verify that the return of
such a portfolio coincides with rf .
(iv) Determine the portfolio zc which replicates the payoff c D .2; 3; 6/. Verify that
p> zc D Q.c/ D m> c.
(v) Determine the portfolio z13 which replicates the Arrow security 13 , with payoff
.0; 0; 1/, and verify that p> z13 D m3 .
Exercise 4.26 Consider an economy with a single traded asset and three possible
states of the world (i.e., N D 1 and S D 3), with dividend D D .0:5; 1; 2/> and
price p D 1. For the payoff c D .1; 2; 3/>, determine the super-replication and
sub-replication prices qu .c/ and ql .c/.
Exercise 4.27 Consider an economy with two traded assets and three possible
states of the world (i.e., N D 2 and S D 3), with price-dividend couple
2 3
15
2:45
D D 44 25 and pD :
2:35
31
(i) Characterize the set of possible state prices.

(ii) Determine the space of attainable payoffs.
(iii) Verify that the arbitrage free price of any attainable payoff does not depend on
the specific state price vector chosen.
(iv) Determine the interval of arbitrage free prices for the payoff c D .2; 1; 3/> .
(v) Consider a Call option with strike price 2 written on the second security. Does
the introduction of such a derivative make the market complete?
Exercise 4.28 Consider an economy with three possible states of the world and
three traded securities (i.e., S D N D 3), with price-dividend couple
2 3 0 1
2 3 2 6
D D 42 3 55 and @4A :
5 4=3 3 k
(i) Determine the values of k for which there are no arbitrage opportunities in the
market.
(ii) Determine the range of arbitrage free prices for a Call option written on the
first security with strike price 3.
(iii) For a fixed value of k such that arbitrage opportunities are present in the market,
construct a portfolio which is an arbitrage opportunity.
Exercise 4.29 Consider an economy with three possible states of the world and
three traded securities (i.e., S D N D 3), with dividend matrix given by
2 3
341
D D 42 2 45 :
452
(i) Verify that the market is complete.

(ii) Given the price vector p D .2:85; 3:45; 2:35/>, determine the state price
vector.
(iii) Determine the risk free rate rf implicit in the price-dividend couple . p; D/.
(iv) Determine the portfolio zc which replicates the payoff c D .1; 4; 3/> and verify
that the value at time t D 0 of such a portfolio coincides with the arbitrage free
price of the payoff c computed in terms of the state price vector m.
(v) Determine the arbitrage free price of a Call option written on the second
security with strike price 2. Verify that the value at time t D 0 of the replicating
portfolio zcall coincides with the arbitrage free price of the Call option.
Exercise 4.30 In the setting of Proposition 4.27, show that, if, for any i D 1; : : : ; I,
the utility function ui W R2 ! R is strictly increasing in the first argument and
non-decreasing in the second argument, then the existence of an optimal portfolio
excludes the existence of arbitrage opportunities of the second kind.
Exercise 4.31 In the setting of Proposition 4.27, show that, if, for any i D 1; : : : ; I,
the utility function ui W R2 ! R is non-decreasing in the first argument and strictly
increasing in the second argument and there exists a portfolio zO 2 RN such that
DOz > 0, then the existence of an optimal portfolio excludes the existence of arbitrage
opportunities.
4.6 Exercises 199
Exercise 4.32 In the setting of Proposition 4.27, show that:

(i) if the utility function ui is strictly increasing in its first argument and non-
decreasing in the second, then the existence of an optimal portfolio implies
the validity of the Law of One Price;
(ii) if the utility function ui is non-decreasing in its first argument and strictly
increasing in the second and there exists a portfolio zO with DOz > 0, then the
existence of an optimal portfolio implies the validity of the Law of One Price.
Exercise 4.33 Consider an economy with two possible states of the world (with
equal probabilities of occurrence) and a single traded asset with payoff .1; 1/ and
price p 2 .0; 1/. Consider Problem (PO4) with the linear expected utility function
u.x0 ; x1 ; x2 / D x0 C x1 =2 C x2 =2. Show that there are no arbitrage opportunities but,
nevertheless, there does not exist an optimal portfolio.
Exercise 4.34 Consider an economy with S possible states of the world, such that
e1 : : : eS , and such that the price-dividend couple . p; D/ is arbitrage free and
complete. In correspondence of an equilibrium allocation of an economy populated
by risk averse agents, the price q1 of the first Arrow security is greater or smaller
than 1=rf ?
Exercise 4.35 Show that in a complete market with no aggregate risk, homoge-
neous beliefs and a risk free rate equal to one, the price at time t D 0 of any security
is given by the expectation (with respect to the original probability measure ) of
the dividend at time t D 1.
Exercise 4.36 (See Huang & Litzenberger [971], Section 6.10) Consider an
economy with N traded securities (among which a risk free asset), I agents
with
PI utility functions of the generalized power form, as in Proposition 4.15, with
iD1 i D 0 and identical cautiousness coefficient b, and suppose that all the agents
are only endowed with units of traded securities. Consider a Call option written on
the n-th asset, with strike price k, and denote by pcall;n its price at time t D 0. Suppose
that
1b ! !! 2 !
eQ dQ n e e e n
log ı ; log N I ;
e0 pn n e n n2
where the random variable eQ denotes the random aggregate endowment at t D 1

and dQ n denotes the random dividend of the n-th security, for some correlation
coefficient .
(i) Prove that the equilibrium of the economy can be characterized in terms of the
no-trade equilibrium of a single representative agent with a utility function of
the generalized power form.
(ii) Show that

" 1b #
eQ
p call;n
D ı E maxfdQ n kI 0g : (4.64)
e0
(iii) Show that formula (4.64) admits the explicit representation
k
pcall;n D pn N. 1 / N. 2 /; (4.65)
rf
where
pn
log C log.rf / n
1 WD k
C and 2 WD 1 n
n 2
Rx 2 =2
and N.x/ WD p1 ez dz.
2 1
Exercise 4.37 As at the end of Sect. 4.4, consider an arbitrage free financial market
with a firm operating in two dates t 2 f0; 1g. Denote by V1 the revenues of the firm
at date t D 1 and suppose that the net firm profits are taxed, in the sense that at date
t D 1 the following tax has to be payed:
T1 .!s / WD .V1 .!s / K/C ;
for any s D 1; : : : ; S, where > 0 represents a tax rate and K 0 is the nominal
value of the firm debt. In particular, if K D 0 (unleveraged firm), then the tax to be
payed at t D 1 is simply equal to V1 .!s /, for s D 1; : : : ; S. We denote by V0un;tax the
value at date t D 0 of an unleveraged firm subject to taxation. On the other hand, if
K > 0 (leveraged firm), then we denote by V0lv;tax the value at t D 0 of a leveraged
firm and by Bt the market value of the debt at date t, for t 2 f0; 1g. Assume that both
the firm equity and debt are traded in the financial market. Prove that the value at
date t D 0 of a leveraged firm subject to taxation is given by
V0lv;tax D V0un;tax C B0 ; (4.66)
where V0un;tax denotes the market value at t D 0 of an unleveraged firm subject to

taxation. In particular, the value of a leveraged firm subject to taxation is increasing
with respect to the market value of the debt.
Chapter 5
Factor Asset Pricing Models: CAPM and APT
If stocks are priced rationally, systematic differences in average

returns are due to differences in risk.
Fama & French (1995)
In Chap. 4, we have discussed the role of financial markets in allocating risk among
agents and we have analysed asset prices and returns by means of two different but
related approaches (see Sect. 4.4): equilibrium analysis and no-arbitrage valuation.
In the present chapter, we will exploit these two approaches to extract information
on asset risk premia. In particular, we will present two models: the Capital Asset
Pricing Model (CAPM) and the Arbitrage Pricing Theory (APT). The CAPM is
based on equilibrium analysis, while the APT, as the name suggests, on the absence
of (asymptotic) arbitrage opportunities. The key feature of both the CAPM and the
APT consists in a linear relation between the asset risk premia and the risk premia
associated to one or several risk factors. The equilibrium analysis developed in
the previous chapter provides us with information about asset prices, returns and
risk premia starting from the knowledge of all the ingredients of the economy (in
particular, the preferences and the initial endowments of the agents) or, when the
aggregation property holds, referring to a representative agent economy. In contrast,
under suitable assumptions, the CAPM and the APT provide linear relations for the
asset risk premia without requiring a detailed knowledge of all the ingredients of the
economy.
This chapter is structured as follows. In Sect. 5.1, we apply the equilibrium
analysis developed in the previous chapter to derive the Consumption Capital Asset
Pricing Model (CCAPM), which consists of a linear relation between asset risk
premia and the risk premium of a market portfolio. In Sect. 5.2, assuming that
all the agents’ hold portfolios belonging to the portfolio frontier (see Chap. 3),
we derive the Capital Asset Pricing Model (CAPM), while in Sect. 5.4 we present
the Arbitrage Pricing Theory (APT), with asset returns being generated by a linear
multi-factor model. The extensive literature on the empirical analysis of the CAPM
and of the APT will be surveyed in Sects. 5.3 and 5.5, respectively. At the end of
the chapter, we provide a guide to further readings as well as a series of exercises.

202 5 Factor Asset Pricing Models: CAPM and APT
5.1 The Consumption Capital Asset Pricing Model

(CCAPM)
In this section, we derive several fundamental asset pricing relations, first by relying
on equilibrium arguments and then on no-arbitrage principles.
Equilibrium and Asset Risk Premia
In an economy with complete markets (or effectively complete markets, see

Sect. 4.3), equilibrium allocations are Pareto optimal and, therefore, an equilibrium
allocation of the economy can be characterized in terms of the optimum for a
representative agent in a no-trade equilibrium. By relying on a representative agent
analysis, we now derive general valuation principles which do not rely on specific
assumptions on the agents’ utility functions or on the distributions of the asset
returns.
We assume that the economy comprises I agents, with time additive, state-
independent, strictly increasing and risk averse preferences. We assume that the
initial endowments of the agents are given in terms of wealth at date t D 0 and
of units of the available securities. We also assume that markets are effectively
complete, so that every equilibrium allocation is Pareto optimal (see Sect. 4.3). In
correspondence of the equilibrium, the representative agent consumes the aggregate
endowment of the economy. We also assume that a risk free asset (denoted by n D 0)
is traded in the economy together with N risky assets. For each i D 1; : : : ; I, the
initial endowment of agent i is represented by the couple .xi0 ; ei /, with xi0 > 0
denoting the initial wealth and ei D .ei0 ; ei1 ; : : : ; eiN / 2 RNC1 the endowment in
terms of units of the NPC 1 assets. The aggregate endowmentPof the Peconomy is
I I N
then given by xm 0 WD iD1 x0 at date t D 0 and by xs WD
i m i
iD1 nD0 en dsn at
date t D 1 in correspondence of the state of the world !s , for s D 1; : : : ; S, where
ds0 D rf , for all s D 1; : : : ; S, p0 D 1 and where the superscript m stands for market
(i.e., the economy as a whole). We denote by xQ m the random variable taking values
.xm1 ; : : : ; xS /. To simplify the notation, in this chapter we shall refer to equilibrium
m
allocation/prices without denoting them with the superscript .

In the present setting, letting u0 and u1 be the representative agent’s utility func-
tions for consumption at time t D 0 and consumption at time t D 1, respectively,
and letting ı be the representative agent’s discount factor, in equilibrium it holds
that (see equation (4.52))

u01 .Qxm / Q
pn D ı E 0 m dn ; for all n D 0; 1; : : : ; N:
u0 .x0 /
5.1 The Consumption Capital Asset Pricing Model (CCAPM) 203
Equivalently, assuming that all assets have strictly positive prices (i.e., pn > 0 for
all n D 0; 1; : : : ; N) and dividing by pn the above relation, we get

u01 .Qxm /
ıE Q
r n D 1; for all n D 0; 1; : : : ; N; (5.1)
u00 .xm0/
where rQn denotes the random return of the n-th asset. In particular, applying the
above formula to the risk free asset, it holds that

u01 .Qxm / 1
ıE 0 m D : (5.2)
u0 .x0 / rf
By combining the above relations, we can write

0 m
EŒdQ n u1 .Qx / Q
pn D C ı Cov ; d n ; for all n D 1; : : : ; N;
rf u00 .xm
0/

u01 .Qxm / EŒQrn
ı Cov ; Q
r n C D 1; for all n D 1; : : : ; N:
u00 .xm0 / rf
In particular, from the last equation we can immediately derive an expression for the
asset risk premia:

ıu01 .Qxm /
EŒQrn rf D rf Cov ; Q
r n ; for all n D 1; : : : ; N: (5.3)
u00 .xm
0/
Formula (5.3) provides a fundamental relation between the risk premium of an

asset and the representative agent’s marginal rate of substitution of consumption.
Indeed, as we have already seen in the previous chapter, the risk premium of the
n-th asset is positive if and only if its random return rQn is negatively correlated
with the representative agent’s marginal rate of substitution of consumption, for
any n D 1; : : : ; N. This property is coherent with the diversification and the
insurance principles. Indeed, if the return of an asset is negatively correlated with
the marginal utility of aggregate consumption, then such an asset will pay more in
correspondence of the states of the world where the aggregate consumption is higher
and less when the aggregate consumption is lower. For this reason, such an asset will
be held by a risk averse agent only in exchange for a positive risk premium. On the
contrary, a risk averse agent can accept a negative risk premium for an asset which
provides insurance against states of the world where the aggregate consumption is
low, i.e., for an asset whose return is positively correlated with the marginal utility
of aggregate consumption.
Relation (5.3) also shows that only systematic risk is priced in equilibrium.
Indeed, if the return of an asset is purely idiosyncratic (i.e., independent from the
aggregate consumption), then its risk premium will be null and its expectation
will be equal to the risk free rate. This is true since purely idiosyncratic risk
can be diversified away and, due to the mutuality principle (see Sect. 4.1), in
correspondence of an equilibrium allocation it will not affect any agent. Therefore,
purely idiosyncratic risk will not be priced in equilibrium.
Note that, letting m Q WD ıu01 .Qxm /=u00 .xm
0 / and recalling that EŒm
Q D 1=rf
(see equation (5.2)), relation (5.3) can be equivalently rewritten as

Q rQn /
Cov.m; Var.m/Q
EŒQrn rf D D ˇnm m ; for all n D 1; : : : ; N;
Q
Var.m/ EŒm
Q
(5.4)
where ˇnm WD Cov.m; Q rQn /= Var.m/
Q corresponds to the regression coefficient of rQn on
mQ and m WD Var.m/=EŒ
Q Q can be interpreted as a risk premium on systematic
m
risk, as measured by the representative agent’s marginal rate of substitution.
Formula (5.4) shows that the risk premium of an asset is proportional to the
coefficient appearing in the linear regression of the asset return on the stochastic
Q
discount factor m.
The above relations between the representative agent’s marginal rate of substi-
tution in correspondence of the aggregate consumption and the asset risk premia
allow us to derive an upper bound on the Sharpe ratio of any traded asset, known as
the Hansen-Jagannathan bound (see Hansen & Jagannathan [891]). Recall that the
Sharpe ratio (with respect to the risk free rate rf ) is defined as the ratio of the return
risk premium EŒQrn rf over the standard deviation .Qrn /.
ıu01 .Qxm /
Q WD
Proposition 5.1 Under the assumptions of the present section, letting m u00 .xm
,
0/
the following holds:
jEŒQrn rf j .m/
Q
; for all n D 1; : : : ; N: (5.5)
.Qrn / EŒm
Q
Proof From equation (5.3), we can write:
1
EŒQrn rf D rf Cov.m;
Q rQn / D rf m;Q
Q rn .m/.Q
Q rn / D m;Q
Q rn .m/.Q
Q rn /;
EŒm
Q
where m;Q
Q rn denotes the correlation coefficient between m Q and rQn and where the third
equality follows from (5.2). Hence, since jm;Q
Q rn j 1, it holds that:
1
jEŒQrn rf j .m/.Q
Q rn /;
EŒm
Q
from which (5.5) follows directly, for every n D 1; : : : ; N. t

u
According to the above result, in equilibrium the Sharpe ratios are bounded from
above and from below by the volatility-expectation ratio of the stochastic discount
factor. Furthermore, the bound (5.5) shows that all assets lie inside a mean-variance
frontier, which is determined by the volatility-expectation ratio of the stochastic
discount factor. At the end of the present section we will see that an analogous bound
can be derived from no-arbitrage arguments, by replacing the stochastic discount
factor with the likelihood ratio of a risk neutral probability measure.
We have so far presented several relations arising from equilibrium arguments
on the risk premia of the single traded assets. However, since returns on portfolios
composed of the traded assets correspond to linear combinations of the returns on
the traded assets, relation (5.3) also applies to returns on arbitrary portfolios of the
N C 1 available securities.
P PIn particular,Pit can Pbe applied to the market portfolio.
Denoting by rQ m D . IiD1 NnD0 ein dQ n /=. IiD1 NnD0 ein pn /, the return of the market
portfolio, it holds that

ıu01 .Qxm / m
EŒQrm rf D rf Cov ; Q
r : (5.6)
u00 .xm
0/
This relation implies that, under the assumptions of the present section, the risk
premium associated to the market portfolio is positive (i.e., EŒQrm > rf ), as shown
in Exercise 5.1. This follows since rf > 0 (see formula (5.2)) and since the marginal
utility is a strictly decreasing function.
From relations (5.3) and (5.6) we immediately obtain the following formula,
which synthesizes the content of the Consumption Capital Asset Pricing Model
(CCAPM):

Cov u01 .Qxm /; rQn m
EŒQrn rf D 0 EŒQr rf ; for all n D 1; : : : ; N: (5.7)
Cov u1 .Qx /; rQ
m m
Relation (5.7) shows that the risk premium of any asset (or portfolio) is proportional
to the risk premium of the market portfolio, where the proportionality factor depends
on the covariance with the representative agent’s marginal utility of aggregate
consumption at time t D 1.
In the CCAPM equation (5.7), the relation between the risk premium of an
asset (or portfolio) and the risk premium associated with the market portfolio is
made complex by the appearance of the representative agent’s marginal utility of
aggregate consumption. A more explicit relation can be derived by considering the
portfolio which is minimally correlated with the representative agent’s marginal rate
of substitution, as shown in the following proposition, the proof of which is given
in Exercise 5.2.
Proposition 5.2 Under the assumptions of the present section, let wO 2 RN be a

portfolio satisfying

Corr u01 .Qxm /; rQ wO D min Corr u01 .Qxm /; rQ w ;
w2RN
where rw denotes the return of portfolio w. Then it holds that
Cov.Qrn ; rQ wO / wO
EŒQrn rf D EŒQr rf ; for all n D 1; : : : ; N: (5.8)
Var.QrwO /
Relation (5.8) shows that the risk premium of every security is proportional to
the risk premium of the portfolio having minimal correlation with the stochastic
discount factor, with the proportionality factor being given by the coefficient
appearing in the linear regression of the random return of the security with respect
to the random return of such a portfolio.
The CCAPM relation can be made more explicit by introducing additional
assumptions on the preference relations of the agents and/or on the distribution of
the asset returns. For instance, if the returns are jointly distributed according to
a multivariate normal law, then we can apply Stein’s lemma (see Lemma 3.9) to
relation (5.3), yielding

ıu001 .Qxm / EŒu001 .Qxm /
EŒQrn rf D rf E 0 m Cov .Qxm ; rQn / D Cov.Qxm ; rQn /;
u0 .x0 / EŒu01 .Qxm /
for all n D 1; : : : ; N, where in the second equality we have made use of

formula (5.2). Since this relation holds for all N risky assets, it also holds for
arbitrary portfolios, in particular for the market portfolio itself. Hence:
EŒu001 .Qxm /
EŒQrm rf D Cov.Qxm ; rQm /:
EŒu01 .Qxm /
Putting together the last two equations, we obtain the relation
Cov.Qxm ; rQn / m Cov.Qrm ; rQn / m

EŒQrn rf D EŒQ
r rf D EŒQr rf ; (5.9)
Cov.Qx ; rQ /
m m Var.Qr /
m
for all n D 1; P: : : ; P
N, where the last equality follows from the simple observation that
rQ m D xQ m =. IiD1 NnD0 ein pn /. Formula (5.9) represents the Capital Asset Pricing
Model (CAPM), on which we shall focus in Sect. 5.2. Note that this derivation of
the CAPM formula (5.9) is based on equilibrium arguments, under the assumption
that agents have increasing and strictly concave utility functions and asset returns
are jointly distributed according to a multivariate normal distribution. Observe that
EŒu001 .Qxm /=EŒu01 .Qxm / corresponds to the harmonic mean of the global absolute
risk aversion coefficients of the agents in the economy (see Constantinides [492]).
To this regard, see Exercise 5.3.
The Case of a Power Utility Function
We now specialize the CCAPM relation (5.7) to the case where agents’ preferences
are represented by power utility functions and asset returns have a general distribu-
tion, not necessarily multivariate normal. More specifically, we assume that
1
ui .x/ D .i C bx/11=b ; for all i D 1; : : : ; I;
b1
with b … f0; 1g. Recall that, as shown in Proposition 4.5, in the case of power
utility functions with common cautiousness coefficient, Pareto optimal allocations
are characterized by a linear sharing rule. Moreover, since the agents’ endowments
are defined in terms of the available securities and there exists a risk free asset,
markets are effectively complete. Hence, Pareto optimality is always attained in
correspondence of an equilibrium allocation. In this case, the aggregation property
holds and there exists a representative agent with utility functions u0 and u1 of
the power type (as computed in the proof of Proposition 4.15), with the same
cautiousness coefficient b. Equation (5.7) then gives that
P
Cov . IiD1 i C bQxm /1=b ; rQn
EŒQrn rf D P .EŒQrm rf /; for all n D 1; : : : ; N:
I 1=b
Cov . iD1 i C bQx /
m ; rQ m
(5.10)
In the special case where i D 0, for all i D 1; : : : ; I, it holds that

Cov .Qxm /1=b ; rQn
EŒQrn rf D .EŒQrm rf /; for all n D 1; : : : ; N;
Cov .Qxm /1=b ; rQ m
and, similarly, formula (5.1) becomes

" 1 #
xQ m b
ıE rQn D 1; for all n D 0; 1; : : : ; N: (5.11)
xm 0
Analogously, equations (5.3) and (5.2) can respectively be written as

1b ! " 1b #
xQ m xQ m 1
EŒQrn rf D rf ı Cov ; rQn and ı E D :
xm 0 xm 0 rf
In particular, from the last relation, which determines the equilibrium risk free rate
rf , we can observe that rf is high when agents are impatient (i.e., ı is small), when
the consumption growth rate xQ m =xm 0 is high and when risk aversion 1=b is high
(or, equivalently, when the elasticity of intertemporal substitution of consumption
is low). In this sense, if the elasticity of intertemporal substitution is low, then an
increase in the risk free rate has a limited effect on expected saving-consumption
growth.
In the case of power utility functions, we can obtain more explicit relations
if returns are distributed according to a log-normal law. Indeed, suppose that the
aggregate consumption growth rate xQ m =xm 0 and the random return r Qn are jointly
1=b
log-normally distributed. Then, the couple ı.Qxm =xm 0 / ; Q
r n is itself distributed
as a bivariate log-normal and, using the elementary properties of the log-normal
distribution together with relation (5.11), it holds that1
1
EŒlog.Qrn / D log.ı/ C E log.Qxm / log.xm
0/
b

1 1 2
Var log.Qrn / C 2 Var log.Qxm =xm
0 / Cov log.Q
r n /; log.Q
x m m
=x0 / ;
2 b b
1 1
log.rf / D log.ı/ C E log.Qxm / log.xm
0/ 2
Var log.Qxm =xm
0/ :
b 2b
The last equation for the equilibrium risk free rate rf is in line with the results of
Sect. 3.4 on optimal saving and consumption. Indeed, when the aggregate consump-
tion xQ m is more volatile, it is optimal to save more (compare with Proposition 3.26)
and, therefore, the equilibrium risk free rate will be lower. It is confirmed that rf is
high when agents are impatient (i.e., ı is small), when the consumption growth rate
xQ m =xm
0 is high and when risk aversion 1=b is high. From the last two relations, we
obtain the following expression for the risk premium of the n-th asset:
1 1
EŒlog.Qrn / log.rf / D Var log.Qrn / C Cov log.Qrn /; log.Qxm =xm
0/ :
2 b
The risk premium of the logarithmic return depends positively on the covariance
between the logarithmic growth rate of the aggregate consumption and the logarith-
mic return of the asset and also depends on the coefficient of relative risk aversion.
These results are in line with the diversification and insurance principles. A large
risk premium is obtained when the coefficient of relative risk aversion is high and/or
the covariance is large.
Applying Proposition 5.1 to the present setting of an economy populated by I
agents with power utility functions (with identical cautiousness coefficient and with
i D 0 for all i D 1; : : : ; I) and under the assumption of a log-normal distribution,
we obtain the following bound on the Sharpe ratios:
m
ı. xxQm /1=b q
jEŒQrn rf j 1
i D e b2 .log.Qx =x0 // 1 log.Qxm =xm
1 2 m m
0
h m 0/ :
.Qrn / E ı. xxQm /1=b b
0
(5.12)
1
Recall that, if a random variable Qx is distributed according to log.Qx/
N .; 2 /, then it holds
2 2 2
that EŒQx D eC =2 and Var.Qx/ D e2C .e 1/.
The interpretation of this approximation is that in equilibrium the bound on the

Sharpe ratio is higher if the economy is riskier (i.e., the aggregate consumption is
more volatile) or if investors are more risk averse.
Note that with b D 1 the utility function is quadratic. In this case, for-
mula (5.10) yields the classical CAPM relation, without any additional assumption
on the distribution of asset returns:
Cov.Qxm ; rQn / m Cov.Qrm ; rQn / m

EŒQrn rf D EŒQ
r rf D EŒQr rf ; (5.13)
Cov.Qxm ; rQ m / Var.Qrm /
for all n D 1; : : : ; N. In particular, observe that relation (5.13) coincides with the
CAPM relation (5.9). The CAPM has been previously derived for general utility
functions under the assumption of normally distributed returns. On the other hand,
equation (5.13) has been derived under the assumption of quadratic utility functions,
without any assumption on the specific distribution of the asset returns (to this
regard, see also Exercise 5.4). In Sect. 5.2 the CAPM relation will be derived under
a different set of assumptions, namely under the key assumption that agents choose
to hold portfolios belonging to the mean-variance portfolio frontier, by relying on
the results of Sect. 3.2.
Absence of Arbitrage and Asset Risk Premia
So far, we have derived several formulations of the CCAPM relation (as well
as of the CAPM relation, under two different sets of assumptions) by relying
on equilibrium arguments, together with additional assumptions on the form of
the utility functions and/or on the returns’ distribution in order to obtain more
explicit relations. As we have seen in Sect. 4.4, there is a deep connection between
equilibrium analysis and the no-arbitrage paradigm. Hence, it is not a surprise
that a CCAPM-type relation can also be derived without explicitly making use of
equilibrium arguments, under the assumption that the price-dividend couple . p; D/
does not admit arbitrage opportunities. In this framework, CCAPM-type relations
can be derived in terms of the likelihood ratio defining a risk neutral probability
measure (see the discussion following Proposition 4.22).
As in the previous chapter, let us assume that the probability space is finite, mean-
ing that there is a finite number S of elementary states of the world .!1 ; : : : ; !S /,
with associated probabilities of occurrence .1 ; : : : ; S /. Recall that, in view of
Proposition 4.22 and the following discussion, if the price-dividend couple . p; D/
does not allow for arbitrage opportunities, then there exists a likelihood ratio `Q with
values ` D .`1 ; : : : ; `S / 2 RSCC such that D .1 ; : : : ; S / D .`1 1 ; : : : ; `S S /
defines a risk neutral probability measure. Recall also that, when the market
is incomplete, there exist infinitely many risk neutral probability measures or,
equivalently, infinitely many likelihood ratios (see Proposition 4.26). We can state
the following result, which holds with respect to any likelihood ratio `. Q
Proposition 5.3 Let . p; D/ be an arbitrage free price-dividend couple and let

` D .`1 ; : : : ; `S / be the likelihood ratio associated to some risk neutral probability
measure. Then the following hold:
Q
EŒQrn rf D Cov.Qrn ; `/; for all n D 1; : : : ; N; (5.14)
and the Sharpe ratio of any traded asset satisfies the following bound:
jEŒQrn rf j Q
.`/; for all n D 1; : : : ; N: (5.15)
.Qrn /
Proof Suppose that . p; D/ does not admit arbitrage opportunities and let be any
Q Then, in view of
risk neutral probability measure, with associated likelihood ratio `.
formula (4.35), it holds that, for any n D 1; : : : ; N,
0 D E ŒQrn rf D EŒ`Q rQn rf D Cov.`;

Q rQn / C EŒQrn EŒ`Q rf ;
from which (5.14) follows, since EŒ`Q D 1. From (5.14), we obtain
Q rn / .`/.Q
jEŒQrn rf j D j Q̀;Qrn j.`/.Q Q rn /;
where Q̀;Qrn denotes the correlation coefficient between `Q and rQn . This implies the
bound (5.15). u
t
According to the above proposition, in an arbitrage free economy, the risk
premium of an asset is equal to the covariance between its return and a likelihood
ratio. Of course, this relation can be extended to the risk premium of a portfolio
of the available securities. Observe also that the CCAPM relation (5.7) can also
be obtained from equation (5.14), recalling that in a representative agent economy
a stochastic discount factor (or, equivalently, a likelihood ratio) can be defined in
terms of the representative agent’s marginal rate of substitution of consumption, as
follows from equation (4.53). This observation provides a link between CCAPM
relations obtained by equilibrium arguments and CCAPM relations derived from
no-arbitrage considerations.
Abstract CAPM-type formulae analogous to relations (5.9) and (5.13) can also
be obtained starting from no-arbitrage arguments, as we are now going to show.
The return on the market portfolio rQ m appearing in relations (5.9) and (5.13)
will be replaced by the return on a portfolio replicating a likelihood ratio. Let
us first consider the case of a complete market economy (the number of linearly
independent traded assets equals the number of possible states of the world). Then,
Proposition 5.3 together with the completeness of the market yield the following
corollary.
Corollary 5.4 Let . p; D/ be an arbitrage free price-dividend couple. Suppose that
the market is complete and let ` D .`1 ; : : : ; `S / be the likelihood ratio associated
to the (unique) risk neutral probability measure . Then there exists a portfolio
z` 2 RN with associated return rQ` such that the following holds:
Cov.Qrn ; rQ ` / `
EŒQrn rf D `
EŒQr rf D ˇn` EŒQr` rf ; (5.16)
Var.Qr /
where ˇn` WD Cov.Qrn ; rQ ` /= Var.Qr` /, for all n D 1; : : : ; N.

Proof If the price-dividend couple . p; D/ is arbitrage free and the market is
complete, then there exists a unique likelihood ratio ` 2 RSCC , see Proposition 4.26.
Due to the completeness of the market, there exists a portfolio z` such that Dz` D `.
Following the notation introduced in the previous chapter, let us denote by V.`/ the
market value at time t D 0 of the portfolio z` , i.e., V.`/ D p> z` , so that
Q
rQ ` D Dz` =V.`/ D `=V.`/:
With this notation, relation (5.14) directly implies that
EŒQrn rf D V.`/ Cov.Qrn ; rQ` /; for all n D 1; : : : ; N:
Of course, such a relation also holds for any portfolio composed of the N C 1
available assets, due to the linearity of the covariance operator. Hence, in particular,
it holds for the portfolio z` , thus leading to
EŒQr` rf D V.`/ Var.Qr` /:
By combining the last two equations we immediately obtain relation (5.16). t

u
According to the above corollary, in a complete market economy, the risk
premium of an asset is equal to the risk premium of the portfolio replicating the
(unique) likelihood ratio multiplied by the ˇn` coefficient, which corresponds to the
regression coefficient of the return rQn with respect to the random variable rQ ` . Note
that, despite the similarities, formula (5.16) and formulae (5.9) and (5.13) have been
derived under quite different assumptions and, in general, the portfolio z` will be
different from the market portfolio.
The risk premium expression in (5.16) allows us to rewrite the pricing func-
tional Q.c/ introduced in Sect. 4.4. Indeed, in a complete arbitrage free market,
formula (5.16) can be equivalently rewritten as
EŒdQ n EŒdQ n
pn D D ; for all n D 1; : : : ; N; (5.17)
EŒQrn rf C Cov.Qrn ; rQ ` /
where WD .EŒQr` rf /= Var.Qr` /. Hence, due to the market completeness assump-

tion, for any contingent consumption plan cQ represented by the vector c 2 RS ,
relation (5.16) implies that
EŒQc
Q.c/ D ; (5.18)
rf C Cov.Qrc ; rQ ` /
where rQ c denotes the return on the portfolio zc which replicates the payoff cQ (due to
the market completeness assumption, such a portfolio always exists). In the above
formula, the term can be interpreted as a correction coefficient for the discount
factor which accounts for the risk aversion of the agents, similarly as a risk neutral
probability measure differs from the physical probability measure due to the agents’
risk aversion. For this reason, the quantity appearing in the denominator of (5.17)–
(5.18) is sometimes called risk adjusted discount factor. Of course, if all agents are
risk neutral, then D 0 and all assets have the same expected return, equal to the
risk free rate rf .
The above results can be extended to incomplete markets. Recall that, in view
of Proposition 4.26, in the absence of arbitrage opportunities, the risk neutral
probability measure (and, hence, the likelihood ratio) is unique if and only if the
market is complete. Hence, if the market is arbitrage free but incomplete then there
exist infinitely many likelihood ratios with respect to which relation (5.14) holds
true. However, it can be shown that there exists a unique payoff Ò 2 I.D/ such
that the relation Q.c/ D EŒÒ cQ holds for all attainable contingent consumption
plans cQ (i.e., for all c 2 I.D/). This can be shown as a consequence of the
Riesz representation theorem, which in the present context can be stated as follows
(compare also with LeRoy & Werner [1191], Theorem 17.7.1 and Section 17.10).2
Lemma 5.5 Let Q W I.D/ ! R be a linear function. Then there exists a unique
vector Ò 2 I.D/ such that Q.c/ D EŒÒ cQ for every random variable cQ with values
c 2 I.D/.
Regardless of whether markets are complete or not, the absence of arbitrage
opportunities is equivalent to the existence of a linear pricing functional Q (see
Theorem 4.20). Due to the above lemma, there exists a unique element Ò 2 I.D/
such that, for every attainable payoff c 2 I.D/, the pricing functional Q.c/ can be
represented as the expectation of the inner product between Ò and c. In particular,
note that Ò is unique regardless of whether markets are complete or not. On the asset
span I.D/, the vector Ò has the same pricing implications of the pricing functional
Q, in the sense that EŒ`Q Oc D Q.c/, for all c 2 I.D/, so that the market value
of every attainable consumption plan can be equivalently computed by taking the
expectation of the inner product of c with the vector `. O However, unattainable
contingent consumption plans c … I.D/ cannot in general be priced as EŒÒ cQ ,
since Lemma 5.5 does not ensure in general that the vector Ò is strictly positive.
As shown in Exercise 5.7, the payoff Ò has also interesting properties in relation
with mean-variance theory. In particular, the return on the portfolio which replicates
Ò minimizes the second moment among all possible portfolio returns.
2
Similarly as in the previous chapter, with some abuse of notation we denote equivalently by cQ
and c 2 RS the random variable Qc with values c D .c1 ; : : : ; cS / 2 RS . Moreover, we denote by
EŒ ÒQc the expectation of the product between the random variable cQ and the random variable taking
values Ò.
5.2 The Capital Asset Pricing Model (CAPM) 213
By relying on Lemma 5.5, we can extend the risk premium relation (5.16) to
general incomplete markets as shown in the following proposition.
Proposition 5.6 Let . p; D/ be an arbitrage free price-dividend couple. Then there
exists a unique portfolio z` 2 RN with associated return rQ ` such that relation (5.16)
holds true, for all n D 1; : : : ; N.
Proof By Theorem 4.20, the absence of arbitrage opportunities implies the exis-
tence of a strictly positive linear pricing functional Q. Hence, in view of Lemma 5.5,
there exists a unique vector Ò 2 I.D/ such that
pn D EŒÒ dQ n ; for all n D 0; 1; : : : ; N:
In terms of returns, this can be rewritten as
1 D EŒÒ rQn ; for all n D 0; 1; : : : ; N:
Applied to the risk free asset, this gives EŒÒ D 1=rf . Define ` WD `=EŒ O Ò . Then,
O `
since ` 2 I.D/, there exists a unique portfolio z (under the standing assumption of
non-redundant assets) such that Dz` D `. Thus, denoting by `Q the random variable
taking values ` D .`1 ; : : : ; `S /, we can write, for all n D 1; : : : ; N,
1 Q 1
Q rQn / C EŒQrn D 1 V.`/ Cov.Qr` ; rQn / C EŒQrn ;
1D EŒ` rQn D Cov.`;
rf rf rf
from which relation (5.16) follows by the same arguments used in the proof of
Corollary 5.4. u
t
5.2 The Capital Asset Pricing Model (CAPM)
In the previous section, we have derived the CAPM relation by equilibrium

arguments under the assumption of normally distributed returns or quadratic utility
functions (see relations (5.9) and (5.13)). Now, we present an alternative derivation
of the CAPM by relying on the assumption that all agents choose to hold portfolios
belonging to the mean-variance portfolio frontier (see Sect. 3.2). In a nutshell, if
every market participant holds a frontier portfolio, then also the market portfolio,
being a convex combination of the individual portfolios, will belong to the portfolio
frontier. Hence, Proposition 3.13 (see also Proposition 3.16 when a risk free asset is
traded in the market) implies that a linear relation holds true between the expected
return of any portfolio and the expected return of the market portfolio. This will
lead to the CAPM relation (5.9). Note that the hypotheses of normal distribution
or of quadratic utility function imply that agents hold a portfolio belonging to the
portfolio frontier (compare with Sect. 2.4).
As in the previous section, we consider an economy populated by I risk averse

agents. We assume that N risky assets are available for trade (later we shall
also introduce a risk free asset with return rf ), with random returns .Qr1 ; : : : ; rQN /
with finite second moments. The N risky assets are in strictly positive supply.
As always, we assume a frictionless and perfectly competitive financial market,
meaning that there are no transaction costs, taxes or trading constraints and that
the economic model is common knowledge among the agents. We denote by
wi0 the wealth of agent i at date t D 0 and by win the proportion of wealth
invested by agent i in the n-th asset, for i D 1; : : : ; I and n D 1; : : : ; N. We
assume that the wealth of the agents is fully invested in the financial assets. The
aggregate wealth P of the economyPI (i.e.,
PN thei total market capitalization) is then
I
given by wm 0 D w
iD1 0
i
D iD1 nD1 en pn and the proportion of aggregate
wealth invested in thePIn-th asset (i.e., the market capitalization
P PN Pof the n-th asset)
I I
n D iD1 wn w0 =w0 D . iD1 en pn /=. nD1 iD1 en pn /, where
i i m
is given by wm i i
the last equality follows since in equilibrium the aggregate demand

P of the risky
assets equals the aggregate supply. In turn, this implies that NnD1 wm D 1. We
PN n
denote by rQ m the return on the market portfolio, i.e., rQ m D nD1 n r
w m
Qn . As in
Sect. 3.2, we call portfolio frontier the set of all portfolios which solve the mean-
variance optimization problem (3.14) as the expected return varies. We then have
the following proposition, which establishes the CAPM relation.
Proposition 5.7 Suppose that all the I agents hold portfolios belonging to the
mean-variance portfolio frontier. Then, under the assumptions of the present section,
the market portfolio wm belongs to the portfolio frontier. Moreover, if wm does not
coincide with the minimum variance portfolio wMVP , denoting by rQ zc.m/ the return of
the frontier portfolio having zero correlation with the market portfolio, the following
relation holds true, for all n D 1; : : : ; N:
Cov.Qrn ; rQm / m
EŒQrn EŒQrzc.m/ D EŒQr EŒQrzc.m/ DW ˇnm EŒQrm EŒQrzc.m/ :
Var.Qr /
m
(5.19)
Proof wm is a convex combination of the individual portfolios wi 2 RN , i D
1; : : : ; I. Hence, in view of Property 1 of the portfolio frontier (see section “The Case
of N Risky Assets”), if wi belongs to the portfolio frontier, for all i D 1; : : : ; I, then
also wm will belong to the portfolio frontier. Relation (5.19) then follows directly
from (3.32) by taking wp D wm . t
u
By linearity, relation (5.19) also applies to arbitrary portfolios composed of
the N assets. Note that, in the context of the mean-variance portfolio frontier,
relation (3.32) is a property which holds with respect to any frontier portfolio wp .
Instead, the CAPM relation (5.19) is based on the assumption that, in equilibrium,
the market portfolio belongs to the portfolio frontier.
The result of Proposition 5.7 holds true in the case where the two mutual funds
separation property holds, as shown in the following corollary.
Corollary 5.8 Suppose that the two mutual funds separation property holds. Then,
under the assumptions of the present section, the market portfolio wm belongs to the
portfolio frontier. Moreover, if wm ¤ wMVP , then relation (5.19) holds true.
Proof As shown in section “Mutual Fund Separation and Mean-Variance Portfolio
Selection”, if the two mutual funds separation property holds, then the two mutual
funds belong to the portfolio frontier. Hence, since the portfolio frontier is stable
with respect to convex combinations, it follows that the market portfolio also
belongs to the portfolio frontier. The result then follows from Proposition 5.7. t
u
Observe that the sign of the coefficient ˇnm appearing in (5.19) is determined by
the sign of the correlation coefficient between the return of asset n and the return on
the market portfolio. In the ˇ-expected return plane, relation (5.19) is represented
by a line (Security Market Line) which intersects the vertical axis in correspondence
of EŒQrzc.m/ and with slope EŒQrm EŒQrzc.m/ . We have thus established the following
important result: in equilibrium, the expected return of a portfolio is a linear affine
combination of the expected return of the market portfolio and of the expected return
of the portfolio belonging to the portfolio frontier with zero covariance with the
market portfolio. The slope EŒQrm EŒQrzc.m/ of the Security Market Line is positive
or negative depending on the efficiency of the market portfolio (see Property 4 in
section “The Case of N Risky Assets”). Indeed, if the market portfolio is efficient
(i.e., wm 2 EPF, so that EŒQrm > EŒQrzc.m/ , always assuming that wm ¤ wMVP ),
then the Security Market Line has a positive slope, while, if the market portfolio
is inefficient (meaning that EŒQrm < EŒQrzc.m/ ), then the Security Market Line has
a negative slope. In the first case, high expected returns correspond to high ˇ
coefficients, so that, in order to get a high expected return, one needs to invest
in a rather risky portfolio (with risk being measured in terms of the covariance
with the return on the market portfolio). As discussed before, this result can be
naturally interpreted in terms of the insurance and diversification principles: having
the possibility of investing in the market portfolio, the agent invests in a risky asset
with a large covariance with the latter (thus increasing the overall variance if an
agent is exposed to both the asset and the market portfolio) if and only if the asset
expected return is high (see also below for an illustration of this phenomenon). On
the contrary, if the Security Market Line has a negative slope, then high expected
returns correspond to negative ˇ coefficients.
In particular, the Security Market Line has a positive slope when agents have
increasing and strictly concave utility functions and returns are distributed according
to a multivariate normal law, as shown in the following corollary.
Corollary 5.9 Under the assumptions of the present section, suppose that the
returns .Qr1 ; : : : ; rQN / are jointly normally distributed with non-identical expectations
and that all the I agents have strictly increasing and concave utility functions. Then
it holds that

EŒQrn D EŒQrzc.m/ C ˇnm EŒQrm EŒQrzc.m/ ; for all n D 1; : : : ; N; (5.20)
and EŒQrm EŒQrzc.m/ > 0.

Proof If the N traded assets have jointly normally distributed returns with non-
identical expectations, then Proposition 3.21 implies that the two funds separation
property holds true with respect to any frontier portfolio wp and to its zero-
correlation portfolio wzc. p/ . Moreover, since the agents have strictly increasing and
concave utility functions, they will choose to hold only efficient portfolios. Indeed,
due to Proposition 2.13, the agents’ preferences are increasing with respect to the
expected return and decreasing with respect to the variance. In turn, this implies that
the market portfolio, being a convex combination of the portfolios of the individual
agents, also belongs to EPF. The result then follows from Proposition 5.7, while
EŒQrm EŒQrzc.m/ > 0 is due to the efficiency of the market portfolio (compare with
Property 4 in section “The Case of N Risky Assets”). t
u
Relation (5.20), together with the inequality EŒQrm EŒQrzc.m/ > 0 (efficiency of
the market portfolio), represents the Zero-ˇ Capital Asset Pricing Model, see Black
[242]. The efficiency of the market portfolio implies that, in order to get a high
expected return in equilibrium, one must look for a risky portfolio (i.e., a portfolio
characterized by a positive and high covariance with the market portfolio). An asset
with a negative covariance will be characterized by a low expected return, lower
than the return of the portfolio having zero covariance with the market portfolio.
A conclusion analogous to Corollary 5.9 can be derived under the assumption
that all the agents have quadratic utility functions (with possibly different risk
aversion coefficients), as shown in the following corollary. Note that we also need to
assume that expected returns of portfolios lie in the region where the agents’ utility
functions are increasing.
Corollary 5.10 Under the assumptions of the present section, suppose that all the
i
I agents have quadratic utility functions of the form ui .x/ D x b2 x2 , for all i D
P
1; : : : ; I. Assume furthermore that EŒwi0 NnD1 rQn win < 1=bi for every wi0 2 R and
wi 2 RN , i D 1; : : : ; I. Then relation (5.20) holds true and EŒQrm EŒQrzc.m/ > 0.
P
Proof As shown in Sect. 2.4, letting W e i WD wi0 NnD1 rQn win , the expected utility of
agent i can be written as follows,
i i bi i 2
e DE W
E ui W e E W e
2

bi e i i bi i
e Var We :
D 1 E W EW
2 2
i
Since it has been assumed that E W e < 1=bi , for all i D 1; : : : ; I, the expected
utility is strictly increasing with respect to the expected return of a portfolio and
strictly decreasing with respect to its variance. As a consequence, every agent will
choose to hold a mean-variance efficient portfolio. The result then follows as in the
proof of Corollary 5.9. t
u
It is interesting to compare the result of the above corollary with the CAPM
relation (5.13). In both cases, quadratic utility functions have been assumed.
Relation (5.13) has been obtained by relying on the existence of a representative
agent, while Corollary 5.10 is based on the assumption that the expected returns on
optimal portfolios lie in the region where the utility functions are increasing. Note
also that market completeness is not required.
Let us now extend the results of the present section to an economy with N risky
assets and a risk free asset with return rf > 0. Recall from section “The Case of
N Risky Assets and a Risk Free Asset” that, in this case, the shape of the portfolio
frontier depends on the relation between the risk free rate rf and the expected return
A=C associated to the minimum variance portfolio composed of the risky assets
only. In particular, if rf ¤ A=C, then the portfolio frontier is represented by two
half-lines and every frontier portfolio can be written as a linear combination of the
tangent portfolio we and the risk free asset (see Proposition 3.15). On the basis
of these observations, we can establish the following simple but important result
(compare also with Sharpe [1529], Lintner [1221], Mossin [1357]).
Proposition 5.11 Suppose that a risk free asset with return rf ¤ A=C is traded
and that all the I agents hold portfolios belonging to the mean-variance portfolio
frontier. Assume furthermore that the net supply of the risk free asset is zero (and that
of the risky assets is strictly positive). Then, under the assumptions of the present
section, the market portfolio wm coincides with the tangent portfolio we and, hence,
belongs to the portfolio frontier. Moreover, the following relation holds true:

EŒQrn rf D ˇnm EŒQrm rf ; for all n D 1; : : : ; N; (5.21)
with ˇnm D Cov.Qrn ; rQ m /= Var.Qrm /.

Proof As explained in section “The Case of N Risky Assets and a Risk Free
Asset”, if rf ¤ A=C, then every portfolio belonging to the portfolio frontier can
be represented as a linear affine combination of the tangent portfolio we and of the
risk free asset. Hence, if all the agents hold portfolios belonging to the portfolio
frontier and the aggregate supply of the risk free asset is zero, it follows that
the market portfolio coincides with the tangent portfolio we . Relation (5.21) then
follows directly from (3.41) by taking wp D wm . t
u
Relation (5.21) is illustrated in Fig. 5.1. Of course, relation (5.21) immediately
extends to arbitrary portfolios composed of the N C1 available securities. In the case
where rf D A=C, the portfolio frontier including the risk free asset is represented
by the two asymptotes to the portfolio frontier composed of the risky assets only.
As discussed in section “The Case of N Risky Assets and a Risk Free Asset”, in
this case the overall net investment in the risky assets is zero and the total wealth
is invested in the risk free asset. Hence, the economy is in equilibrium only if the
aggregate supply of the risk free asset is strictly positive and the aggregate supply
of the risky assets is null. Therefore, under the assumptions of the present section
(strictly positive supply of the risky assets), an equilibrium cannot exist and the
CAPM does not hold.
Fig. 5.1 Capital Asset Pricing Model
Proposition 5.11 establishes the CAPM relation in the presence of a risk free
asset. However, Proposition 5.11 does not assert the efficiency of the market
portfolio, i.e., we do not know a priori whether EŒQrm > rf . As shown in the next
proposition, the efficiency of the market portfolio can be established under rather
natural assumptions on the agents’ utility functions.
Proposition 5.12 Under the assumptions of Proposition 5.11, suppose furthermore
that all the I agents have strictly increasing and concave utility functions. Then the
market portfolio is efficient, i.e., EŒQrm > rf .
Proof Under the present assumptions and in view of Proposition 3.7, the optimal
portfolio of every agent will have an expected return greater or equal than the risk
free rate rf . Recall also that, due to Proposition 5.11, the market portfolio coincides
with the tangent portfolio. Hence, if rf > A=C, then it holds that EŒQrm < rf and
no investor will choose to invest a strictly positive amount of wealth in the market
portfolio, thus leading to a contradiction with the assumption of a strictly positive
supply of risky assets. This implies that rf < A=C. In this case, the market portfolio
is efficient, since the tangent portfolio belongs to the upper part of the portfolio
frontier (see section “The Case of N Risky Assets and a Risk Free Asset”). t
u
Similarly to the case of formula (5.17), the CAPM relation (5.21) can also be
used to establish a valuation rule as follows:
EŒdQ n
pn D ; for all n D 1; : : : ; N: (5.22)
rf C ˇnm EŒQrm rf
The denominator rf C ˇnm .EŒQrm rf / represents a risk adjusted discount factor for
the n-th asset. Note that, as long as the market portfolio is efficient (i.e., EŒQrm > rf ),
there exists a positive relationship between the ˇ coefficient and the risk adjusted
discount factor, or, equivalently, a negative relationship between the price of an asset
and its ˇ coefficient.
Summing up, in the Capital Asset Pricing Model, asset risk premia are measured
with respect to a single risk factor, represented by the return on the market portfolio.
The sensitivity of an asset return with respect to this single factor is captured by the
ˇ coefficient, which represents the coefficient in the linear regression of the asset
return on the market portfolio return. Indeed, in view of Proposition 3.16, for an
arbitrary portfolio w it holds that
rQ w D rf C ˇwm .Qrm rf / C "Qwm ;
where the random variable "Qwm is uncorrelated with the market portfolio return, so
that
2
Var.Qrw / D ˇwm Var.Qrm / C Var.Q"wm /:
According to this representation, the return rQ w is decomposed into a constant

term, equal to the risk free rate rf , a risky component related to the return of
the market portfolio (market risk) and, finally, a residual component "Qwm . The
latter component is called idiosyncratic risk. The market risk is the systematic
risk that cannot be diversified away and, therefore, it is compensated through
the risk premium ˇwm .EŒQrm rf /, unlike the idiosyncratic risk, which can be
diversified away (provided that the residual risk components associated to different
assets are uncorrelated). For this reason, jˇwm j is an indicator of the riskiness of
a portfolio w. As already mentioned, these facts can be interpreted in terms of
the diversification and insurance principles. Indeed, take the point of view of an
agent who holds the market portfolio and considers the possibility of investing
in the portfolio w. If ˇwm > 0, then the return rQw is positively correlated with
the market portfolio. Hence, in order to invest a positive amount of wealth in
the portfolio w, the investor will require a compensation for the higher risk of
his overall position. On the contrary, if ˇwm < 0, then the agent can accept a
negative risk premium in order to benefit from the negative correlation between
rQ w and rQ m (insurance principle). As shown in the notes at the end of the chapter,
the CAPM can be extended to the case where agents exhibit heterogeneous beliefs
about the expected returns of the risky assets (see also Hens & Rieger [938,
Proposition 3.5]).
5.3 Empirical Tests of the CAPM
The CAPM establishes a linear relation between the risk premium of an asset
(or of a portfolio) and its ˇ coefficient with respect to the market portfolio. This
is an ex-ante relation, since it involves the expectations of random returns, and it
cannot be directly tested as such on financial time series. As a matter of fact, no
time series for the risk premia and for the ˇ coefficients are directly available.
The approaches to test the CAPM fall in two classes: time series approaches and
“two pass” approaches. In particular, the latter consist in first estimating the beta
coefficients and the expected returns and then testing the validity of the CAPM
relation.
The CAPM has been derived in a two-period economy. Therefore, in testing the
CAPM we implicitly assume that the CAPM holds period by period (asset returns
are assumed to have a stationary probability distribution). The classical assumption
is that asset returns are identically and independently distributed (i.i.d.) over time
according to a normal multivariate distribution (an assumption consistent with the
CAPM).
One of the main problems in testing the CAPM is represented by the fact that the
market portfolio is not observable. To overcome this problem, a proxy of the market
portfolio is typically employed and usually the proxy is represented by a stock index.
Note that the CAPM is a general equilibrium model for a closed economy, so that
the market portfolio represents the composition of the wealth of the whole economy,
while any stock index is only a proxy of the market portfolio (see Roll [1457]
concerning the problems due to the non-observability of the market portfolio). The
risk free rate is usually set equal to the interest rate of short term Treasury bills.
Let us briefly illustrate the classical two pass procedure for testing the CAPM.
Seminal examples of the two pass procedure are Fama & MacBeth [678], Black et al.
[245], Blume & Friend [258]. To implement the first step, we must transform the ex-
ante CAPM model into an ex-post model. The ex-ante relation can be transformed
into an ex-post relation by making the rational expectations hypothesis: asset returns
are generated by a stationary model which is common knowledge among all the
agents of the economy. Under this hypothesis, the expected return of an asset should
be equal to its historical mean.
At any date t 2 f0; 1; : : :g, asset returns can be written in the following form

rQn;t D EŒQrn;t C ˇnm;t rQtm EŒQrtm C Qn;t ; for all n D 1; : : : ; N; (5.23)
where rQn;t denotes the random return of asset n at date t, EŒQ t D 0, EŒQ t Qt> D
˙t , for some variance-covariance matrix ˙t , ˇnm;t WD Cov.Qrn;t ; rQtm /= Var.Qrtm / and
rQtm denotes the market portfolio return, with Cov.Qrtm ; Qn;t / D 0, for all n D
1; : : : ; N and t 2 f0; 1; : : :g. As can be easily checked, relation (5.23) is always
satisfied, independently of the validity of the CAPM. Suppose now that the returns’
distribution is stationary in the sense that ˙t D ˙ and ˇnm;t D ˇnm , for all
t 2 f0; 1; : : : ; g and n D 1; : : : ; N. Under the rational expectations hypothesis and
5.3 Empirical Tests of the CAPM 221
imposing the CAPM restrictions in (5.23), the CAPM in ex-post form is obtained,
for t 2 f0; 1; : : :g:
rQn;t rf D ˛n C ˇnm .Qrtm rf / C Qn;t ; for all n D 1; : : : ; N: (5.24)
An equivalent expression is obtained for the zero-ˇ CAPM.

In the first step of the two pass procedure, on the basis of model (5.24),
ˇ coefficients are estimated through the ordinary least squares (OLS) estimator
(time series regression). Note, however, that in the presence of heteroskedastic and
correlated returns OLS estimates are inefficient relative to generalized least square
estimators. In the second step, the average returns of the securities are regressed
cross-sectionally against the ˇ coefficients estimated in the first step. Letting rOn and
ˇOn be the expected return and the ˇ coefficient estimated for portfolio p in the first
step, we can consider the following regression:
rOn D 0 C 1 ˇOn C vQn : (5.25)
We can test the following implications of the CAPM on relation (5.25):

• validity of the linear relation (5.25) between the risk premium of an
asset/portfolio and the ˇ coefficient;
• 0 D rf ;
• the market return is the only risk factor;
• the market portfolio belongs to the efficient part of the portfolio frontier
composed of the risky assets together with the risk free asset and its risk premium
is equal to 1 (and, therefore, 1 > 0).
Roll [1457] points out that the first three implications are a direct consequence
of the fact that the market portfolio belongs to the efficient portfolio frontier.
So, the only hypothesis to be tested is that the market portfolio belongs to the
efficient portfolio frontier. In other words, the above hypotheses cannot be tested
independently.
To test the zero-ˇ CAPM, instead of the (observable) risk free rate there is the
expected return of the frontier portfolio having zero correlation with the market
portfolio, a portfolio with a non-observable return. In this case, the two pass
procedure is similar to the one described above for the CAPM, with the hypotheses
to be tested on model (5.25) becoming 0 0 and 1 > 0.
Early Tests
The CAPM could be directly tested on asset returns: unfortunately, the residual
(noise) component Qt for a single asset is very large and, therefore, the estimates of
ˇ turn out to be twisted (errors in variables problem). In Miller & Scholes [1343] it
was observed that assets with high (low) ˇ are characterized by a risk premium
lower (higher) than that predicted by the CAPM. To overcome this drawback
(errors in the estimates of the ˇ of single assets), Black et al. [245] and Fama &
MacBeth [678] proposed to proceed to an aggregation in the second step of the
procedure by building a set of asset portfolios with dispersed ˇ by sorting assets
based on estimated betas. On the one hand this aggregation procedure reduces
the measurement errors of the ˇ coefficients, on the other hand it reduces the
power of regression tests. Moreover, an aggregation process induces selection bias
problems (data snooping biases). Grouping assets to build portfolios based on some
stocks’ empirical characteristics (estimated beta, size, price-earnings ratio) creates
potentially significant biases in the test statistics, in particular the null hypothesis
is quite likely to be rejected even when it is true (see Lo & MacKinlay [1235]).
Note that a two pass approach is by its own nature affected by the errors in variables
problem.
Early empirical evidence reported in Black et al. [245], Fama & MacBeth [678],
Blume & Friend [258] is consistent with the mean-variance efficiency of the market
portfolio: the relation between a portfolio risk premium and the ˇ coefficient is
linear and there is no other risk factor. The market risk premium is positive but
1 is significantly lower than the market risk premium observed empirically (1:08
instead of 1:42 in Black et al. [245]). Moreover, 0 rf turns out to be significantly
different from zero (0:519 in Black et al. [245]). However, these findings can be
easily interpreted in the context of the zero-ˇ CAPM.
As an illustrative example, let us give some more details on the results of Fama
& MacBeth [678]. Their data sample is given by 1926–1968 NYSE stocks and the
market portfolio is the NYSE Index. The return frequency is monthly. They start
by considering the period 1926–1929 and perform the first pass for each stock
estimating the ˇ coefficients. The assets are ranked according to their ˇ and are
partitioned in twenty portfolios. For each portfolio the monthly return is computed
for the period 1930–1934. The first pass of the procedure is repeated obtaining
an estimate of the ˇ coefficients of the portfolios. The second pass is performed
for each month in the 1935–1938 period. The procedure is repeated by rolling the
estimation and the testing samples.
The use of a market portfolio proxy to test the CAPM can be the origin of a bias.
Roll [1457] points out the problem: a negative empirical evidence for the model
described above simply means that the market portfolio proxy does not belong
to the frontier and nothing can be established about the validity of the CAPM
without knowing the relation between the true market portfolio and its proxy. For
example, the author argues that the negative results obtained in Black et al. [245]
for the CAPM are compatible with the classical CAPM and an ill-specified market
portfolio. A test with a proxy of the market portfolio provides implications on the
fact that the true market portfolio belongs to the frontier only if its ˇ coefficient is
equal to one and the error terms Qn;t are uncorrelated with the market portfolio return.
The relevance of this critique (Roll’s critique) has been analysed in two directions.
In Stambaugh [1561] it is shown that the results of an empirical test do not vary in
a sensible way with the composition of the proxy of the market portfolio (adding
bonds and real estate to the index). In Shanken [1525] and Kandel & Stambaugh
[1067] it is shown that a negative result on the CAPM with respect to a proxy of
the market portfolio implies the rejection of the CAPM provided that the correlation
between the proxy and the true market portfolio is high enough (larger than 0:7).
These results led to reconsider the relevance of Roll’s critique.
CAPM restrictions can be tested by estimating (5.24) through the time series
approach. Under the hypothesis that the CAPM holds, the coefficient ˛n should be
equal to zero. This fact is a direct consequence of the fact that the market portfolio
(or its proxy) belongs to the efficient part of the portfolio frontier composed by the
N risky assets together with the risk free asset. We can also test that the CAPM is
correctly specified adding other regressors.
CAPM Anomalies
The empirical evidence for the CAPM in the ’70s was substantially positive.
However, in the ’80s a large amount of literature showed that the return of the
market portfolio is not the only risk factor. In particular, some characteristics of
the stocks turn out to be significant in order to explain asset risk premia (CAPM
anomalies). Portfolios composed of assets of companies with some characteristics
turn out to have a Sharpe ratio higher than that predicted on the basis of the market
portfolio proxy. For a survey on this literature see Hawawini & Keim [916] and
Schwert [1510]. With no intent to be exhaustive, the list of anomalies include:
Price earnings ratio: in Basu [179], Ball [121, 122] it is shown that the
company’s price-earnings ratio turns out to be relevant in order to explain the
portfolio risk premium: assets with a low (high) price-earnings ratio have an average
return higher (lower) than that predicted by the CAPM.
Size: in Banz [140], Reinganum [1443] it is shown that the size of the company,
represented by the market capitalization, explains portfolio risk premia better than
the ˇ coefficient. Size as a second risk factor, besides the market return, contributes
to explain the residual returns variability. The relation is negative: returns of small
companies are higher than those predicted by the CAPM. This result suggests the
existence of a positive small cap premium which has inspired many small cap funds
in the last decades. However, note that Dimson & Marsh [578], Schwert [1510]
observe that the small cap premium has been small or even negative in the last
twenty years.
January effect: Reinganum [1444], Loughran [1246], Keim [1079] showed that a
large part of the abnormal returns on small firms occurs during the first month of the
year: short selling pressure on small companies reduces the price of small companies
in December with a rebound effect in January. Schwert [1510] has shown that this
effect is confirmed in a more recent analysis.
Leverage: in Bhandari [221] it is shown that a positive relationship exists between
leverage and asset returns. Employing leverage as a third factor together with the
market return and size helps explaining the residual returns variability.
Book to market value ratio: in Stattman [1566] it is shown that the ratio of the
book value of equity over its market value is positively correlated with the average
return of the assets.
Mean reversion and momentum effect: in De Bondt & Thaler [532] it is shown
that the past performance of a company turns out to be significant: portfolios
composed of assets with a poor performance in the past three/five years exhibit
an average return higher than that predicted by the CAPM (mean reversion effect).
On the other hand, assets with high returns over the past three to twelve months
continue to have high returns in the next future (momentum effect), see Jegadeesh
[1023], Jegadeesh & Titman [1026].
Liquidity: risk premia are related to market liquidity, expected return is an
increasing and concave function in the bid-ask spread and other forms of illiquidity,
see Amihud & Mendelson [59], Chalmers & Kadlec [393], Brennan & Subrah-
manyam [297], Eleswarapu [635], Amihud [57], Easley et al. [615], Bekaert et al.
[185]. This phenomenon agrees with the size anomaly.
There is a high degree of interdependence among the above anomalies. These
results were not welcomed by the academic community. In the first place, these
results have little theoretical foundation and the performance of multi-factor models
(APT for instance) are not necessarily better than that of the CAPM. Moreover,
these empirical findings are affected by a series of problems (e.g., data snooping,
infrequent trading and liquidity bias of small stocks, selection bias, measurement
errors in ˇ). The debate on these topics exploded in the ’90s after the seminal
contribution Fama & French [669]. According to Fama & French [669], the positive
relationship between portfolio risk premia and ˇ established in early contributions is
not confirmed in the period 1963–1990 (on this point see also Reinganum [1442]).
The average slope from the regression of returns on ˇ is found to be 0:15% per
month (with t-statistics 0:46, flat relation). For the sample 1981–1990, the market
risk premium is even negative. In a cross-sectional regression of asset returns,
the ˇ coefficient is less significant than factors like size (0:15% as a coefficient
with t-statistics 2:58), book to market value of equity ratio (0:5% as a coefficient
with t-statistics 5:71), leverage, earnings-price ratio. In particular, the size effect is
strong and robust but the book to market value ratio is more powerful than size
in explaining the cross-sectional variability. High-beta stocks do not exhibit higher
returns than low-beta stocks of the same size or with the same book to market equity
ratio. Moreover, adding ˇ as a second factor to size, book to market equity, leverage
or earnings-price ratio does not help explaining average returns (in the first case
the coefficient of the regression is even negative while the coefficients of the other
variables are statistically different from zero). The size of the company and the
book to market value ratio obscure the leverage and earnings-price ratio effect. This
analysis has shown that firms with high book to market value of equity (low market
price relative to the book value of assets), earnings to price or cash flow to price
ratios (value stocks) have higher average returns than stocks with low ratios (growth
stocks). Value stocks are good candidates to be undervalued by the market and, on
the contrary, growth stocks are good candidate to be overvalued by the market.
Three different schools of thought can be identified on the literature on CAPM

anomalies: those who criticize the statistical robustness of the anomalies, those
who interpret the evidence inside the classical asset pricing theory (i.e., anomalies
subsume a risk factor not captured by ˇ), and those who interpret the evidence as a
signal of market irrationality (behavioral finance). In our presentation, we follow
Boudoukh et al. [273] identifying these three schools respectively as loyalists,
revisionists and heretics.
Loyalists
The results of Fama & French [669] have been discussed in several papers. In
Kothari et al. [1126], three different criticisms are put forward. In the first place,
due to the low statistical power of the tests for a positive market risk premium,
the results of Fama & French [669] provide little support for rejecting the null
hypothesis of a market risk premium equal to the one observed historically. This
relies on the presence of a strong noise component in the asset returns. The second
critique concerns the fact that an estimate of ˇ based on annual returns (instead of
monthly returns as in Fama & French [669]) produces a stronger positive relation
between average returns and ˇ (the market risk premium is positive and significant,
see also Handa et al. [885]). The third point concerns the data set (COMPUSTAT)
employed in a large part of the empirical analysis. In this data set new assets are
inserted as time goes on: when an asset is inserted in the data set, its time series is
integrated by several years, before the year of insertion. This procedure induces
a survivorship bias: companies characterized at some date t by a high book to
market value and subsequently by small returns have a low survival probability
and, therefore, are unlikely to be inserted in the data set at a future time s > t.
On the contrary, companies with high returns are more likely to survive and to be
included in the data set. Adding some years to the time series when an asset is
inserted in the data set introduces a bias towards the existence of a positive relation
between risk premium and the book to market value ratio. Moreover, some authors
observe that the COMPUSTAT database does not include many firms experiencing
financial distress. Using a different database, some authors have shown that the
relation between risk premium and the book to market value ratio is much weaker
than that reported in Fama & French [669]. An insignificant effect of the book to
market ratio is detected in Breen & Korajczyk [288].
The results obtained in Fama & French [669] and, in particular, the non-existence
of a relationship between the ˇ coefficient and expected returns can be traced back
to the statistical procedure employed to test the CAPM. A classical resolution relies
on the Roll’s critique: if the market portfolio proxy does not belong to the portfolio
frontier, then other factors besides ˇ turn out to be significant. Roll & Ross [1462]
show that there is a region delimited by a parabola inside the mean-variance frontier
containing market index proxies with no relation to expected returns. The parabola
is not far away from the mean-variance frontier and, therefore, a slight inefficiency
of the market proxy with a market portfolio on the frontier may generate the result
of Fama & French [669].
The absence of a relation between the ˇ coefficient and expected returns is not
confirmed using generalized least squares instead of ordinary least squares. In this
case, the estimated impact of the ˇ coefficient on expected returns is strong. Using
this method, it can be shown that there exists a positive quasi-linear relationship
(also under a grossly inefficient market portfolio proxy), with a slope equal to zero
occurring only when the mean return of the market proxy is equal to that of the
global minimum variance portfolio, see Kandel & Stambaugh [1070] and Lewellen
et al. [1213]. The relation becomes linear as the market proxy tends to the efficient
frontier and the R2 coefficient is positively related to the efficiency of the portfolio
(on maximum likelihood estimators of the CAPM see Shanken & Zhou [1528]).
In Kim [1091, 1092] it shown that the results obtained in Fama & French [669]
are biased because of the two-pass procedure employed to test the CAPM. ˇ
estimate errors in the first step result in an underestimation of the price of ˇ risk
and in an overestimation of other factors associated with variables observed without
error (size, book to market value ratio). Taking into account this phenomenon, the ˇ
coefficient has a statistically significant explanatory power for average returns and
the size factor becomes much less relevant and in some cases even insignificant (see
also Jegadeesh [1025]). On the other hand, book to market value of equity ratio
is still significant. Kan & Zhang [1063] show that, due to the error in variables
problem, a useless factor (a factor independent of all asset returns) may be priced
according to a two pass procedure. On the weak statistical significance of tests
showing the capability of explanatory variables (other than the ˇ coefficient) to
explain cross-sectionally risk premia see Lewellen et al. [1213]. In Knez & Ready
[1107] it is shown that size completely disappears when outliers are removed
from the database. Evidence that the book to market effect is driven by extreme
observations is also provided. Black [244] and MacKinlay [1269] suggest that the
anomalies can be traced back to the statistical methodology employed (e.g., data
snooping).
The relevance of the survivorship bias has been discussed in several papers.
Using a database not affected by the survivorship bias, Davis [526] shows that the
book to market value of equity ratio and the earnings-price ratio are significant for
explaining asset returns. In Chan et al. [406], Fama & French [672], Kim [1092] it
is shown that the survivorship bias of the COMPUSTAT database does not suffice to
explain the relationship between book to market value and risk premium. Moreover,
annual and monthly ˇ estimates produce the same inferences about the existence of
a positive risk premium. The size effect and the bad performance of the ˇ coefficient
in a cross-sectional regression are confirmed in a data set immune to the problems
associated with the COMPUSTAT data set, see Breen & Korajczyk [288].
Revisionists and Heretics
The literature surveyed above has shown that some of the anomalies put forward
in Fama & French [669] can be explained only partially through biases associated
to the statistical procedure employed in testing the CAPM. This evidence opened
the door to a debate, with on the one side those in favor of the hypothesis that
the anomalies provide evidence of the fact that other factors than the market return
are priced by the market and on the other side those supporting the idea of market
irrationality.
In Ball [121], Chan et al. [402], Chan & Chen [401], Fama & French [670, 671,
673], Berk [197], Chen & Zhang [425], Cochrane [464], Berk et al. [199], Cohen
et al. [472], Vassalou & Xing [1613], Pástor & Veronesi [1408], the connection
between book to market value ratio, size, earnings and returns is justified by
arguing that these variables (in particular prices) capture some risk components
not represented by the ˇ coefficient. These variables capture the financial distress
conditions of a company (high book to market value ratio and small size) and are
good proxies of future earnings. According to Fama & French [671], the positive
relation between book to market value ratio and risk premia can be explained as
follows: high book to market value companies (value companies) are less profitable
than low book to market value companies (growth companies) for at least five years
before and after portfolios are formed. As a consequence, in equilibrium high book
to market value companies are characterized by high expected returns. Moreover,
high book to market value companies are riskier than low book to market value
companies, thus yielding a distress premium. A similar interpretation has been
proposed for the size of a company in Chan & Chen [401], Berk [197]. Campbell
& Vuolteenaho [358] and Campbell et al. [350] show that the cash flows of value
(growth) stocks are particularly sensitive to permanent (temporary) movements in
aggregate stock prices. They conclude that systematic risks of individual stocks with
similar accounting characteristics are primarily driven by the systematic risks of
their fundamentals.
Gomes et al. [810], Berk et al. [199], Santos & Veronesi [1498], Zhang [1680]
theoretically show that the cross-sectional dependency of returns to book to market
ratio and to size can be rationalized in terms of a ˇ measurement problem in an
equilibrium perspective. Gomes et al. [810] provide a general equilibrium model
linking expected returns to firm characteristics such as size and book to market
ratio. A one-factor model with the market portfolio as the only factor is consistent
with the fact that the above firm characteristics predict future returns since they are
correlated with the true (stochastic) ˇ coefficient. Berk et al. [199] provide a similar
argument on the basis of a two factor model with time varying risk. Changes in
risk are linked to firm-specific variables generating the book to market and the size
effect. According to these works, there are no additional risk factors, factor loadings
are only due to measurement errors and changes in investment opportunities and
apparent risk factors are only related to the market factor.
According to this interpretation, there is a common variation in earnings which

is not captured by the ˇ coefficient. On the basis of this observation, Fama &
French [670, 673] propose a three factor linear model: a factor associated with the
market portfolio, a factor associated with firm size (excess return on a portfolio of
small firms over a portfolio of large firms) and a factor associated with the book
to market value ratio (excess return on a portfolio of high book to market stocks
over a portfolio composed of low book to market stocks). These factors are priced
(with positive risk premium) and capture the size, book to market and the other
anomalies highlighted in Fama & French [669], thus yielding a large R2 coefficient
in the cross-sectional regression. Extending the relation (5.24), the model tested by
the authors is
rQn;t rf D ˛n C nm .Qrtm rf / C ns SMBt C nh HMLt C Qn;t ;
for all n D 1; : : : ; N, where SMBt is the small minus big market capitalization risk
factor and HMLt is the high minus low value premium risk factor. SMBt measures
the return of a portfolio invested in stocks of companies with relatively small market
capitalization, while HMLt measures the return of a portfolio invested in companies
with high levels of the book to market value of equity ratio. The abnormal returns
resulting from the three factor model are not significantly different from zero for
portfolios sorted by size, book to market value, earnings price ratio. The three factor
model captures the long term reversals of returns but not the momentum effect, see
Fama & French [673] and Brennan et al. [293]. In Carhart [367], a fourth factor
associated to the stock performance over the past few months is included in order to
capture the momentum effect. Note that the size and book to market premia seem to
have diminished over time (see Cochrane [464]). The risk factor interpretation can
be rationalized inside the APT (common factors in shocks to expected earnings) or
the ICAPM (variation in the investment opportunity set). Brennan et al. [299], Liew
& Vassalou [1218], Petkova [1416], Petkova & Zhang [1417], Lettau & Wachter
[1201], Bansal et al. [134], Lettau & Ludvigson [1195], Santos & Veronesi [1497]
find empirical support for the second interpretation, while Chen [416], including
also the momentum effect, finds no support for it (only size may proxy changes of
the investment opportunity set). There is little evidence that momentum is a risk
factor.
An alternative interpretation of these anomalies is based on the presence of
irrational agents in the market, see De Bondt & Thaler [532, 533], Lakonishok et al.
[1159], Haugen & Baker [915] and Chap. 9. In this perspective, anomalies are due to
irrational traders who extrapolate the strong earnings growth of low book to market
value companies too far into the future (high market price) and the poor growth
of high book to market value companies (overreaction). Low book to market value
companies then have low average returns because future earnings growth is weaker
than expected and high book to market value companies have high average returns
because future earnings growth is stronger than expected.
The two interpretations described above go in opposite directions. The first one
(revisionists) brings the anomalies back to risk factors and, therefore, in the context
of classical asset pricing theory, proposing a multi-factor model for asset returns.
The second interpretation (heretics) emphasizes agents’ irrationality and, therefore,
the incapacity of classical asset pricing theory to explain asset risk premia (the stock
market is not efficient and assets are incorrectly priced). The empirical analysis
in Davis et al. [527], Fama & French [670, 671, 673], Lewellen [1209], Cohen
et al. [472], Fama & French [676] favors the first interpretation. MacKinlay
[1269] concludes that multi-factor pricing models alone are not enough to solve
CAPM anomalies, there is room for other explanations: market imperfections,
irrational agents or the inefficiency of the statistical methodology. Moskowitz
[1354], analysing the contribution of size, book to market and momentum to the
covariance matrix of asset returns, finds that size and market portfolio have a
significant explanatory power, while the contribution of the book to market ratio
is weak and that of momentum is negligible.
In Haugen & Baker [915], Lakonishok et al. [1159], La Porta et al. [1168], Griffin
& Lemmon [826], the empirical evidence favors a market overreaction hypothesis.
The underperformance of stocks with low book to market ratios concentrates on
earnings announcement dates. In La Porta [1167] a similar conclusion is reached
analysing expectations by stock market analysts on earnings growth rates, showing
that analysts’ forecasts about future earnings growth are too extreme because
they excessively extrapolate past earnings movements (i.e., analysts overreact).
A conclusion against the multi-factor interpretation is also provided in Daniel &
Titman [520], Daniel et al. [523], Daniel & Titman [522]: the authors argue that
the three-factor model proposed in Fama & French [670] does not capture risk
components (there are no premia for these factors), instead the model appears to
explain average returns only because the factor loadings are correlated with firm
characteristics (size and book to market value ratio). It is simply the characteristics
of the firm and not the covariance of its return with the factor that determine
the expected stock return, e.g. firms grouped according to similar characteristics
show similar properties. Evidence is provided showing that, after controlling for
the size and book-to-market ratio effects, returns are not related to loadings of the
above multi-factor model. Baker & Wurgler [110] show that investor sentiment has
strong effects on the cross-section of stock prices. This evidence agrees with the
irrational interpretation of the anomalies: firms with similar characteristics become
incorrectly priced at the same time. Ferson et al. [697], Daniel & Titman [520] and
Kan & Zhang [1063] show that attribute-sorted portfolios chosen according to an
empirically observed relation to cross-section of stock returns can appear to be risk
factors even when this is not the case.
The Conditional CAPM
The above empirical tests concentrate on an unconditional CAPM. However, the

CAPM may fail unconditionally but nevertheless hold conditionally. The CAPM
has been tested allowing for time-varying ˇ, time-varying market variance and
time-varying risk premia (conditional CAPM; see Ang & Liu [69] for an analytical
model). Allowing for time-varying ˇ and expected returns, the conditional mean-
variance efficiency of the stock index is rejected in Ferson et al. [694], but a single
risk premium model is not rejected if its premium is time-varying and the factor
is not restricted to correspond to a market factor. In Bollerslev et al. [264], the
CAPM is tested assuming a GARCH model for asset returns and it is shown that
the conditional covariance of asset returns with the market portfolio return changes
significantly over time with an auto-regressive component and the time variability
is significant in order to explain time-varying asset risk premia (time-varying ˇ).
Empirical evidence against the conditional CAPM (time-varying expected returns
and co-variances) is provided in Bodurtha & Mark [261], Harvey [908, 909],
Lewellen & Nagel [1212], Adrian & Franzoni [25], while more positive results are
obtained in Brennan et al. [299], Liew & Vassalou [1218], Petkova [1416], Petkova
& Zhang [1417], Lettau & Wachter [1201], Vassalou [1612], Avramov & Chordia
[93]. Also allowing for time-varying risk premia, the conditional CAPM does
not seem to be able to capture size and book to market effects, see He et al.
[921] (positive evidence is however reported in Ang & Chen [66]). Allowing for
heteroskedasticity in aggregate stock returns and time-varying ˇ, results against the
CAPM are obtained (large size effect) in Schwert & Seguin [1511]. These works are
unanimous in establishing that conditional co-variances change over time (we will
come back to these topics in Sect. 7.3).
The market portfolio misspecification hypothesis has been further investigated
in Jagannathan & Wang [1012]. In the papers discussed above, the proxy of the
market portfolio is usually given by a stock index. In Jagannathan & Wang [1012],
the conditional CAPM allowing for a time-varying ˇ coefficient and expected
returns is tested by also considering human capital (labor income growth is a proxy
for the return on human capital). Compared with the 1% of the cross-sectional
variation in average returns explained by the classical CAPM, the conditional
CAPM explains 30% allowing for time-varying ˇ and, when human capital is
included, the percentage of explained variation is larger than 50%, with size and
book to market value having little explanatory power. The model allows for three
betas (for the market, human capital and time variability). Positive results for the
conditional CAPM including human capital have been obtained in Jagannathan
et al. [1010], Campbell [333] and in Lettau & Ludvigson [1195], Santos & Veronesi
[1497] by handling wealth or labor income over total consumption as a conditional
variable. These models are able to capture the value effect.
5.4 The Arbitrage Pricing Theory (APT)
The Capital Asset Pricing Model provides a linear relation between the asset risk
premia and the risk premium of a single risk factor, represented by the market
portfolio. Note also that, in order to derive the CAPM relation, we had to make
assumptions on the agents’ utility functions and/or on the distribution of the asset
returns or, as in Sect. 5.2, on the fact that all the agents hold portfolios belonging
to the portfolio frontier. As we have seen in the preceding section, several empirical
5.4 The Arbitrage Pricing Theory (APT) 231
tests of the CAPM suggest that a multi-factor model is needed to explain asset
returns. In the present section, we shall establish a linear relation between the
asset risk premia and several risk factors by relying on the absence of arbitrage
opportunities and on the existence of a linear model generating asset returns, without
explicit assumptions on the agents’ utility functions. Such a relation is known as
Arbitrage Pricing Theory (APT) and goes back to the seminal contributions of Ross
(see Ross [1464, 1466]).
The existence of a linear multi-factor model generating asset returns means that
the return of every asset can be written as a linear combination of a finite number
of random variables (risk factors) together with a random component specific for
each asset (often called the idiosyncratic risk component). As the name suggests,
the APT is based on the assumption of absence of arbitrage opportunities. In
particular, when there is no idiosyncratic risk component, the APT provides an exact
linear relationship between asset risk premia and risk factors. In the presence of an
idiosyncratic risk, an analogous relation holds in an approximate form and, under
suitable additional assumptions, even in an exact form.
Let us start by discussing the case where there are no idiosyncratic risk
components. We assume that N risky assets are traded in the economy, together with
a risk free asset with return rf > 0. The returns of the N risky assets are generated
by a linear model, with respect to K risk factors . fQ1 ; : : : ; fQK /:
X
K
rQn D EŒQrn C bnk fQk ; for all n D 1; : : : ; N; (5.26)
kD1
where (without loss of generality) EŒ fQk D 0, for all k D 1; : : : ; K. The coefficient

bnk is the factor loading of the risk factor fQk on the n-th asset, for k D 1; : : : ; K and
n D 1; : : : ; N. Of course, for the linear model (5.26) to be of interest, we assume that
N > K. Letting fQ D . fQ1 ; : : : ; fQK /, relation (5.26) can be written in vector notation as
rQ D EŒQr C BfQ ;
where EŒQr is the vector of the expected returns of the N assets and B 2 RNK is the
matrix of the factor loadings.
Under the assumption of absence of arbitrage opportunities (see Definition 4.17),
the following proposition shows that the linear model (5.26) implies that asset risk
premia are determined by a linear relationship with respect to the factor loadings
matrix B.
Proposition 5.13 Suppose that asset returns are generated by the linear
model (5.26) and that there are no arbitrage opportunities. Then there exists a
vector 2 RK such that
EŒQr rf 1 D B: (5.27)
The vector satisfies k D Cov.`; Q fQk /, for all k D 1; : : : ; K, where `Q represents

the likelihood ratio of some risk neutral probability measure. Moreover, under the
non-redundancy condition rank.B/ D K, the quantity Cov. Q̀; fQk / does not depend on
the specific risk neutral probability measure chosen.
Proof For any portfolio w 2 RN (representing the proportions of wealth invested in
the N risky assets, thereby normalizing the initial wealth to 1), the associated return
is given by w> rQ D w> EŒQr C w> BfQ . Considering a portfolio w such that w> 1 D 0
(zero cost at t D 0) and w> B D 0, the absence of arbitrage opportunities implies
that w> EŒQr D 0 (otherwise it would be possible to create a portfolio with zero cost
and strictly positive return, contradicting the absence of arbitrage opportunities).
Equivalently, this means that ker.B> / \ ker.1> / ker.EŒQr > /. Hence, there exists
a couple .; 0 / 2 RKC1 such that
EŒQr D 10 C B:

If the kernel of B> is a subset of the kernel of 1> , then it holds that 1 D B , for
some 2 RK . In this case, the last relation implies that

EŒQr D B.0 C / D B 0 rf C C rf 1;
thus proving relation (5.27). Suppose now that the kernel of B> is not a subset of the
kernel of 1> , so that there exists a vector w 2 RN such that w> B D 0 but w> 1 ¤ 0.
We claim that, due to the absence of arbitrage opportunities, it holds that 0 D rf .
Indeed, suppose on the contrary that 0 > rf (the case 0 < rf can be treated in
an analogous way). Let wN WD w=.w> 1/ and consider the following strategy: borrow
one unit of wealth at the risk free rate rf and invest that unit of wealth in the N risky
assets according to the portfolio w. N This strategy has zero cost at time t D 0 and
delivers a return of wN > 1.0 rf / D 0 rf > 0 at time t D 1, thus contradicting
the absence of arbitrage opportunities.
In order to prove the second part of the proposition, recall that, in view of
Proposition 4.22, the absence of arbitrage opportunities is equivalent to the existence
of a risk neutral probability measure (or, equivalently, of a corresponding
likelihood ratio Q̀). Hence, in view of equation (4.35) together with the linear
model (5.26), we have, for all n D 1; : : : ; N,
X
K X
K
rf D E ŒQrn D EŒQrn C bnk E Œ fQk D EŒQrn C bnk Cov. Q̀; fQk /;
kD1 kD1
where we have used the fact that E Œ fQk D EŒ`QfQk D Cov. Q̀; fQk / C EŒ`Q EŒ fQk together
with the assumption that EŒ fQk D 0, for all k D 1; : : : ; K. Finally, it remains to show
that the quantity Cov.`;Q fQk / does not depend on the specific choice of the likelihood
ratio. Indeed, as in Lemma 5.5, let Ò be the unique element of I.D/ such that Q.c/ D
Oc, for all contingent consumption plans cQ taking values c 2 I.D/. Then, arguing
EŒ`Q
as in the proof of Proposition 5.6, it can be shown that
EŒÒ rQn EŒÒ EŒQrn C Cov.`;

O rQn / X
K
rf D D D EŒQrn C rf O fQk /:
bnk Cov.`;
EŒÒ EŒÒ kD1
The last two equations, together with the non-redundancy condition rank.B/ D K,
imply that Cov.`;Q fQk / D rf Cov. Ò; fQk /, for all k D 1; : : : ; K, for any likelihood ratio
`Q (and, equivalently, for any risk neutral probability measure). t
u
According to the above proposition, the risk premium of the n-th asset satisfies
the following linear relation in an exact form:
X
K
EŒQrn rf D bnk k ; for all n D 1; : : : ; N:
kD1
The coefficient k admits the interpretation of a risk premium associated to the

risk factor fQk . The risk premium k is expressed in terms of the covariance
between the risk factor fQk and the vector Ò appearing in the Riesz representation
of a pricing functional (see Lemma 5.5). In particular, this shows that the linear
relationship (5.27) does not rely on market completeness. Note also that, as can be
seen by examining the proof of Proposition 5.13, if a risk free asset is not traded in
the market, then we can still prove the existence of a couple .0 ; / 2 RKC1 such
that EŒQr D 10 C B.
As in the case of the CAPM relation (see formula (5.22)), Proposition 5.13
allows to represent the price pn of the n-th asset as the
P expectation of the dividend
discounted by the risk adjusted discount factor rf C KkD1 bnk k , i.e.,
EŒdQ n
pn D PK ; for all n D 1; : : : ; N:
rf C kD1 bnk k
Let us now extend the previous analysis by allowing for the presence of
idiosyncratic risk components in the linear multi-factor model (5.26). In this case, as
will be shown below, a linear relation between the asset risk premia and the factor
loadings holds in an approximate form and, under suitable assumptions, even in
exact form.
In order to establish an approximate linear relation, we consider a sequence of
economies indexed by the number N 2 N of available securities (for the moment,
we do not assume that a risk free asset is traded), i.e., we consider a sequence of
economies with an increasing number of assets. For each N 2 N, the asset returns in
the N-th economy are assumed to satisfy the following linear model (the superscript
N denotes the N-th economy):
X
K
rQnN D EŒQrnN C bNnk fQkN C QnN ; for all n D 1; : : : ; N; (5.28)
kD1
where, for all N 2 N, we assume that EŒQ nN D EŒQ mN QnN D 0, for all m; n D 1; : : : ; N
with m ¤ n, EŒ fQkN D 0, for all k D 1; : : : ; K, and Var.Q nN / 2 , for all n D
1; : : : ; N, for some constant 2 . These assumptions imply that the idiosyncratic risk
components are uncorrelated among themselves (note, however, that they can be
correlated with the risk factors) and have uniformly bounded variances. The linear
model (5.28) can be expressed in vector notation as follows:
rQ N D EŒQrN C BN fQ N C Q N ;
where, for all N 2 N, EŒQrN is the vector of the expected returns, BN 2 RNK is the
factor loadings matrix and Q N is an N-dimensional random vector with zero mean
and uncorrelated components. For any given portfolio wN 2 RN , representing the
proportions of wealth invested in the N available assets in the N-th economy, the
associated return is given by
.wN /> rQ N D .wN /> EŒQrN C .wN /> BN fQ N C .wN /> Q N :
In the present context, we can formulate the following definition, see Huberman
[979]. For N 2 N, we denote by 1N the unit vector in RN .
Definition 5.14 A sequence of portfolios fwN gN2N such that .wN /> 1N D 0 is said
to be an asymptotic arbitrage opportunity if the following hold:

lim EŒ.wN /> rQN D C1 and lim Var .wN /> rQ N D 0:
N!1 N!1
Intuitively, an asymptotic arbitrage opportunity represents a sequence of invest-

ment strategies such that the expected return explodes while the associated risk
(as measured by the variance) decreases to zero as the dimension of the economy
increases. We have the following proposition (see Ross [1464] and Huberman
[979]).
model (5.28), for all N 2 N, and that there are no asymptotic arbitrage
opportunities. Then, for each N 2 N, there exists a couple .N0 ; N / 2 RKC1
and a positive constant A such that
N
X X
K 2
EŒQrnN N0 bNnk Nk A; for all N 2 N: (5.29)
nD1 kD1
Proof For each N 2 N, take the orthogonal projection of the vector EŒQrN onto the
linear space spanned by 1N 2 RN and the columns of the matrix BN , so that
EŒQrN D N0 1N C BN N C cN ;
where .N0 ; N / 2 RKC1 , for some vector cN 2 RN such that .cN /> 1N D 0 and
.cN /> BN D 0. Note that
X
N N
X X
K 2
kcN k2 WD .cNn /2 D EŒQrnN N0 bNnk Nk :
nD1 nD1 kD1
Arguing by contradiction, suppose that (5.29) does not hold. Then there exists an
increasing subsequence fN 0 gN 0 2N such that
0
lim kcN k2 D C1: (5.30)
N 0 !1
0 0 0
For each N 0 2 N, define a portfolio dN WD ˛N 0 cN , where ˛N 0 WD kcN k2p for some
0
p 2 .1; 1=2/. By the properties of cN , it holds that .dN /> 1N 0 D 0. Noting that
N0 > N0 N0
.c / B D 0, the return of the portfolio d is given by
0 0 0 0 0
.dN /> rQ N D ˛N 0 kcN k2 C ˛N 0 .cN /> Q N :
0 0 0
By definition of ˛N 0 , it holds that EŒ.dN /> rQ N D kcN k2.1Cp/ , so that, by (5.30), we
have that
0 0
lim EŒ.dN /> rQN D C1
N 0 !1
and that
0 0 0 0
Var .dN /> rQN 2 ˛N2 0 kcN k2 D 2 kcN k2C4p
and, therefore,
0 0
lim
0
Var .dN /> rQN D 0:
N !1
In view of Definition 5.14, we have thus shown that the failure of (5.29) leads to an
asymptotic arbitrage opportunity, thus proving the proposition. t
u
Proposition 5.15 implies that, in an economy comprising a large number of traded
assets, for most of the assets the expected return is approximately linear with respect
to the factor loadings matrix, i.e.,
EŒQrN 1N0 C BN N :
Due to (5.29), the quality of such a linear approximation improves as the dimension
of the economy increases. Equivalently, the mean quadratic error of the linear
approximation for the expected returns is smaller than A=N and, therefore, it
decreases to zero as N increases to infinity. This result hinges on the crucial assump-
tion that the variances of the idiosyncratic risk components .Q N / are uniformly
bounded, so that, in the limit, the idiosyncratic risk can be diversified away by
choosing a sequence fwN gN2N of portfolios such that limN!1 kwN k2 D 0.
As shown in Huberman [979], if the economies are stationary, in the sense that
the vector of expected returns EŒQr as well as the factor loadings matrix B do not
depend on N, then in the absence of asymptotic arbitrage opportunities it holds that
1
X X
K 2
EŒQrn 0 bnk k < 1:
nD1 kD1
Let us now extend the result of Proposition 5.15 to the case where, besides the
risky assets, there also exists a risk free asset. More precisely, for all N 2 N, we
assume that in the N-th economy there is a risk free asset with rate of return rfN > 0.
The proof of the following proposition is analogous to that of Proposition 5.15 (see
Exercise 5.16).
model (5.28), for each N 2 N, and that there are no asymptotic arbitrage
opportunities. Assume furthermore that, for each N 2 N, a risk free asset with
rate of return rfN > 0 is traded in the N-th economy. Then, for each N 2 N, there
exists a vector N 2 RK and a positive constant A such that
N
X X
K 2
EŒQrn rf
N N
bnk k
N N
A; for all N 2 N: (5.31)
nD1 kD1
In the above approximate relations, the quantity N can be interpreted as the

vector of risk premia associated to the risk factors. Moreover, if there exist K
portfolios such that, for all k D 1; : : : ; K, the return of the k-th portfolio is perfectly
correlated (in the limit) with the k-th factor and has zero exposure with all the other
risk factors (mimicking portfolio), then Nk can be interpreted as the risk premium
of such a portfolio, see Ingersoll [999] and Admati & Pfleiderer [17]. Note that, in
general, the uniqueness of the coefficients is not guaranteed.
Let us now consider an economy with a fixed number N 2 N of risky assets and
a risky free asset (for simplicity of notation, we shall omit the superscript N in the
notation of (5.28)). In this economy with N risky assets and a risk free asset, the
notion of arbitrage opportunity is defined as in Definition 4.17. Recall that, due to
Proposition 4.22, the absence of arbitrage opportunities is equivalent to the existence
of a risk neutral probability measure. Hence, denoting by `Q the likelihood ratio of
some risk neutral probability measure, then, equation (4.35) together with (5.28)
gives
X
K
rf D EŒ`Q rQn D EŒQrn C bnk EŒ`Q fQk CEŒ`Q Qn ; for all n D 1; : : : ; N: (5.32)
kD1
Recall also that, in view of Lemma 5.5, there exists a unique vector Ò 2 I.D/ such
that Q.c/ D EŒÒ cQ for all random variables cQ with values c 2 I.D/ and for any
pricing functional Q./. Hence, we can derive a relation analogous to (5.32) in terms
of Ò as follows (compare with the proof of Proposition 5.6):
EŒÒ rQn X EŒÒ fQk EŒÒ Qn

K
rf D D EŒQrn C bnk C ; for all n D 1; : : : ; N:
EŒÒ kD1 EŒÒ EŒÒ
(5.33)
The above relations directly lead to the following result, which gives a sufficient
condition for the validity of the APT relation in exact form (compare also with
Proposition 5.13 and see also LeRoy & Werner [1191, Chapter 20]).
model (5.28) and that the residual risk Qn is uncorrelated with the factors
. fQ1 ; : : : ; fQK /, for all n D 1; : : : ; N. Assume furthermore that there are no arbitrage
opportunities and that the likelihood ratio `Q of some risk neutral probability measure
belongs to span.1; fQ1 ; : : : ; fQK /. Then the following relation holds:
X
K X
K
EŒQrn D rf bnk EŒ`Q fQk D rf Q fQk /;
bnk Cov.`; for all n D 1; : : : ; N:
kD1 kD1
Proof Note that, since `Q 2 span.1; fQ1 ; : : : ; fQK / and
EŒQ n D EŒQ n fQk D Cov.Q n ; fQk / D 0;
for all k D 1; : : : ; K and n D 1; : : : ; N, it holds that EŒ`Q Qn D 0. The claim then

follows directly from equation (5.32). t
u
Note that, in view of Lemma 5.5 and equation (5.33), the APT relation in exact
form obtained in the above proposition can also be established under the assumption
that the vector Ò appearing in the Riesz representation of a pricing functional
belongs to span.1; fQ1 ; : : : ; fQK /. If Ò … span.1; fQ1 ; : : : ; fQK / then the APT relation
only holds as an approximation, with the error term being given by EŒÒ Qn =EŒÒ ,
n D 1; : : : ; N. In view of
for equation (5.33), we call pricing error the quantity
EŒQr rf 1 C B EŒÒ fQ =EŒÒ . The following proposition provides an upper bound
for the pricing error (see LeRoy & Werner [1191, Section 20.4]).
model (5.28) and that the residual risk Qn is uncorrelated with the factors
. fQ1 ; : : : ; fQK /, for all n D 1; : : : ; N. Assume furthermore that there are no arbitrage
opportunities. Then the pricing error satisfies the following bound:
v
u N ˇ ˇ q
uXˇ X K
EŒÒ fQk ˇˇ2
t ˇEŒQr r C O OfQ 2 ;
ˇ n f bnk
O ˇ ˇˇ O ˇˇ E ` ` (5.34)
nD1 kD1 EŒ` EŒ`
where Ò 2 I.D/ is the vector appearing in the Riesz representation of a pricing

functional Q (see Lemma 5.5) and ÒQf is the orthogonal projection of Ò onto the
linear space span.1; fQ1 ; : : : ; fQK /.
Proof As in Lemma 5.5, let Ò be the vector in the Riesz representation of a pricing
functional (which exists due to the assumption of absence of arbitrage opportunities,
see Theorem 4.20). Denote by ÒQf the orthogonal projection (with respect to the
norm induced by the inner product .Qx; yQ / 7! EŒQxyQ ) of Ò onto the linear space
span.1; fQ1 ; : : : ; fQK /. As a consequence, Ò can be decomposed as Ò D ÒQf C .Ò ÒQf /.
Moreover, since
Ò 2 span.rf ; rQ1 ; : : : ; rQN / span.1; fQ1 ; : : : ; fQK / ˚ span.Q 1 ; : : : ; QN /;
as follows from equation (5.28), it holds that Ò ÒQf 2 span.Q 1 ; : : : ; QN /. Hence,

PN
there exists a vector ˛ 2 RN such that Ò ÒQf D nD1 ˛n Qn , noting that the
elements .Q 1 ; : : : ; QN / are linearly independent and represent a basis for the linear
space span.Q 1 ; : : : ; QN /. Since EŒQ i Qj D 0 for all i; j 2 f1; : : : ; Ng with i ¤ j, this
implies that, for all n D 1; : : : ; N,
XN
E Ò Qn D E .Ò Òf C Òf /Q n D E .Ò Òf /Q n D
Q Q Q
˛i EŒQ i Qn D ˛n Var.Q n /
iD1
and
Q 2
X
N
E Ò Òf D ˛n2 Var.Q n /:
nD1
ˇ ˇ
In view of equation (5.33), the pricing error of asset n is given by ˇEŒÒ Qn =EŒÒ ˇ, so
that, using the fact that Var.Q n / for all n D 1; : : : ; N,
N ˇ ˇ
X ˇ X EŒÒ fQk ˇˇ2 1 X O 2 1 X 2
K N N
ˇEŒQrn rf C b D E `
Q D ˛n Var.Q n /2
ˇ nk
O ˇ O 2 n
O 2
nD1 kD1 EŒ ` EŒ` nD1 EŒ` nD1
2 X 2 2 O OQf 2
N
˛n Var.Q n / D E `` ;
EŒÒ 2 nD1 EŒÒ 2

u
The upper bound on the pricing error given in the above proposition is partic-
ularly interesting. Indeed, it shows that the pricing error will be small when the
vector Ò is close to its projection ÒQf on the factor span. In particular, if it holds that
Ò 2 span.1; fQ1 ; : : : ; fQK /, then Ò D ÒQf , so that the pricing error is null, thus confirming
the validity of the APT in exact form as obtained in Proposition 5.17. Moreover,
note that the upper bound given in (5.34) does not depend on the number N of
traded assets, as in Proposition 5.16. In the limit, as the dimension of the economy
(i.e., the number of traded assets) increases to infinity, the average pricing error will
become arbitrarily small.
Extensions of the Arbitrage Pricing Theory
The Arbitrage Pricing Theory has been extended in several directions (see also
Connor [476]). In particular, the assumption that the residual risks are uncorrelated
can be significantly relaxed. To this regard, note that the result of Proposition 5.17
does not rely on the absence of correlation among the residual risks, but only on
the absence of correlation between the residual risks and the factors . fQ1 ; : : : ; fQK /.
In Ingersoll [999], the APT in the form of Proposition 5.16 has been generalized
to the case where the residual risks QnN are possibly correlated among themselves
(but uncorrelated with the risk factors), under the assumption that the elements
of the factor loadings matrix are uniformly bounded. In this case, letting ˙ N be
the covariance matrix of the residual risks, it can be shown that the absence of
asymptotic arbitrage opportunities implies that there exists a positive constant A
such that
>
EŒQrN N0 1N BN N .˙ N /1 EŒQrN N0 1N BN N A; for all N 2 N:
(5.35)
In particular, if the matrix ˙ N is diagonal (i.e., the residual risks are uncorrelated),
this implies that
N P
1 X EŒQrnN N0 KkD1 bNnk Nk 2
lim p D 0:
N!1 N
nD1 Var.Q nN /
Note that this result does not rely on the assumption that residual risks have
uniformly bounded variances, as considered in Proposition 5.15.
An extension of the APT to the case of correlated residual risks (uncorrelated
with the risk factors) has also been proposed in Chamberlain [395] and Cham-
berlain & Rothschild [397]. In this case, in order to establish a result similar to
Proposition 5.15, two hypotheses on the idiosyncratic risk components and on
the factors are introduced. Indeed, in the presence of correlated residual risks,
the idiosyncratic risk component is not necessarily eliminated via diversification
by investing a small proportion of wealth in each asset. We say that a sequence
fwN gN2N of portfolios is diversified if it holds that .wN /> 1N D 1, for all N 2 N,
and limN!1 kwN k2 D 0. The first hypothesis requires that the influence of the
residual risks QnN , for n D 1; : : : ; N and N 2 N, can be eliminated in the limit via
diversification, in the sense that limN!1 EŒ..wN /> Q N /2 D 0 for every sequence
fwN gN2N of diversified portfolios. As shown in Chamberlain [394], this hypothesis is
satisfied if the largest eigenvalue of the covariance matrix ˙ N of the residual risks is
uniformly bounded from above over all N 2 N. The second hypothesis is called the
pervasiveness condition and amounts to assume that each factor influences a large
number of assets in the N-th economy, for each N 2 N. This condition requires that
the smallest eigenvalue of the matrix BN .BN /> explodes as N increases to infinity.
Under these two hypotheses, asset returns can be described by an approximate K-
factor model and relations analogous to (5.29) or (5.35) can be established.
Equilibrium versions of the APT have been proposed in Chen & Ingersoll [423],
Connor [475], Dybvig [605]. In an economy with a finite number of assets, a linear
relation for the asset risk premia holds in exact form if in equilibrium there exists a
perfectly diversified portfolio (i.e., not influenced by idiosyncratic risk) representing
the optimal choice of an agent (under suitable regularity assumptions on the agent’s
utility function), as shown in the following proposition (see also Chen & Ingersoll
[423]).
Proposition 5.19 Suppose that the economy comprises a risk free asset with return
rf and N risky assets, whose returns are generated by the linear model (5.28) with
(possibly correlated) residual risks .Q 1 ; : : : ; QN / satisfying
EŒQ n j fQ1 ; : : : ; fQK D 0; for all n D 1; : : : ; N:
Then, if there exists a portfolio which is unaffected by the residual risks and
represents the optimal choice for an agent with a continuously differentiable,
increasing and strictly concave utility function, the APT holds in exact form.
Proof Let us denote by W e the agent’s optimal wealth. Under the present assump-
e
tions, W can be expressed as a function of the risk factors, so that u0 .W
e / D
Q Q e
g. f1 ; : : : ; fK /, since W is assumed not to be affected by the residual risks
.Q 1 ; : : : ; QN /. From the first order conditions of the agent’s optimal portfolio problem
(see equation (3.4)), it holds that

e /.Qrn rf / D E g. fQ1 ; : : : ; fQK /.Qrn rf / D 0;
E u 0 .W for all n D 1; : : : ; N:
Replacing equation (5.28) into the above expression we get

X
K
E g. fQ1 ; : : : ; fQK / EŒQrn rf C bnk fQk C Qn D 0;
kD1
so that, by rearranging the terms, we get, for all n D 1; : : : ; N,

!
X
K
EŒg. fQ1 ; : : : ; fQK /fQk EŒg. fQ1 ; : : : ; fQK /Q n
EŒQrn D rf C bnk :
kD1
EŒg. fQ1 ; : : : ; fQK / EŒg. fQ1 ; : : : ; fQK /
The last term is null, since

E g. fQ1 ; : : : ; fQK /Q n D E g. fQ1 ; : : : ; fQK /EŒQ n j fQ1 ; : : : ; fQK D 0;

u
5.5 Empirical Tests of the APT 241
In an economy with a finite number of assets, an error bound for each asset has
been established in a general equilibrium setting by Dybvig [605] and Grinblatt &
Titman [834]. If markets are complete, idiosyncratic noise components are mutually
independent and independent of fQk , the (representative) agent has an increasing and
strictly concave utility function with a non increasing and bounded coefficient of
absolute risk aversion, each asset is in positive net supply, then the market portfolio
is diversified (in the sense that idiosyncratic risk is diversified away) and a bound
to the approximation error holds for each asset, see Dybvig [605]. The bound to
the approximation error of an asset depends on the variance of its idiosyncratic
risk component, on the bound to the coefficient of absolute risk aversion and its
weight on the market portfolio. In an economy with an infinite number of assets,
Connor [475] provides an exact APT equilibrium version if the market portfolio
is well diversified. In this case, every investor holds a well diversified portfolio (K
mutual funds separation, on these results see also Milne [1345]). In this perspective,
a problem is represented by Roll’s critique: to test an equilibrium version of the APT,
the market portfolio should be observable (see Shanken [1524]). In the special case
where consumption and idiosyncratic risk components are distributed according to
a bivariate normal law, then, if the number of assets grows in such a way that the
weight of each asset in the market portfolio tends to zero, the residual also tends to
zero, see Constantinides [492].
5.5 Empirical Tests of the APT
The APT cannot be easily tested in its approximate form and, as a consequence,
most of the empirical studies concentrate on testing the exact linear relation
associated with the APT. A strong critique to testing the APT in exact form was
put forward in Shanken [1523] (Shanken critique). The returns of two sets of assets
generating the same set of portfolio returns may conform to different factor models
and even the number of the factors can differ. In particular, the risk factors for
a set of returns cannot be identified in a unique way, since they depend on the
assets generating them. As a consequence, the implications of the APT in exact
form on portfolio risk premia change as reference assets (and factors) change,
see also Gilles & LeRoy [781]. The risk premia implications are consistent only
if the assets have the same expected return. Transforming assets as in Shanken
[1523], factor and idiosyncratic randomness are handled as idiosyncratic noise,
making the factor model useless. In a finite economy, an approximation bound is
a mathematical tautology with no economic content and, therefore, it is untestable.
In Reisman [1445], this critique was further examined, showing that if the APT
approximation is to be interpreted as an equality, then expected returns should be
a linear function of the ˇ coefficients with respect to virtually any set of reference
variables correlated with the true factors. The upper bound to the approximation
error increases as the correlation between the true factors and the reference variables
decreases (Nawalkha [1372] has shown that under some conditions no loss in pricing
accuracy occurs decreasing the correlation with the true factors). The factor model
can be manipulated rather arbitrarily by repackaging a given set of securities, so
that factors are in practice indeterminate. This is an intrinsic limit of the APT in
exact form. These results have been discussed in Grinblatt & Titman [835] and
Dybvig & Ross [609], where it is shown that if the APT holds in exact form then
the factor model cannot be manipulated as suggested in Shanken [1523]: as the
number of assets of the economy tends to infinity, the variance of the returns of
some assets becomes unbounded or the inverse of the transformed matrix explodes.
Al-Najjar [38] has shown that in a large economy repackaging identifies a unique
factor model. For a reply to the above arguments see Shanken [1524]. The Shanken
critique suggests considering the approximation error in empirically testing the APT
and, therefore, testing models deriving an approximation error bound for each asset.
This is accomplished through equilibrium versions of the APT, as discussed at the
end of the previous section.
Some methodological considerations on testing the CAPM also apply to APT
tests. The APT is a single-period model, therefore one assumes that the factor
model holds period by period and that the economy is stationary. The classical
hypothesis is that the asset returns vector conditionally on the factors is identically
and independently distributed over time as a multivariate normal random variable.
As for the CAPM, the (approximate) linear relation of the APT is an ex-ante result,
but the factor model is an ex-post model. To test the model one assumes that agents
form rational expectations and that the expected return of an asset is equal to the
historical average return.
In general, the APT does not provide an exact formula for asset risk premia. In a
finite asset economy, the APT provides an approximate linear relation between asset
risk premia and factor risk premia: an exact linear relation holds if a portfolio of
the factors belongs to the mean-variance frontier or if there exists a well diversified
optimal portfolio. Moreover, unless an equilibrium version of the APT is tested,
the theory does not provide a bound to the approximation error for each asset, but
only a bound to the global approximation error. Empirically, the APT in exact form
is usually tested without checking for the validity of the hypothesis yielding an
exact pricing relation. Because of the non-uniqueness factor representation result
obtained in Shanken [1523], it is difficult to test a violation of the bound on the
approximation error.
Imposing the restrictions of the APT in exact form on the multi-factor
model (5.28), the following ex-post model is obtained:
X
K
rQn;t D 0 C bnk .k C ıQk;t / C Qn;t ; for all n D 1; : : : ; N:
kD1
As for the CAPM, the APT in exact form concerns a linear relation between asset
risk premia, factor risk premia and the matrix B. There are two main classes of tests
of the APT: cross-sectional regressions and time series regressions. Given a factor
structure, in the first case risk premia () are estimated taking as observed the matrix
5.5 Empirical Tests of the APT 243
B, in the second case a joint estimation is performed (i.e., constrained time series
estimate of B and ). The literature on empirical tests of the APT is quite large and
we refer to Connor & Korajczyk [481] for a survey. In what follows, we concentrate
our attention on cross-sectional regression tests.
A cross-sectional regression test consists of three steps: a) risk factor identifica-
tion; b) estimation of the matrix B; c) test of a linear relation between risk premia and
the coefficients in B. As for the CAPM, a two-pass procedure or a joint estimation
procedure is adopted (i.e., betas and relative prices are simultaneously estimated).
The methodology to estimate the matrix B and to test the APT depends on the nature
of the factors, see Campbell et al. [348]. In some cases, factors are traded portfolio
returns (when this is not the case we can look for portfolios mimicking the factors,
as explained in the previous section). Given the multi-factor model and the estimated
matrix B, risk premia can be estimated by a variety of methods including ordinary
least squares and generalized least squares. The two-pass procedure may induce an
error in variables problem, see Shanken [1526].
The sequence of steps described above implies that a test of the APT is a joint test
of three different hypotheses: a) risk factors are correctly identified; b) the matrix
B is correctly estimated; c) the exact linear relation between asset risk premia and
coefficients holds.
The risk factor identification problem is a model selection problem. Two main
classes of factor models have been considered in the literature, depending on the
selection approach: statistical factor models and economic factor models. Let us
first consider statistical factor models. There are two main statistical approaches to
identify the factors: factor analysis and principal component analysis. In both cases,
factors do not have a direct economic interpretation. A factor analysis procedure
has been used in Roll & Ross [1461], Chen [422], Lehmann & Modest [1178] and
a principal component analysis procedure in Connor & Korajczyk [478] and Jones
[1038]. Both methods produce a consistent estimate of B in large samples, while in
finite samples there is no clear argument to choose between them. Factor analysis is
computationally more expensive than principal component analysis. The number of
selected factors is three/four in Roll & Ross [1461], Chan et al. [408], five in Chen
[422], Lehmann & Modest [1178] and six in Connor Korajczyk [480]. Factors are
chosen by repeating the estimation procedure and varying their number. A test on
the number of factors has been proposed in Connor Korajczyk [480]. Typically, no
economic interpretation can be given to risk premia signs.
There are two main classes of economic factor models: models with factors
referring to firm characteristics and models with factors referring to macroeconomic
variables. The first type of factor models was stimulated by the empirical tests of
the CAPM showing that factors such as size, book to market value ratio, dividend
yield, cash flow to price ratio, turn out to be significant in order to explain asset
risk premia, as discussed in Sect. 5.3 (see also Fama & French [670, 673], Chan
et al. [408]). Three factors (a market factor, one associated with size and one
with book to market value ratio) explain sufficiently well the cross-section of
asset risk premia. Building a linear model with macroeconomic factors, one tries
to identify those macroeconomic-financial market factors affecting a company’s
value (i.e., affecting future earnings and the discount factor). In Chen et al. [424]
the following five variables have been considered: industrial production growth,
unexpected inflation, expected inflation, yield spread between long and short interest
rates (maturity premium), yield spread between corporate high and low-grade bonds
(default premium). Aggregate consumption growth (as suggested by the CCAPM),
market factor (as suggested by the CAPM) and oil price are not significant. Many
other macroeconomic factors have been considered in empirical studies. In other
studies, an index of the market turns out to be significant. An APT model based
on macroeconomic factors has been tested with positive evidence in Burmeister &
McElroy [323] and Elton et al. [638].
The APT can be tested in several directions. In a cross-sectional regression
setting, given the factors identified through one of the procedures illustrated above
we can verify that risk factors are priced (test the statistical significance of the risk
premia associated with the factors). In Roll & Ross [1461] it is shown that factors are
significant: in the macroeconomic factor model analysed in Chen et al. [424] four
factors are significant, only the term spread factor is marginally significant. Another
test consists in verifying that other variables are not significant in explaining risk
premia. In Roll & Ross [1461] it is shown that the standard deviation of an asset
return has no incremental power over the four factors. The result is confirmed in
Chen [422]. As far as the size of the company is concerned, results are controversial:
in Chen [422] and Chan et al. [402] it is shown that the size anomaly is explained
by the macroeconomic APT model proposed in Chen et al. [424]. On the contrary,
in Reinganum [1441], Brennan et al. [293] it is shown that the size anomaly is not
explained by the APT. The evidence on the other anomalies relative to the CAPM
(e.g., book to market value of equity ratio) is mixed, see Brennan et al. [293].
Another strategy to test the APT consists in verifying that the coefficients are
constant as the set of returns of the economy changes. As reference returns vary,
the multi-factor model as well as risk premia change (however if there exists a
risk free asset then 0 D rf for every set of returns). In Roll & Ross [1461], this
hypothesis has not been rejected, while a negative result has been obtained in Brown
& Weinstein [316]. Using the time series test approach, a testable implication of the
APT in exact form is that 0 D rf which is equivalent to the condition established
in Grinblatt & Titman [836]. A restriction on the matrix B is derived using results
in Huberman & Kandel [982]. Mixed results for exact pricing restrictions were
provided in Lehmann & Modest [1178] and Connor & Korajczyk [478], showing
that the APT does not explain the size effect. In Connor & Korajczyk [478] the
authors compare the APT with the CAPM: the results are mixed with a slight
preference for the APT (the same conclusion is reached in Mei [1322]).
In Chen [422] the APT is compared with the CAPM in a cross-sectional
regression setting. Expected returns are estimated according to the APT with five
factors and to the CAPM: the APT explains returns better than the CAPM. Residuals
from the CAPM cross-sectional regression can be explained by factor loadings
employed in the APT, while residuals from the APT cannot be explained by the
market portfolio ˇ. Data support the APT as a better model for asset returns.
A similar conclusion is drawn in Fama & French [673]. The performance of the
three classes of factor models has been evaluated in Connor [477]. The ranking is
as follows: factor models based on firm’s characteristics, statistical factor models,

macroeconomic factor models. Firm-specific factors perform well compared with
macroeconomic variables and statistical factors, see also Chan et al. [408] and
Moskowitz [1354].
Summing up, empirical tests often reject exact linear APT pricing restrictions.
Comparing the APT to the CAPM, the APT performs well in explaining cross-
sectional differences in asset returns. Moreover, the APT seems to be able to explain
some pricing anomalies relative to the CAPM.
Dybvig & Ingersoll [607] investigate the relation between the CAPM and the
complete markets equilibrium model. The results are striking: in particular, if the
CAPM pricing relation holds for all assets, the markets are complete and the market
portfolio generates sufficiently high returns in some state of the world, then there
exist arbitrage opportunities. As a consequence, if agents’ utility functions are
always increasing, then in equilibrium the CAPM relation cannot hold for all the
assets. If asset returns are unbounded, then the CAPM can hold in equilibrium only
if each investor has achieved his level of satiation. Under the complete markets
assumption with unbounded returns, if the CAPM holds for all the assets, then in
equilibrium agents exhibit quadratic utility functions.
In our analysis, we have focused our attention on asset pricing relations in
correspondence of an equilibrium allocation. The existence of the equilibrium
can be established by relying on the results presented in Chap. 1. Clearly, an
assumption which implies that the agents hold portfolios belonging to the mean-
variance efficient portfolio frontier is that the agents’ preferences are represented by
mean-variance utility functions (see Sect. 2.4). However, this assumption might be
problematic in order to establish the equilibrium since quadratic preferences are not
always monotone, see Nielsen [1376]. The existence of an equilibrium in a market
with a risk free asset is ensured if all the agents agree on the asset expected returns or
if their coefficient of absolute risk aversion goes to infinity as the standard deviation
of a portfolio goes to infinity. Without a risk free asset, a sufficient condition for
the existence of an equilibrium is that the agents agree on the expected returns
of the assets and that their risk aversion coefficients are bounded. In particular,
this last condition eliminates satiation problems (on this point see Allingham [51]
and Nielsen [1375]). Non-monotonicity of preferences can originate negative prices
in equilibrium. Imposing bounds on agents’ risk aversion, the positivity of the
equilibrium prices can be proved as in Nielsen [1377].
The CAPM admits a straightforward generalization to the case where agents
exhibit heterogeneous beliefs about the expected returns of the risky assets (nev-
ertheless, all the agents are assumed to have identical beliefs about the covariance
matrix V of the traded assets). More precisely, we suppose that, from the point of
view of agent i, the expected return of asset n is given by in , for i D 1; : : : ; I and
n D 1; : : : ; N. In order to derive an analogue to the CAPM relation (5.21) in the case
of heterogeneous beliefs, we suppose that the agents are characterized by quadratic

utility functions of the form
bi 2
ui .x/ D x x ; for all i D 1; : : : ; I:
2
Denoting by W e i the optimal wealth at time t D 1 of agent i, for i D 1; : : : ; I,
we suppose that EŒW e i < 1=bi , for all i D 1; : : : ; I (in order to ensure that
the optimal portfolios lie in the region where the utility functions are increasing).
Following the notation adopted in section “Closed Form Solutions and Mutual Fund
Separation”, let us denote by wi n the optimal amount of wealth invested at time
t D 0 by agentP i in
P the n-th asset, for i D 1; : : : ; I and n D 1; : : : ; N. Note that
the quantity NnD1 IiD1 wi n represents the total market capitalization of the risky
assets. Consequently, for each n D 1; : : : ; N, the relative market capitalization of
asset n is given by the quantity
PI i
iD1 wn
wmn WD PN PI i
;
kD1 iD1 wk
for all n D 1; : : : ; N. Equivalently, .wm 1 ; : : : ; wN / denotes the market portfolio

m
in terms of proportions of the total market capitalization invested in each of the

assets. The return rQ m associated to the market portfolio is then given by
risky P
rQ D NnD1 wm
m
Qn . Let us also introduce the following notation, for all n D 1; : : : ; N:
n r
X
I
i
N n WD PI in ;
iD1 jD1 j
with i WD .1 bi EŒWe i /=bi , in view of the explicit characterization of the optimal

portfolio given in equation (3.9). Recall also that, as explained in section “Closed
Form Solutions and Mutual Fund Separation”, the quantity i represents the inverse
of the coefficient of absolute risk aversion of agent i evaluated in correspondence of
his expected optimal wealth. Let us also define
X
I
i X
N
N WD
M
PI i;M
; where i;M
WD n n ;
wm i
iD1 jD1 j nD1
We are now in a position to state the following result, the proof of which is given in
Exercise 5.17 (see Hens & Rieger [938, Proposition 3.5]).
Proposition 5.20 Under the above assumptions and using the notation introduced
above, the following CAPM relation holds under heterogeneous beliefs:
N n rf D ˇnm .N M rf /; for all n D 1; : : : ; N (5.36)
where ˇnm WD Cov.Qrm ; rQn /= Var.Qrm /, for all n D 1; : : : ; N.

Equation (5.36) represents the Security Market Line in an economy where agents
have heterogeneous expectations about the asset returns. Note that the expected
returns N n and N M appearing above represent weighted averages of the expected
returns as perceived by the individual agents, with the weights depending on the
risk aversion of the individual agents. In other words, expected returns as perceived
by less risk averse agents play a more important role in the determination of
.N 1 ; : : : ; N N / and of N M than the expected returns as perceived by more risk averse
agents.
The CAPM has been extended in several directions. In particular, in Brennan
[289] and Litzenberger & Ramaswamy [1222] the model has been extended to an
economy with dividends and taxes and in Mayers [1313] to an economy with non-
tradeable assets and human capital. Furthermore, an analysis of the CAPM with
short sales prohibition is given in Sharpe [1530]. An intertemporal version of the
CAPM has been introduced in Merton [1330] (for more details, see also Sect. 6.4).
Foreign exchange risk has been introduced in the CAPM by Solnik [1555] and in
Stulz [1574].
For a survey on the econometric issues and techniques for testing the CAPM,
we refer the reader to Shanken [1527] and Campbell et al. [348]. Shanken [1526]
provides an econometric view of maximum likelihood methods and of traditional
two pass approaches to estimating factor models, also addressing the errors-in-
variables problem. The traditional inference procedure, under standard assumptions,
overstates the precision of price of risk estimates and in Shanken [1526] an
asymptotically valid correction methodology is provided.
The assumption of a normal distribution plays a particularly important role in
the literature related to the CAPM. On the one hand, this assumption brings the
advantage that finite sample properties of asset pricing models can be derived, on the
other hand the normal distribution is only one of the distributions compatible with
the CAPM and there is a substantial evidence on the non-normality of asset returns
(for instance, heavy tails and heteroskedasticity). An alternative distributional
assumption compatible both with the CAPM and with the presence of heavy tails
in asset returns is that of a multivariate t-Student distribution. By employing a
generalized method of moments estimator, MacKinlay & Richardson [1271] show
that under this assumption the bias introduced by the normality hypothesis is small,
but it can be relevant when the Sharpe ratio of a portfolio is high and/or the number
of degrees of freedom is small. Examples of the time series approach to test the
CAPM and the zero-ˇ CAPM are provided in Jobson & Korkie [1033], MacKinlay
& Richardson [1271], Gibbons et al. [777], Gibbons [775], Shanken [1527], Ferson
[687]. Tests can be performed by means of a maximum likelihood estimator
assuming a given distribution for the asset returns (multivariate normality or t-
Student) or by means of a generalized method of moments estimator (without
distributional assumption).
As reported in Sect. 5.3, the empirical evidence on the CAPM has produced
mixed results and several anomalies have been discovered. In recent years, other
anomalies have been detected taking the CAPM as the benchmark. Brennan et al.
[293], Chordia et al. [439], Datar et al. [525], Liu [1229], Ang et al. [68] have
shown a negative and significant relation between returns and turnover rate, trading
volume and volatility, taken as proxies of market liquidity. A relation also holds
for other measures of market liquidity (e.g., the price impact of trades and fixed-
variable cost of trades), see Brennan & Subrahmanyam [297] and Amihud [57].
In particular, according to Brennan & Subrahmanyam [297], there is an increasing
relation between adverse selection, illiquidity costs (variable transaction costs) and
asset returns. Easley et al. [615] find a positive relationship between the probability
of information-based trades and asset returns (adverse selection premium), with the
probability of informed trading being lower for high volume stocks, see Easley et al.
[618]. In the analysis of Duarte & Young [591], it has been shown that this effect is
mainly due to illiquidity and not to adverse selection. Hou & Moskowitz [968] find
that a positive risk premium is associated with the delay with which prices respond
to information, an effect which is not explained by liquidity and other microstructure
effects. Pástor & Stambaugh [1407] find that the stocks whose returns are more
exposed to marketwide liquidity fluctuations are characterized by higher expected
returns. In Lamont et al. [1163] it is shown that there is a financial constraint factor,
in the sense that constrained firms earn lower returns than unconstrained firms.
Naranjo et al. [1370] and Brennan et al. [293] show that stock returns are increasing
in dividend yields. Ang et al. [67] show that returns are increasing in downside
risk (i.e., the risk that asset returns are more correlated with the market when the
latter is failing rather than when the market is rising). Harvey & Siddique [910]
and Boyer et al. [275] show that conditional skewness helps to explain the cross-
sectional variation in expected returns and negative co-skewness with the market
return induces a positive risk premium. See Chen et al. [417] for an analysis of the
relationship between return skewness and other anomalies. Hou & Robinson [969]
show that firms in more concentrated industries earn lower returns and Dittmar [579]
finds that kurtosis affects asset returns. In Diether et al. [577], Yu [1668], Sadka &
Scherbina [1489], Anderson et al. [61, 62], Zhang [1677], Johnson [1036], Doukas
et al. [583] it is shown that analysts’ disagreement is related to expected returns.
Note that many of these anomalies are not captured by the Fama & French [670]
model.
For international evidence on the CAPM anomalies we refer to Hawawini &
Keim [916], Fama & French [674], Haugen & Baker [915], Rouwenhorst [1479],
Asness et al. [79]. Fama & French [674] propose an international two-factor model
with a factor for relative distress in order to capture the international evidence of a
value effect (see also Fama & French [677] and Griffin [825]). A positive evidence
on the international version of the conditional CAPM is reported in De Santis &
Gerard [560].
Considering a non-linear habit formation model, Santos & Veronesi [1498] show
that cross-sectional regularities can be explained by the variability of the cash
flows of the companies. In Yogo [1665] cross-sectional anomalies are explained by
considering a non-separable utility function. Zhang [1680], Chen & Zhang [419],
Liu et al. [1228] explain asset price anomalies by assuming costly reversibility of
investments through a q-theory model. Bansal et al. [134, 135], Hansen et al. [889]
show that value-growth anomalies can be reconciled with an intertemporal model
assuming recursive preferences (see Chap. 9) and considering long run risk: the
cash flow growth of value (growth, respectively) portfolios has positive (negligible,
respectively) covariation with consumption in the long run.
For a survey of theoretical and empirical results on the APT we refer to
Connor [476], Huberman [980], Connor & Korajczyk [481]. APT restrictions
can be obtained by allowing for private information among the agents on asset
returns, see Stambaugh [1562] and Handa & Linn [886]. It can be shown that,
under some conditions on the preference relation, there are no asymptotic arbitrage
opportunities in the market. The conditions include the existence of an agent with
strictly increasing and continuous preferences and of an optimal portfolio (see
Jarrow [1019]). Bounds on the pricing kernel in the spirit of the Hansen-Jagannathan
bound can be found in Bansal & Lehman [136], Bernardo & Ledoit [204], Cochrane
& Saa’-Requejo [468], Snow [1553]. Huberman et al. [984] and Huberman &
Kandel [982] have identified necessary and sufficient conditions for the existence
of mimicking portfolios: there exists a set of mimicking portfolios if and only if
the minimum variance portfolio has positive systematic risk, in the sense that it is
affected by the risk factors.
It is of interest to study under which conditions one can obtain a linear relation
in an exact form, in the spirit of Propositions 5.13 and 5.17. In Ingersoll [999],
assuming an approximate factor model, it is shown that in the limit as the dimension
of the economy approaches infinity (i.e., N ! 1/, the returns on all well
diversified portfolios satisfy a linear relation in exact form. Along the same lines,
in Chamberlain [394] it is shown that in the limit as the dimension of the economy
approaches infinity a linear pricing relation holds in exact form if and only if there
exists a well diversified portfolio which belongs to the mean-variance frontier. A
related result has also been obtained in Grinblatt & Titman [836] for an economy
with a finite number of assets and a risk free asset. Indeed, assuming that the
N traded assets have a non-singular covariance matrix, Grinblatt & Titman [836]
consider a set of K N reference portfolios, i.e., a set of K portfolios such that
there exists a linear combination of them which does not coincide with the minimum
variance portfolio constructed from the N assets. Interpreting the K reference
portfolios as risk factors, it can be shown that the linear relation of the APT holds in
exact form if and only if there exists a combination of the K reference portfolios
which belongs to the mean-variance efficient portfolio frontier. A similar result
has also been obtained in Huberman & Kandel [982], discussing the relationship
between exact arbitrage pricing and K mutual funds separation. It is worth pointing
out that these results provide a bridge between the CAPM, which relies on the fact
that the market portfolio belongs to the portfolio frontier, and the APT in exact
form, which relies on the fact that a combination of reference portfolios belongs
to the portfolio frontier (as a consequence, similar methods aiming at verifying
the efficiency of a portfolio can be employed to test the CAPM and the APT, see
Shanken [1525]). Reference portfolios playing the role of risk factors are often
called mimicking portfolios.
The asset pricing models presented in this chapter are all linear models. A non-
linear model with the market return playing the role on the only risky factor has
been proposed in Bansal & Viswanathan [138] and a multi-factor non-linear model
has been introduced in Dittmar [579]. The empirical performance of such non-linear
models turns out to be good and outperforms linear models. Moreover, it is shown
that both human capital and the market return are relevant risk factors.
5.7 Exercises
Exercise 5.1 Consider the asset pricing relation (5.6). Prove that the covariance
between the return on the market portfolio and the stochastic discount factor
ıu01 .Qxm /=u00 .xm
0 / is negative. Deduce that the risk premium EŒQ
rm rf of the market
portfolio is positive.
Exercise 5.3 As in Sect. 5.1, let us consider an economy comprising I individuals
with increasing and strictly concave utility functions ui , N risky assets with normally
distributed returns and a risk free asset with risk free rate of return rf . Denoting by
e i the optimal consumption of agent i, for i D 1; : : : ; I, define the global absolute
W
risk aversion coefficient as the quantity
00
EŒui .We i /
i WD :
0
EŒui .We i /
By relying on equilibrium arguments, express the risk premium of the market

portfolio in terms of the global absolute risk aversion coefficients of the agents.
Exercise 5.4 In the setting of Exercise 5.3, remove the assumption that asset returns
are normally distributed and assume instead that the I agents have quadratic utility
functions of the form
bi 2
ui .x/ D ai x x; for all i D 1; : : : ; I:
2
Assume furthermore that the initial endowments of the agents and the asset returns
are such that the optimal consumption plans of the agents always remain in the range
e i < ai =bi , for all i D 1; : : : ; I). By
where the utility functions are increasing (i.e., W
relying on equilibrium arguments, express the risk premium of the market portfolio
in terms of the global absolute risk aversion coefficients of the agents.
Exercise 5.5 Consider an economy populated by I agents with exponential utility
functions of the form
1
ui .x/ D eai x ; with ai > 0 for all i D 1; : : : ; I:
ai
5.7 Exercises 251
Suppose that there are two traded assets: a risk free asset with rate of return rf
and a risky asset with return distributed according to a normal law with mean
and variance 2 . Suppose that the aggregate supply of the risky asset is equal to 1.
Determine the risky asset risk premium in correspondence of an equilibrium of the
economy.
Exercise 5.6 Consider a representative agent economy as in Sect. 5.1. Show that
the risk premium on the market portfolio is increasing with respect to the absolute
risk aversion coefficient of the representative agent and, if the representative agent’s
utility function is DARA, then it is decreasing with respect to the economy’s
aggregate wealth.
Exercise 5.7 Suppose that there are no arbitrage opportunities and let Q./ be a
pricing functional. Let Ò 2 I.D/ denote the unique vector such that Q.c/ D EŒ`c, O
for all c 2 I.D/, as in Lemma 5.5. Let z 2 R be the portfolio such that ` D Dz`
` N O
O
and denote by rQ` its return, i.e., rQ ` D `=V. Ò/. Prove the following properties:

(i) for any arbitrary portfolio z 2 RN with return rQ z , it holds that E rQ z rQ ` D
` 2
E .Qr / .
(ii) the portfolio z` belongs to the mean-variance portfolio frontier.
Exercise 5.8 In an economy with N C 1 traded assets, suppose that there are no
arbitrage opportunities and let `Q be the likelihood ratio of any risk neutral probability
measure. Call the quantity m Q WD `=r Q f stochastic discount factor (see Sect. 4.4).
Show that
EŒlog.Qrn / EŒlog.m/;
Q for all n D 1; : : : ; N:
Exercise 5.9 Consider an economy with a risk free lending rate rl lower than the
corresponding risk free borrowing rate rb (i.e., rb > rl ), reflecting the presence of
transaction costs in the risk free market. Suppose that all the agents choose to hold
mean-variance efficient portfolios. By relying on the same arguments adopted in
Sect. 5.2 and assuming that the net supply of the risk free asset is zero, show that
the following Zero-ˇ CAPM relation is obtained:

with EŒQrm EŒQrzc.m/ > 0 and rb EŒQrzc.m/ rl .

Exercise 5.10 Consider an economy where all the agents choose to hold mean-
variance efficient portfolios but it is not possible to borrow at the risk free rate rf (i.e.,
only investing in the risk free asset is allowed). By relying on the same arguments
adopted in Sect. 5.2 and assuming that the risk free asset is in strictly positive net
supply, show that the following Zero-ˇ CAPM relation is obtained:

with EŒQrm EŒQrzc.m/ > 0 and EŒQrzc.m/ rf .

Exercise 5.11 Consider the setting of the Modigliani-Miller economy described at

the end of Sect. 4.4, with an unleveraged firm generating revenues V1 at date t D 1
and, as in (4.62), denote by run the expected return on the shares of the unleveraged
firm. Consider then a leveraged firm generating the same revenues at date t D 1
but partly financed by debt with nominal value K. As in (4.62), we denote by rlv
and rd the expected rates of return on the stock and on the debt, respectively, of the
leveraged firm. Suppose that the CAPM holds for the stock of the unleveraged firm,
the stock of the leveraged firm and the debt (bond) of the leveraged firm, with respect
to some market portfolio with expected rate of return rm WD EŒQrm and a risk free
asset with return rf . Show that the beta coefficient of the stock of the leveraged firm
(denoted by ˇlv ) can be expressed as a linear combination of the beta coefficients
associated to the stock of the unleveraged firm and to the debt (denoted respectively
by ˇun and ˇd ).
Exercise 5.12 Consider three risky assets with expected returns, standard devia-
tions and correlations with the market portfolio given by the following vectors:
2 3 2 3 2 3
1:07 0:3 0:2
D 41:085 D 4 0:2 5 4
D 0:45 :
1:1 0:15 0:8
Suppose that the market portfolio is characterized by expected return m D 1:09

and standard deviation m D 0:1.
(i) Does the CAPM relation hold if the risk free rate is equal to rf D 1:04?
(ii) Compute the risk premium for an asset with ˇ D 0:5.
(iii) Show that in equilibrium there cannot exist an asset with expected return 0 D
1:2 and ˇ 0 D 0:5.
Exercise 5.13 Consider an economy with three possible states of the world with
equal probabilities of occurrence. The economy’s aggregate endowment is given
by e D .2; 5; 3/ and there are three traded assets. The first asset is risk free, has
price p1 D 1 at time t D 0 and delivers the constant payoff 1 at time t D 1. The
two remaining assets are risky: the first asset has random return rQ2 D .0; 3; 1/ and
unitary price at time t D 0 and the second asset has random payoff dQ 3 D .0; 0; 2/ and
price p3 at time t D 0. The aggregate supply of the three assets is given by .2; 1; 0/.
Determine the equilibrium price of the third asset assuming that the CAPM relation
holds.
Exercise 5.14 Consider an economy with two traded assets with returns rQ1 and rQ2
together with a market portfolio with return rQ m . The covariance matrix of the random
vector .Qr1 ; rQ2 ; rQ m / is given by
0 1
0:16 0:02 0:064
@ 0:02 0:09 0:032A :
0:064 0:032 0:040
5.7 Exercises 253
Moreover, it holds that EŒQrm D 1:12 and rf D 1:04. Consider the portfolio w
investing 3=4 and 1=4 in the first and in the second assets, respectively.
(i) Compute the ˇ coefficients of the first and of the second asset as well as of the
portfolio w with respect to the market portfolio.
(ii) Write the equation defining the Security Market Line.
(iii) Compute the risk premium of the two assets as well as of the portfolio w.
Exercise 5.15 Consider an economy with three traded assets, whose returns are
generated by the following linear model with respect to two risk factors fQ1 and fQ2
(with EŒ fQ1 D EŒ fQ2 D 0):
8
ˆ
ˆ rQ1 D 0:1 C 0:3 fQ1 0:2 fQ2 ;
<
rQ D 0:5 0:4 fQ1 C 0:3 fQ2 ;
ˆ 2
:̂
rQ3 D 0:2 0:2 fQ1 C 0:4 fQ2 :
(i) Determine the portfolios of the three traded assets which have unitary exposure
to one risk factor and zero exposure to the other risk factor.
(ii) Determine the risk free rate implicit in the economy.
(iii) Consider an additional asset with return rQ4 D ˛ C 0:1 fQ1 C 0:3 fQ2 . Determine the
constant ˛ such that there is no arbitrage opportunity in the economy extended
with this fourth asset.
(iv) Verify that the APT is satisfied in exact form.
(v) Verify that an asset with return rQ5 D 0:2 0:1 fQ1 C 0:1 fQ2 would generate an
arbitrage opportunity.
(vi) Assuming that the first three assets are available for trading at unitary price,
determine the value of a consumption plan equal to 0:4 0:2 fQ1 C 0:4 fQ2 .
Chapter 6
Multi-Period Models: Portfolio Choice,
Equilibrium and No-Arbitrage
It is perfectly true, as philosophers say, that life must be

understood backwards. But they forget the other proposition,
that it must be lived forwards. . . And if one thinks over the
proposition it becomes more and more evident that life can
never really be understood in time simply because at no
particular moment can I find the necessary resting-place to
understand it backwards.
Kierkegaard (Journal of the year 1843)
The analysis developed in the previous chapters refers to a stylized two-period

economy, where the uncertainty is related to the possible state of the world realized
at t D 1 and agents have to take decisions at the initial date t D 0. In this chapter,
we consider a more general economy in a multi-period setting, where agents can
make dynamic decisions and revise their strategies according to the evolution of the
uncertainty on the state of the world. In such a dynamic context, a first and basic
issue to consider is how to model the evolution of time. To this regard, one can
distinguish between discrete time and continuous time models. In this book, despite
the powerful mathematical tools that can be employed for the analysis of continuous
time models, we shall limit ourselves to a discrete time setting, since this allows us
to capture all the main ideas avoiding the mathematical technicalities of continuous
time models. More specifically, we shall consider discrete time models on a finite
probability space (i.e., the number of elementary states of the world is finite),
thus allowing for an easy description of the resolution of uncertainty. The goal of
the present chapter is to extend to a multi-period setting the fundamental results
on portfolio optimization, equilibrium and no-arbitrage discussed in the previous
chapters for a static economy. Chapter 7 contains an extensive presentation of the
main empirical findings and puzzles arising in the context of multi period classical
asset pricing theory.
We consider a pure exchange economy comprising T C 1 dates t 2 f0; 1; : : : ; Tg.
For simplicity, we assume that there exists a single good, which can be generically
interpreted as wealth or as a reference consumption good and is available for trade
at every date t 2 f0; 1; : : : ; Tg (at the expense of a heavier notation, the results
presented below can be easily generalized to an economy with multiple goods). To
represent the uncertainty, we introduce a probability space .˝; F ; P/, where, as

256 6 Multi-Period Models: Portfolio Choice, Equilibrium and No-Arbitrage
usual, ˝ D f!1 ; : : : ; !j˝j g represents the finite set of all elementary states of the
world, F is the partition generated by all the elements ! 2 ˝ and P is a probability
measure on .˝; F / representing agents’ beliefs. Let us recall that a partition of ˝
is a collection of non-empty pairwise disjoint subsets of ˝ (called events) such that
their union coincides with ˝.
In the previous chapters, we have considered two-period models (i.e., t 2 f0; 1g),
the uncertainty being due to the fact that agents at the initial date t D 0 do not know
which state of the world will be realized at the future date t D 1. At the initial date
t D 0, the agents only know the possible set of states of the world and their beliefs
concerning the possible realizations are represented by a probability measure. In a
multi-period setting, we need to make information dynamic, in the sense that agents
partially learn over time the true state of the world, which will be fully revealed
at the terminal date T. In other words, we need to mathematically represent the
evolution of information over time. To this effect, we introduce a refining sequence
F D .Ft /tD0;1;:::;T of partitions of ˝ satisfying the crucial property that, for every
s; t 2 f0; 1; : : : ; Tg with s < t, the partition Ft is finer than Fs , meaning that
every event in Fs can be represented as the union of some events belonging to
Ft . Moreover, we assume that F0 D ˝, meaning that agents do not have any
information about the true state of the world at the initial date t D 0, and that
FT D F , so that the true state of the world is fully revealed at the terminal date
t D T. The sequence of partitions F represents the information flow of our economy.
Indeed, while agents will discover the true state of the world ! 2 ˝ only at the
terminal date T, they will accumulate information over time and learn that ! belongs
to increasingly smaller subsets of the universe ˝.
For any t 2 f0; 1; : : : ; Tg, we denote by At an event of the partition Ft and we
say that event At is realized if ! 2 At , i.e., if the true state of the world is an element
of At . At every date t, the agents know to which event At 2 Ft does ! belong but
cannot distinguish among different states of the world belonging to the same event
At . Given a real-valued random variable W ˝ ! R and t 2 f0; 1; : : : ; Tg, we shall
say that is measurable with respect to Ft if the realization .!/ is fully known
at date t. More specifically, for every date t 2 f0; 1; : : : ; Tg, let us introduce the
notation t WD jFt j, so that Ft D fA1t ; : : : ; At t g. We say that a random
Pt variable
W ˝ ! R is Ft -measurable if it can be represented as .!/ D sD1 1Ast .!/ s ,
with s 2 R, for all s D 1; : : : ; t . The present notions of information flow and
measurability can be generalized by assuming that F is a filtration, namely a refining
sequence of -algebras on ˝.
Due to its refining structure, F is sometimes referred to as an event tree, see
Fig. 6.1. An event tree can be represented by a series of knots ordered from the
left (starting at the initial date t D 0) to the right (up to the final date t D T) and
connected through the branches of a tree. Every knot of the tree represents an event
and, for every t 2 f0; 1; : : : ; Tg, the family of all knots of the tree in correspondence
of t represents the partition Ft . Conditionally on the realization of some event As at
date s, an event/knot At (for s < t) can contain the true state of the world only if At
can be reached through the branches of the tree departing from the knot As (i.e., if
6 Multi-Period Models: Portfolio Choice, Equilibrium and No-Arbitrage 257
0 1 2 T
Fig. 6.1 Information flow (event tree)
At As ). This means that, for every s; t 2 f0; 1; : : : ; Tg with s < t and for every
! 2 ˝, if ! 2 As then ! … At for all the events At which are not reachable from As .
Since we restrict our attention to information flows represented by refining
sequences of partitions (event trees), the knowledge of the event At realized at date
t 2 f1; : : : ; Tg embeds the whole information of the previous events fAs gsD0;:::;t1
realized before date t. In particular, this implies that the knowledge of the event
realized at the current date acts as a sufficient statistic for predicting the realization
of future events and past information is not relevant.
As in the previous chapters, we assume that agents exhibit homogeneous beliefs,
meaning that all agents agree on a common probability measure P. However, we
want to point out that most of the following results can be generalized to the case of
heterogeneous beliefs, provided that all agents agree on the set of possible states
of the world which can occur with strictly positive probability (i.e., the beliefs
of all the agents are equivalent). Moreover, the information flow F is the same
for all the agents, meaning that we do not consider agents having access to some
form of privileged information (this situation will be considered in Chap. 8). Every
elementary event ! 2 ˝ has a strictly positive probability ! > 0 of occurrence.
For every t 2 f0; 1; : : : ; Tg, the probability of an event At 2 Ft is given by

X
At WD P.At / D ! :
!2At
For all s; t 2 f0; 1; : : : ; Tg with s < t and At 2 Ft , As 2 Fs , it follows from the

Bayes rule that
( P.A / At
P.As \ At / t
D if At As I
At jAs WD P.At jAs / D D P.As / As
P.As / 0 otherwise.
The quantity At jAs represents the probability of occurrence of the event At at time
t, given the information represented by the realization of event As at the previous
date s.
The information flow F captures the dynamic feature of information in a multi-
period setting. Since economic agents obviously react to the information received
over time, it is natural to allow them to adjust their investment/consumption
strategies over time. This implies that the notion of random variable is no longer
sufficient to represent the randomness of prices, dividends and consumption plans,
since we need to keep track of the temporal resolution of uncertainty. This leads
to the notion of stochastic process, namely a collection X D .Xt /tD0;1;:::;T of
random variables indexed with respect to time. In particular, we consider stochastic
processes which are adapted to the information flow F, in the sense that, for every
date t 2 f0; 1; : : : ; Tg, the random variable Xt is Ft -measurable. We shall always
represent prices, dividends and consumption plans as adapted stochastic processes.
For an adapted stochastic process X, we denote by Xt .At / the value of the process X
at date t in correspondence of the event At , for any t 2 f0; 1; : : : ; Tg and At 2 Ft .
This chapter is structured as follows. In Sect. 6.1, we analyse by means of
dynamic programming techniques the optimal investment-consumption problem for
an agent and we present explicit solutions in the case of classical utility functions.
In Sect. 6.2, we consider an economy populated by I agents and study the relation
between equilibrium and Pareto optimality. Similarly as in Chap. 4, the notion of
complete markets will play a crucial role. Section 6.3 extends the fundamental
theorem of asset pricing to a multi-period setting, while Sect. 6.4 generalizes the
asset pricing relations presented in Chap. 5. In Sect. 6.5 we present some basic
concepts related to the existence of speculative bubbles. At the end of the chapter,
we provide a guide to further readings as well as a series of exercises.
6.1 Optimal Investment and Consumption Problems 259
6.1 Optimal Investment and Consumption Problems
In this section, we present a general approach based on the dynamic programming

principle to the solution of the optimal investment-consumption problem of a single
agent.
Adopting the setting of Huang & Litzenberger [971, Chapter 7], we assume
that the economy comprises N securities paying non-negative (but possibly null)
random dividends at the dates t D 0; 1; : : : ; T and available for trade at every date
t D 0; 1; : : : ; T. In the terminology of Huang & Litzenberger [971], such assets are
called long-lived securities. Typical examples of long-lived securities are stocks or
corporate bonds maturing at the terminal date T. For each n D 1; : : : ; N, the n-th
security is characterized by its (ex-dividend) price process sn D .snt /tD0;1;:::;T and by
its dividend process dn D .dtn /tD0;1;:::;T and both processes are assumed to be adapted
to the information flow F. As explained in the introduction, this means that, at every
date t D 0; 1; : : : ; T, the prices and the dividends of the N securities are measurable
with respect to the information contained in Ft . This simply means that, at each date
t D 0; 1; : : : ; T, the prices and the dividends of all N securities are observable. In
order to avoid trivial situations, we assume that all securities are in strictly positive
supply in the economy and that the dividend process dn is non-negative and non-null,
for all n D 1; : : : ; N, in the sense that there exists at least one date t 2 f0; 1; : : : ; Tg
such that dtn .At / > 0 in correspondence of some event At 2 Ft . The price process
sn represents the random evolution of the ex-dividend price of the n-th security (i.e.,
the random variable snt is the price at date t of the n-th security after the delivery
of the dividend dtn ) and is also assumed to be non-negative and non-null, for all
n D 1; : : : ; N. Moreover, since the economy is supposed to end at the terminal
date T, it holds that snT D 0, for all n D 1; : : : ; N. Let us also define the cum-
dividend price process pn D . pnt /tD0;1;:::;T by pnt WD snt C dtn , for all n D 1; : : : ; N
and t D 0; 1; : : : ; T. We denote by S D .St /tD0;1;:::;T and D D .Dt /tD0;1;:::;T the
RN -valued price and dividend processes of the N securities, respectively, so that
St D .s1t ; : : : ; sNt /> and Dt D .dt1 ; : : : ; dtN /> , for all t 2 f0; 1; : : : ; Tg.
Having introduced the general securities available for trade, let us describe
how agents can trade, construct portfolios and choose consumption plans. In
particular, since the information and the uncertainty evolve dynamically over time,
as represented by the information flow F, it is natural to allow agents to rebalance
their portfolios over time depending on the resolution of uncertainty and on the
available information. For the sake of simplicity, we assume that agents do not
receive exogenous income such as labor income. This implies that the agents’ wealth
is fully determined by their initial endowments, by their consumption plans, by their
trading strategies and by the dividend and price processes of the traded securities. As
a preliminary, we need to introduce the notion of predictable stochastic process: a
stochastic process X D .Xt /tD0;1;:::;T is said to be predictable if, for all t D 1; : : : ; T,
the random variable Xt is Ft1 -measurable (in particular, note that every predictable
stochastic process is adapted). In other words, a stochastic process is predictable if
its realization at date t is known one period in advance (at date t 1). We are in a
position to formulate the following definition.
Definition 6.1 A couple .; c/ of stochastic processes is said to be a trading-
consumption strategy if D .t /tD0;1;:::;T is an RN -valued predictable stochastic
process and c D .ct /tD0;1;:::;T is an RC -valued adapted stochastic process.
A trading-consumption strategy .; c/ is said to be self-financing if
>
tC1 St D t> .St C Dt / ct ; for all t D 0; 1; : : : ; T 1; (6.1)
T> DT D cT : (6.2)
The processes D .t /tD0;1;:::;T and c D .ct /tD0;1;:::;T appearing in the above
definition have the following interpretation. For all t D 1; : : : ; T, the random
vector t D .t1 ; : : : ; tN /> represents the number of units of the N securities
held in the portfolio on the time period Œt 1; t before trading takes place at
date t. In particular, the predictability requirement on the process ensures that
the portfolio constructed at date t 1 and held until date t only depends on the
information available at the portfolio formation date t 1. Of course, this represents
a minimal informational constraint for a meaningful notion of dynamic trading
strategy (excluding clairvoyance of future dividends and asset prices). The vector
0 D .01 ; : : : ; 0N /> represents the initial holdings in the N securities. Letting
x0 WD 0> .S0 CD0 / denote the initial wealth, the self-financing condition can also be
formulated as in Definition 6.1 starting from wealth x0 at date t D 0, i.e., requiring
that 1> S0 D x0 c. The consumption process c D .ct /tD0;1;:::;T represents (in units
of some reference consumption good or wealth) the quantity consumed by the agent
at t D 0; 1; : : : ; T. In particular, the process c is adapted, meaning that the quantity
consumed at date t is allowed to be contingent on the information released at date t.
Note also that condition (6.2) relies on the assumption that the economy ends at the
terminal date T and, hence, ST D 0 (so that no trading takes places at the final date
T) and the dividends received at T are fully consumed.
To a trading-consumption strategy .; c/ we can associate a wealth process
W./ D .Wt .//tD0;1;:::;T defined by Wt ./ WD t> .St C Dt /, for all t D 0; 1; : : : ; T.
This process represents the value of the portfolio associated to a strategy . In
particular, due to equation (6.1), if .; c/ is a self-financing trading-consumption
strategy, then the wealth process W./ satisfies
Wt ./ D t> .St C Dt / D tC1

>
S t C ct ; for all t D 0; 1; : : : ; T 1: (6.3)
The reason for the self-financing terminology is made clear by condition (6.3),
which requires that the value at date t of the portfolio constructed at the previous
date t 1 (on the left-hand side) is equal to the sum of the cost of the new portfolio
created at date t and the wealth consumed at date t (on the right-hand side). In
particular, there is no injection of wealth at any date t D 1; : : : ; T. Condition (6.3)
can be equivalently rewritten as follows, for all t D 0; 1; : : : ; T 1:
>
WtC1 ./ WD WtC1 ./ Wt ./ D tC1 .StC1 C DtC1 St / ct
(6.4)
> >
D tC1 StC1 C tC1 DtC1 ct :
As shown in Exercise 6.1, condition (6.4) easily leads to the following representa-
tion of the wealth process W./ associated to a self-financing trading-consumption
strategy .; c/:
X
t X
t X
t1
Wt ./ D W0 ./ C s> Ss C s> Ds cs ; for all t D 1; : : : ; T: (6.5)
sD1 sD1 sD0
Note that the self-financing property of a trading-consumption strategy does not

depend on the quantity with respect to which prices and consumption are measured
(in other words, the self-financing property is numéraire invariant). For instance, if
a risk free security is traded in the market, then discounting all the quantities with
respect to it will not affect the self-financing property (see Exercise 6.2).
If a trading-consumption strategy .; c/ is self-financing, then we say that
the consumption plan c is financed by the trading strategy , starting from the
initial wealth x D W0 ./. We denote by A the set of all self-financing trading-
consumption strategies.
Having described the available securities and introduced the notion of trading-
consumption strategy, we are now in a position to formulate the optimal investment-
consumption problem of an agent. We assume that the agent is endowed at date
t D 0 with an initial wealth x0 > 0 (we can also assume that the initial endowment
is expressed as a vector 0 2 RN , with 0n denoting the number of shares of the n-th
security, for n D 1; : : : ; N, so that the initial wealth is x0 D 0> .S0 C D0 /). We
introduce the set
˚
A .x0 / WD .; c/ 2 A W W0 ./ D x0
representing all self-financing trading-consumption strategies starting from initial

wealth x0 . Concerning the agent’s preferences, we assume that they can be repre-
sented by an expected time additive and state independent utility function defined
over consumption plans. More specifically, for each t D 0; 1; : : : ; T, we assume
that the utility associated with consumption at date t is evaluated at t D 0 by the
discounted expected utility ı t EŒu.ct /, where u W RC ! R is a strictly increasing,
strictly concave and twice differentiable function and ı is a constant discount factor
with 0 < ı 1. Hence, given the initial wealth x0 > 0, the utility function u and
the discount factor ı, an agent’s optimal investment-consumption problem can be
formulated as follows:
!
X
T
max u.c0 / C ı t EŒu.ct / : (6.6)
.;c/2A .x0 /
tD1
In problem (6.6) the maximization is with respect to all self-financing trading-

consumption strategies .; c/ starting from the initial wealth x0 . Whenever it
exists (and this will be related to the absence of arbitrage opportunities, see
Proposition 6.31), we denote by . ; c / the optimal solution to problem (6.6).
Apart from some special cases (see below), the optimal investment-consumption
problem (6.6) cannot be solved as a sequence of static single-period optimization
problems. Therefore, one cannot apply step-by-step for each date t D 0; 1; : : : ; T 1
the static optimization techniques introduced in the previous chapters (even though,
as we shall see below, the techniques introduced in the previous chapters for
single-period problems will play an important role in the solution of the multi-
period problem (6.6)). Nonetheless, if problem (6.6) admits an optimal trading-
consumption strategy . ; c /, then the same strategy should remain optimal at
every future date t D 1; : : : ; T. In other words, if an optimal solution . ; c /
to problem (6.6) has been found at the initial date t D 0, then . ; c / is
such that, for every future date t D 1; : : : ; T, there does not exist an alternative
trading-consumption strategy .; Q cQ / starting at date t from wealth Wt . / which
improves the expected utility from date t onwards. This will be shown rigorously in
Proposition 6.2 below.
As a preliminary, we need to introduce some notation. For t 2 f0; 1; : : : ; T 1g,
we define CtC .x/ as the set of all non-negative adapted processes .cs /sDt;:::;T such
that there exists an RN -valued predictable process .Qs /sDtC1;:::;T (starting at date t)
satisfying
QtC1
>
S t D x ct ;
QsC1
>
Ss D Qs> .Ss C Ds / cs ; for all s D t C 1; : : : ; T 1; (6.7)
QT> DT D cT :
In other words, CtC .x/ represents the set of all consumption processes financed
(from date t onwards) by some trading strategy starting at date t from wealth x.1 We
then have the following result (see also Huang & Litzenberger [971, Section 7.8]).
Proposition 6.2 Let . ; c / 2 A .x0 / be an optimal solution to problem (6.6).
Then, for every date t 2 f0; 1; : : : ; T 1g, there does not exist a consumption process
1
The definition of the set CtC .x/ can be straightforwardly extended to the case where wealth x is a
strictly positive Ft -measurable random variable, as considered in Proposition 6.2.
.Qcs /sDt;:::;T 2 CtC .Wt . // such that, on some event AQ t 2 Ft , it holds that
X
T X XT X
ıs As jAQ t u cQ s .As / > ıs As jAQ t u cs .As / ; (6.8)
sDt As 2Fs sDt As 2Fs
Qt
As A Qt
As A
where cQ s .As / and cs .As / denote the values of the processes cQ and c , respectively, at
date s in correspondence of the event As , for As 2 Fs and s 2 ft; : : : ; Tg.
Proof The proposition can be proved by contradiction. Suppose that there exists
a process .Qcs /sDt;:::;T 2 CtC .Wt . // such that (6.8) holds in correspondence of
some event AQ t 2 Ft . Since .Qcs /sDt;:::;T 2 CtC .Wt . //, there exists an RN -valued
predictable process .Qs /sDtC1;:::;T such that (6.7) holds (with c D cQ ). Define then a
new trading-consumption strategy . 0 ; c0 / as follows, for all events As 2 Fs and for
all dates s D 0; 1; : : : ; T:
(
s .As / for s D 0; : : : ; t;
s0 .As / WD
s .As / C .Qs .As / s .As //1As AQ t for s D t C 1; : : : ; TI
(
cs .As / for s D 0; : : : ; t 1;
c0s .As / WD
cs .As / C .Qcs .As / cs .As //1AsAQ t for s D t; : : : ; T;
where s .As /, Qs .As /, cs .As / and cQ s .As / denote the values of the processes , ,
Q c
and cQ , respectively, at date s in correspondence of As 2 Fs and where 1As AQ t takes
the value one if As is a subset of AQ t and zero otherwise. Since . ; c / 2 A .x0 / and
Q cQ / satisfies (6.7), it can be easily checked that . 0 ; c0 / 2 A .x0 /. Moreover, the
.;
definition of c0 and inequality (6.8) imply that
X
T X XT X
ıs As jAt u c0s .As / ıs As jAt u cs .As / ;
sDt As 2Fs sDt As 2Fs
As At As At
for all events At 2 Ft , with strict inequality in correspondence of AQ t 2 Ft

(which is realized with strictly positive probability, since every elementary event
! 2 ˝ has a strictly positive probability of occurrence). Denoting by EŒjFt
the conditional expectation with respect to Ft and using the tower property of
conditional expectation (so that EŒ D EŒEŒjFt ), it holds that
" # " #
X
T X
T X
T X
T
ı s
EŒu.c0s / DE ı s
EŒu.c0s /jFt >E ı s
EŒu.c
s /jFt D ı s EŒu.c
s /;
sDt sDt sDt sDt
which implies that
X
T X
t1 X
T
ı s EŒu.cs / D ı s EŒu.cs / C ı s EŒu.cs /
sD0 sD0 sDt
X
t1 X
T X
T
< ı s EŒu.c0s / C ı s EŒu.c0s / D ı s EŒu.c0s /:
sD0 sDt sD0
This contradicts the optimality of . ; c /, thus proving the claim. t

u
According to the above proposition, once an optimal strategy . ; c / has been
chosen at the initial date t D 0, it will never be optimal to revise that strategy
at a later date t D 1; : : : ; T. The optimal strategy . ; c / must remain optimal
starting from every knot of the event-tree. In other words, the current optimal
decision of an agent already takes into account the future optimal decisions in
response to the resolution of the uncertainty. This result is crucial for the solution of
the dynamic investment-consumption problem and naturally leads to the dynamic
programming principle, which provides a backward recursive technique for the
solution of stochastic control problems of the type (6.6).
For the investment-consumption problem (6.6), we introduce the value function
(or, more precisely, value process) V as
!
X
T
V.x; t/ WD sup u.ct / C ı st
EŒu.cs /jFt ; (6.9)
c2CtC .x/ sDtC1
for t D 0; 1; : : : ; T 1, with V.x; T/ WD u.x/, for all x 2 RC . In (6.9) the

supremum is taken over all consumption processes .cs /sDt;:::;T that can be financed
by some trading strategy .Qs /sDtC1;:::;T starting at date t from wealth x, see (6.7). The
quantity V.x; t/ represents the maximal expected utility that can be reached starting
at date t with an endowment of wealth x, given the information Ft available at t.
Definition (6.9) can be extended to the case where x is an RC -valued Ft -measurable
random variable. It is important to remark that, in general, the quantity V.x; t/ is
an Ft -measurable random variable and cannot be considered as a deterministic
function of x and t (see however the special cases considered below in this section).
The next proposition states the dynamic programming principle for the investment-
consumption problem (6.6) in terms of the value function V.
Proposition 6.3 For every .x; t/ 2 RC f0; 1; : : : ; T 1g, the value function V
defined in (6.9) satisfies the dynamic programming principle:
>
V.x; t/ D sup u.ct / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt ; (6.10)
where the supremum is taken with respect to all Ft -measurable random variables
>
.tC1 ; ct / 2 RN RC satisfying tC1 St C ct D x. Moreover, if . ; c / 2 A .x0 / is
an optimal solution to problem (6.6), then it holds that

V Wt . /; t D u.ct /CıE V WtC1 . /; tC1 jFt ; for all t D 0; 1; : : : ; T1:
(6.11)
Proof Let .x; t/ 2 RC f0; 1; : : : ; T 1g and .cs /sDt;:::;T 2 CtC .x/ (as explained
above, this means that .cs /sDt;:::;T is a consumption stream financed by some trading
strategy .s /sDtC1;:::;T starting at date t from wealth x). For t D T 1, equation (6.10)
coincides with the definition (6.9) of the value function V. So, let us suppose that
t 2 f0; 1; : : : ; T 2g. Then, it holds that
X
T
u.ct / C ı st EŒu.cs /jFt
sDtC1

X
T ˇ
ˇ
D u.ct / C ıE u.ctC1 / C ı s.tC1/ u.cs /ˇFt
sDtC2

h X iˇ
T
ˇ ˇ
D u.ct / C ıE u.ctC1 / C E ı s.tC1/ u.cs /ˇFtC1 ˇFt
sDtC2

h X
T
ˇ iˇˇ
u.ct / C ıE sup u.ctC1 / C E ı s.tC1/
u.cs / FtC1 ˇˇFt
ˇ
C >
c2CtC1 .tC1 .StC1 CDtC1 // sDtC2
>
D u.ct / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt ;
where the last equality follows from definition (6.9). By taking the supremum over
>
all Ft -measurable couples .tC1 ; ct / 2 RN RC satisfying tC1 St C ct D x in the
right-hand side of the above inequality and then the supremum over all c 2 CtC .x/
in the left-hand side, we obtain
>
V.x; t/ sup u.ct / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt :
.tC1 ;ct /2RN RC
>
tC1 St Cct Dx
(6.12)
To prove the converse inequality, let .tC1 ; ct / 2 RN RC be a couple of Ft -

>
measurable random variables such that tC1 St C ct D x. Consider then the value
>
function defined in (6.9) at t C 1 in correspondence of wealth tC1 .StC1 C DtC1 /.
Observe that, by the definition of supremum, for every " > 0 there exists a
C
consumption stream .c"s /sDtC1;:::;T 2 CtC1 >
.tC1 .StC1 C DtC1 // such that
h X
T
ˇ i >
u.c"tC1 / C E ı s.tC1/ u.c"s /ˇFtC1 V tC1 .StC1 C DtC1 / ":
sDtC2
This implies that

>
u.ct / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt

h X
T
ˇ iˇˇ
"
u.ct / C ıE u.ctC1 / C E ı s.tC1/
u.cs / FtC1 ˇˇFt C "
" ˇ
sDtC2
X
T
D u.ct / C ı st EŒu.c"s /jFt C "
sDtC1
V.x; t/ C ";
where the last inequality follows from the fact that .ct ; c"tC1 ; : : : ; c"T / 2 CtC .x/. By
the arbitrariness of ", we get that
>
u.ct / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt V.x; t/:
By taking the supremum in the left-hand side of the last inequality over all elements
>
.tC1 ; ct / 2 RN RC such that tC1 St C ct D x, we get the converse inequality
to (6.12), thus proving that (6.10) holds as an equality. Finally, in correspondence of
an optimal solution . ; c / 2 A .x0 /, relation (6.11) follows from Proposition 6.2
together with the dynamic programming principle (6.10). t
u
The above proposition represents a fundamental result for the solution of
dynamic optimal investment-consumption problems of the type (6.6). Indeed, for
any t 2 f0; 1; : : : ; T 1g, the optimal solution to problem (6.6) can be found by
first searching for an optimal trading-consumption strategy from time t C 1 onwards
and then computing the maximum appearing in the right-hand side of (6.10) over the
single-period interval Œt; tC1. In particular, condition (6.11) shows that, if a trading-
consumption strategy . ; c / 2 A .x0 / is an optimal solution to problem (6.6),

then, for every t D 0; 1; : : : ; T 1, the couple .tC1 ; ct / must be optimal over the
time interval Œt; t C 1 for the maximization problem (6.10). Equation (6.11) is also
known as the Bellman equation.
The dynamic programming principle provides a backward recursive algorithm
for the solution of problem (6.6). At each time step t 2 f0; 1; : : : ; T 1g, the global
maximization problem is reduced to the standard optimization problem (6.10) over
RN RC conditionally on the information Ft available at date t. By means of this
approach, the multi-period optimization problem (6.6) is reduced to a sequence
of simpler single-period optimization problems, as will be illustrated below. The
general strategy in the application of the backward recursive algorithm consists
in starting by considering conditions (6.10)–(6.11) at date T 1, with one period
remaining until the terminal date T. On this time interval, condition (6.10) reduces to
a standard single-period optimization problem, to which the techniques introduced
in the previous chapters can be applied. Having computed the optimal consumption
and the value function at date T 1 conditionally on FT1 , one can then solve (6.10)
and compute the optimal consumption for the time interval ŒT 2; T 1 and then
continue backwards in time until the initial date t D 0. As a result of this procedure,
the optimality of the future decisions is already taken into account at each single
step of the optimization problem. This procedure is illustrated in Exercises 6.4–6.7.
The dynamic programming principle leads to the following properties of the
value function in correspondence of an optimal solution . ; c / to problem (6.6).
We denote by .Wt /tD0;1;:::;T the wealth process associated to the optimal trading-
consumption strategy . ; c /, i.e., Wt WD Wt . /, where Wt . / is defined as
in (6.3). From now on, we always assume that the prices of the N traded securities
are strictly positive,2 i.e., snt > 0, for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, and
we denote by rtn WD pnt =snt1 the random return of asset n in the time interval Œt 1; t.
Proposition 6.4 Let . ; c / 2 A .x0 / be an optimal solution to problem (6.6). If
V.; t/ is differentiable with respect to its first argument, for each t D 1; : : : ; T, then
the following holds, for all t D 0; 1; : : : ; T 1:
u0 .ct / D ıEŒV 0 .WtC1

; t C 1/rtC1
n
jFt ; for all n D 1; : : : ; N; (6.13)
and
u0 .ct / D V 0 .Wt ; t/; (6.14)
where V 0 .; t/ denotes the first derivative of V.; t/ with respect to its first argument.
In particular, (6.13)–(6.14) together imply that, for all t D 0; 1; : : : ; T 1,
u0 .ct / D ıEŒu0 .ctC1 /rtC1

n
jFt ; for all n D 1; : : : ; N: (6.15)
Proof Due to Proposition 6.3, condition (6.11) holds in correspondence of an

optimal solution . ; c / to problem (6.6). Hence, using equation (6.3), we get
>
V.Wt ; t/ D u.Wt tC1
>
St / C ıE V tC1 .StC1 C DtC1 /; t C 1 jFt ;
for all t D 0; 1; : : : ; T 1. Hence, if V.; t/ is differentiable with respect to its

first argument, for all t D 1; : : : ; T, differentiation with respect to tC1 leads to the
optimality condition
>
0 D u0 .Wt tC1
>
St /snt C ıE V 0 tC1 .StC1 C DtC1 /; t C 1 sntC1 C dtC1
n
jFt ;
(6.16)
2
Note that the absence of arbitrage opportunities (see Sect. 6.3) implies that, if dtn .At / > 0 for
some At 2 Ft and t 2 f1; : : : ; Tg, then sns .As / > 0 for all As At and s D 0; 1; : : : ; t 1. In
other words, if a security delivers a strictly positive dividend in correspondence of some future
event which can be realized with strictly positive probability, then its current price must be strictly
positive as well.
for each n D 1; : : : ; N. Equation (6.13) then follows by dividing by snt , which

is assumed to be strictly positive. Moreover, using in sequence the optimality
condition (6.11), the self-financing condition (6.4), equation (6.16) and the self-
financing condition (6.3), it holds that
" #
ˇ
0 @c @WtC1 ˇ
V .Wt ; t/ Du 0
.c
t /
t 0
C ıE V .WtC1 ; t C 1/ ˇFt
@Wt @Wt
@c
D u0 .c
t /
t
@Wt
" !ˇ #
@c XN n
@tC1 ˇ
C ıE V 0
.WtC1 ;t C 1/ 1 t
C .stC1 C dtC1 st / ˇˇFt
n n n
@Wt
nD1
@Wt
!
@c X @tC1 N n
0
Du .c
t /
t
C sn
@Wt
nD1
@Wt
t
" !ˇ #
@c XN n
@tC1 ˇ
0 n ˇ
C ıE V .WtC1 ;t C 1/ 1 t
st ˇFt
@Wt
nD1
@Wt
D u0 .c
t /;

where in the third equality we have used the Ft -measurability of tC1 (since is a
predictable process). This proves (6.14). t
u
Proposition 6.4 shows several important properties of a solution to the optimal
investment-consumption problem (6.6). In particular, relation (6.14) shows that,
in correspondence of an optimum, the marginal utility u0 .ct / of consumption at
time t is equal to the marginal (indirect) utility of wealth V 0 .Wt ; t/ at time t. This
condition (which is often referred to as the envelope condition) intuitively means
that, in correspondence of an optimum, one extra unit of consumption at time t or
one extra unit of wealth invested in the optimal way from date t onwards should
give the same marginal increase of expected utility. Since wealth is only used to
finance future consumption, this is equivalent to saying that, in correspondence
of an optimal solution, the marginal utility of current consumption should equal
the marginal utility of future consumption. Condition (6.15) is often referred to as
the Euler condition for the investment-consumption problem (6.6). Note also that
equation (6.15) can be equivalently rewritten as
" #
u0 .c ˇ
tC1 / ˇ
snt D ıE .dtC1
n
C sntC1 /ˇFt ; for all n D 1; : : : ; N and t D 0; 1; : : : ; T1:
u0 .c
t /
This fundamental relation will be used in Sect. 6.4 to derive several asset pricing
relations.
If u is strictly increasing and strictly concave, then it can be shown that V.; t/ is
increasing and strictly concave in its first argument (representing wealth), for every
t D 0; 1; : : : ; T (see Exercise 6.3). As a consequence, in view of (6.14), there exists
for every t D 0; 1; : : : ; T a random function gt such that ct D gt .Wt /, with gt ./
strictly increasing and Ft -measurable (compare also with Huang & Litzenberger
[971, Section 7.10]). However, the function gt ./ is in general a random function
depending on Ft and, in general, can be explicitly computed only in some special
cases, as considered below in this section.
In most of the cases, it is assumed that the available securities include a
risk free asset delivering a constant rate of return. In that case, the previous
proposition immediately implies the following result, which will be useful for the
characterization of optimal strategies.
Corollary 6.5 Suppose that there are N C 1 traded securities n D 0; 1; : : : ; N, with
security 0 being a risk free asset with constant rate of return rf > 0. Then, under the
assumptions of Proposition 6.4, the following condition holds in correspondence of
an optimal solution . ; c / to problem (6.6), for all t D 0; 1; : : : ; T 1:
EŒV 0 .WtC1

; t C 1/.rtC1
n
rf /jFt D 0; for all n D 1; : : : ; N: (6.17)
Proof Equation (6.13) implies that, for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1,
ıEŒV 0 .WtC1

; t C 1/rf jFt D u0 .ct / D ıEŒV 0 .WtC1

; t C 1/rtC1
n
jFt ;
form which (6.17) immediately follows. t

u
In view of Proposition 6.4, condition (6.17) can be equivalently rewritten as
EŒu0 .ctC1 /.rtC1

n
rf /jFt D 0; for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1:
Moreover, equation (6.15) implies that
u0 .ct / D ırf EŒu0 .ctC1 /jFt ; for all t D 0; 1; : : : ; T 1:
In particular, if ı D 1=rf , then we get
u0 .ct / D EŒu0 .ctC1 /jFt ; for all t D 0; 1; : : : ; T 1;
meaning that, in correspondence of an optimum, the marginal utility of the optimal

consumption process .u0 .ct //tD0;1;:::;T is a martingale process3 (if ı D 1=rf ).
3
We recall that an adapted process X D .Xt /tD0;1;:::;T is said to be a martingale if, for all s; t 2
f0; 1; : : : ; Tg with s t, it holds that EŒXt jFs D Xs .
Furthermore, if the utility function is quadratic this reduces to
ct D EŒctC1 jFt ; for all t D 0; 1; : : : ; T 1;
so that the optimal consumption process c follows a random walk of the form
ctC1 D ct C "QtC1 ; for all t D 0; 1; : : : ; T 1;
where .Q"t /tD1;:::;T is a sequence of random variables satisfying EŒQ"t jFt1 D 0, for
all t D 1; : : : ; T. This result provides a basis for the so-called life-cycle-permanent
income hypothesis. The random walk hypothesis implies that consumption change
is unrelated to predictable and lagged income changes, but it can depend on
unanticipated income changes. Consumption growth should be more volatile than
income growth if aggregate income exhibits positive serial correlation. There is
a large literature on these implications, showing that consumption growth is both
excessively smooth relative to unanticipated (labor) income growth and excessively
sensitive to lagged-predictable (labor) income growth, see Campbell & Deaton
[345], Campbell & Mankiw [349], Carroll [368], Christiano et al. [444], Deaton
[530], Flavin [714], Hall [878], Hall & Mishkin [879]. Moreover, consumption is
affected by the business cycle. We will return to these topics in the following chapter.
Explicit Solutions to the Optimal Investment-Consumption

Problem
We have so far presented a general theory for the solution of optimal investment-
consumption problems of the form (6.6). In the remaining part of this section, we
derive the optimal investment-consumption plans for two important classes of utility
functions, by relying on the dynamic programming principle, following Ingersoll
[1000, Chapter 11]. In particular, we shall see that the optimal solution to a multi-
period optimal investment-consumption problem does not coincide in general with
the sequence of solutions obtained in the static single-period setting with the same
utility function. This feature reflects the fact that, when solving problem (6.6), an
agent needs to optimally face a trade-off between optimizing single-period returns
and hedging against future changes in the set of investment opportunities. In the
context of a multi-period binomial model (see Sect. 6.3), Exercise 6.9 derives the
explicit solutions to some classical optimal intertemporal consumption problems.
Until the end of this section, we shall assume that the economy comprises
N C 1 securities, with the 0-th security being a risk free asset delivering a constant
rate of return rf > 0, as in Corollary 6.5. Moreover, we restrict our attention
to utility functions defined with respect to strictly positive consumption plans.4
4
Note that, if we consider utility functions u W RC ! R satisfying the Inada condition
limx&0 u0 .x/ D C1, then the corresponding optimal consumption plan .ct /tD0;1;:::;T will
necessarily satisfy c
t > 0 for all t D 0; 1; : : : ; T.
As shown in Lemma 6.19 below, if there are no arbitrage opportunities, then

the wealth process .Wt .//tD0;1;:::;T associated to a trading-consumption strategy
.; c/ 2 A .x0 / satisfying cT > 0 almost surely necessarily satisfies Wt ./ ct > 0
for all t D 0; 1; : : : ; T 1. In this case, we can pass to an alternative parametrization
of the trading strategies in terms of proportions of wealth invested in the N C 1
available securities. To this effect, for any .; c/ 2 A .x0 / satisfying Wt ./ ct > 0
for all t D 0; 1; : : : ; T 1, we define
tC1
n
snt
wntC1 WD ; (6.18)
Wt ./ ct
for all t D 0; 1; : : : ; T 1 and n D 0; : : : ; N. The quantity wntC1 represents the

proportion of wealth (which is not consumed at date t) investedP in the n-th asset at
date t. Note also that, due to equation (6.3), it holds that NnD0 wntC1 D 1, for all
t D 0; 1; : : : ; T 1. Moreover, the self-financing equation (6.3) can be rewritten as
follows (for simplicity of notation, if no ambiguity arises about the trading strategy,
we write Wt WD Wt ./):
X
N
0
WtC1 D tC1
n
.sntC1 C dtC1
n
/ C tC1 rftC1
nD1
X
N X
N
D tC1
n
pntC1 C .Wt ct tC1
n
snt /rf
nD1 nD1
!
X
N
D .Wt ct / wntC1 .rtC1
n
rf / C rf ; (6.19)
nD1
where we have conventionally assumed that s00 D 1. In view of equations (6.3)

and (6.19), it is clear that the wealth process .Wt /tD0;1;:::;T can be equivalently
characterized by the couple .; c/ 2 A .x0 / as well as by the couple .w; c/, where
w D .wt /tD1;:::;T is a predictable stochastic process (defined as in (6.18)) and
representing the proportions of wealth (which is not consumed) invested in the N C1
assets.
In the remaining part of this section, restricting our attention to utility functions
defined with respect to strictly positive consumption streams, we shall use the
parametrization of trading strategies introduced in (6.18). In particular, using
condition (6.13) (applied to the risk free asset) together with Corollary 6.5 and the
Ft1 -measurability of the portfolio wt , for all t D 1; : : : ; T, we get the optimality
condition

u0 .ct / D ıE V 0 .WtC1

; t C 1/rf jFt
X
N
0 0
Dı wn
tC1 E V .WtC1 ; t C 1/.rtC1 rf /jFt CıE V .WtC1 ; t C 1/rf jFt
n
nD1
" X ˇ #
N
ˇ
D ıE V 0 .WtC1

; t C 1/ wn ˇ
tC1 .rtC1 rf / C rf ˇFt :
n
(6.20)
nD1
Moreover, since Wt ct > 0 for all t D 0; 1; : : : ; T 1, condition (6.17) holds if
and only if

.Wt ct /E V 0 .WtC1

; t C 1/.rtC1
n
rf /jFt D 0; for all t D 0; 1; : : : ; T 1;
(6.21)
for all n D 1; : : : ; N. As we shall see in the following, condition (6.21) turns out to
be useful for characterizing the optimal portfolio process .wt /tD1;:::;T .
Logarithmic Utility Function
In the context of multi-period investment-consumption problems, the case of a log-

arithmic utility function represents an interesting and special case, not only because
an optimal solution is available in explicit form, but also because the optimal
solution coincides with a sequence of static single-period optimal investment-
consumption plans. In other words, in a dynamic setting, a logarithmic utility
functions leads to a myopic behavior, in the sense that the portfolio optimization
problem at each time step t 2 f0; 1; : : : ; T 1g is a static optimization problem
unaffected by the future investment opportunities, regardless of the probability
distribution of the asset returns. Referring to Proposition 6.6 below for a rigorous
derivation of this property, the explanation is intuitively clear. Indeed, let us consider
for simplicity the problem of maximizing the expected utility EŒlog.CQ T / of terminal
consumption at date T. Recalling that WT D CT , writing WT D W0 TtD1 Wt =Wt1
and using (6.19) we get that
" !#

YT
Wt X T
Wt
EŒlog.CT / D E log W0 D EŒlog.W0 / C E log
tD1
Wt1 tD1
Wt1
" !#
X T XN
D EŒlog.W0 / C E log wt .rt rf / C rf
n n
;
tD1 nD1
from which we immediately see that the maximization of EŒlog.CT / reduces to the
step-by-step (i.e., for each t D 0; 1; : : : ; T 1) maximization of the logarithmic
rate of return between t 1 and t which only depends on the distribution of the
asset returns at date t and is unaffected by the distribution of future returns. In
this sense, the multi-period problem reduces to a sequence of static single-period
portfolio optimization problems, which can be solved by relying on the techniques
presented in Chap. 3.
The following proposition generalizes the above result to the case of the
optimal consumption-investment problem (6.6) (compare also with Ingersoll [1000,
Chapter 11] and see also Samuelson [1493], Merton [1327]). The proof is given in
Exercise 6.4.
Proposition 6.6 Let u.x/ D log.x/ in problem (6.6) and denote by .Wt /tD0;1;:::;T
the corresponding optimal wealth process. Then, the value function satisfies
V.Wt ; t/ D f .t/ log.Wt / C t; for all t D 0; 1; : : : ; T; (6.22)
1ı TtC1
where the function f W f0; 1; : : : ; Tg ! R is explicitly given by f .t/ D 1ı and
. t /tD0;1;:::;T is an adapted process defined recursively by T D 0 and

f .t/ 1 log f .t/
t D ıf .t C 1/ log
f .t/ ıf .t C 1/

X ˇ !
N
ˇ
CE log wtC1 .rtC1 rf / C rf ˇˇFt
n n
C ıEŒ tC1 jFt ;
nD1
for all t D 0; 1; : : : ; T 1, where .wt /tD1;:::;T denotes the optimal portfolio process.
Moreover, the optimal consumption process .ct /tD0;1;:::;T satisfies
Wt 1ı
ct D D W; for all t D 0; 1; : : : ; T; (6.23)
f .t/ 1 ı TtC1 t
and the optimal portfolio process .wt /tD1;:::;T is such that

2 !1 3
ˇ
X
N
ˇ
E4 wn
tC1 .rtC1 rf / C rf
n
.rtC1
n
rf /ˇˇFt 5 D 0; (6.24)
nD1
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1.

Condition (6.23) shows that, for every date t D 0; 1; : : : ; T, the optimal consumption
ct is a deterministic (time-varying) fraction of the current optimal wealth Wt ,
regardless of the distribution of the future returns. Moreover, the optimality con-
dition (6.24) characterizing the optimal portfolio process .wt /tD1;:::;T corresponds
exactly, for every date t D 0; 1; : : : ; T 1, to the optimality condition (3.4) obtained
in the static single-period portfolio optimization problem, in the presence of N risky
assets and a risk free asset. As explained above, this shows that, in the case of
a logarithmic utility function, the optimal portfolio problem is unaffected by the

future distribution of asset returns and, similarly, the optimal consumption does
only depend on the current wealth, regardless of the future investment opportunities
(myopic behavior). We want to emphasize that such a myopic behavior is a
peculiarity of the logarithmic utility function and is not shared by other typical
specifications of utility functions.
By relying on Proposition 6.6 together with equation (6.19) and the optimality
condition (6.20), we can prove the following corollary (see Exercise 6.5), which
shows an important property of the optimal wealth process in the case of a
logarithmic utility function.
Corollary 6.7 Let u.x/ D log.x/ in problem (6.6) and denote by .Wt /tD0;1;:::;T the
corresponding optimal wealth process and by .ct /tD0;1;:::;T the optimal consumption
process, characterized by equation (6.23). Then, for any self-financing trading-
consumption strategy .; c/ 2 A .x0 / such that cT > 0, it holds that

WtC1 ˇˇ Wt ct
E ˇFt D ; for all t D 0; 1; : : : ; T 1;
WtC1 Wt ct
where .Wt /tD0;1;:::;T is the wealth process generated by .; c/.

In particular, in view of the above corollary, if one considers the simpler problem of
maximizing the expected utility from consumption at the terminal date T only (i.e.,
without intermediate consumption), then an analogous reasoning allows to show that
the optimal portfolio process .Wt /tD0;1;:::;T is such that, for any portfolio process
.Wt /tD0;1;:::;T , the ratio .Wt =Wt /tD0;1;:::;T is a martingale. In view of this important
property, the portfolio process .Wt /tD0;1;:::;T is also called the numéraire portfolio.
Power Utility Function
Let us now examine the optimal investment-consumption problem for a power

utility function. In this case, as we shall see below, problem (6.6) cannot in general
be reduced to a sequence of disconnected single-period problems, unlike the case
of a logarithmic utility function considered in Proposition 6.6. The proof of the
following proposition is given in Exercise 6.6.
Proposition 6.8 Let u.x/ D x = in problem (6.6), with 2 .0; 1/, and denote by
.Wt /tD0;1;:::;T the corresponding optimal wealth process. Then, the value function is
explicitly given by

1 .Wt /
V.Wt ; t/ D at ; for all t D 0; 1; : : : ; T; (6.25)

where .at /tD0;1;:::;T is an adapted process defined recursively by aT D 1 and

0 11
" ! ˇ #! 1
1
B X
N
1 ˇ C
at D @1 C ıE wn
tC1 .rtC1 rf / C rf
n
atC1 ˇˇFt A ;
nD1
for all t D 0; 1; : : : ; T 1, where .wt /tD1;:::;T denotes the optimal portfolio process.
Moreover, the optimal consumption process .ct /tD0;1;:::;T satisfies
ct D at Wt ; for all t D 0; 1; : : : ; T; (6.26)
and the optimal portfolio process .wt /tD1;:::;T is characterized by

2 ! 1 3
X ˇ
N
ˇ
rf /ˇˇFt 5 D 0;
1
E 4atC1 wn
tC1 .rtC1 rf / C rf
n
.rtC1
n
(6.27)
nD1

Proposition 6.8 shows that, unlike the case of a logarithmic utility function, for
a power utility function the optimal consumption ct at date t is not a deterministic
fraction of the current wealth Wt , for t D 0; 1; : : : ; T. Indeed, .at /tD0;1;:::;T is in
general a stochastic process. Analogously, the value function is not a deterministic
power function of current optimal wealth, but is state dependent, due to the presence
of the Ft -measurable random variable at in (6.25). In particular, for every t D
0; 1; : : : ; T, the random variable at involves the Ft -conditional expectation of a
function of the future returns and, hence, the optimal consumption and portfolio
processes will depend on the conditional distribution of the future investment
opportunities.
The dependence of the optimal portfolio choice on the future investment
opportunities is due to the fact that the optimal investment-consumption choice
consists in determining the optimal trade-off between optimizing the current
portfolio return and hedging against future changes in the investment opportunity
set (intertemporal hedging, see also Merton [1330]). The fact that the multi-
period optimal investment-consumption problem cannot in general be reduced to a
sequence of myopic single-period problems is precisely due to the fact that an agent
tries to dynamically adjust the portfolio policy in order to hedge against changes
in the future investment opportunities. It is important to point out that the hedging
problem faced in a multi-period optimal investment-consumption problem cannot
be reduced to consumption smoothing. Indeed, even in the absence of intermediate
consumption (i.e., considering only the maximization of expected utility of terminal
consumption), an investor will choose to dynamically adjust the portfolio through
time, depending on the future investment opportunities.
In the case of a power utility function and under additional assumptions, the
optimal investment-consumption plan can be shown to be myopic, as in the case
of a logarithmic utility function. More specifically, the following corollary (see
Exercise 6.7 for a proof) provides a sufficient condition for the optimal portfolio
choice to be state independent and such that, for every date t D 1; : : : ; T, the
portfolio optimality condition coincides with the optimality condition (3.4) arising
in the single-period problem.
Corollary 6.9 Under the assumptions of Proposition 6.8, suppose that, for the
optimal portfolio process .wt /tD1;:::;T ,
! ˇ !
X
N
ˇ
rf / C rf ˇˇFt D 0;
1
Cov atC1 ; wn
tC1 .rtC1
n
(6.28)
nD1
1 P
for all t D 0; 1; : : : ; T 1, i.e., atC1 and . NnD1 wn
tC1 .rtC1 rf / C rf / are
n

uncorrelated conditionally on Ft . Then the optimal portfolio process .wt /tD1;:::;T
is characterized by
2 ! 1 3
X ˇ
N
ˇ
E4 wn
tC1 .rtC1 rf / C rf
n
.rtC1
n
rf /ˇˇFt 5 D 0; (6.29)
nD1

P
In particular, the result of Corollary 6.9 does hold if . NnD1 wn tC1 .rtC1 rf /Crf /
n
1
and atC1 are independent, for all t D 0; 1; : : : ; T 1. In Exercise 6.8 we show that, if
the Ft -conditional distribution of the future returns coincides with the unconditional
distribution, for every date t D 0; 1; : : : ; T 1, then the process .at /tD0;1;:::;T reduces
to a deterministic function of time. In particular, due to equation (6.26), the optimal
consumption is then given by a deterministic fraction of the current wealth and the
optimal portfolio can be characterized as in (6.29), similarly as in the case of a
logarithmic utility function. However, in the presence of correlation, the optimal
portfolio choice also takes into account the future investment opportunities and,
hence, cannot be reduced to the static optimality condition (3.4).
Power Utility Function with a Markovian State Process
Under the assumption of a power utility function, we now consider the case where
an agent is only interested in maximizing the expected utility of consumption at the
terminal date T:

CT
max ı E
T
; (6.30)

where the maximum is taken with respect to all RNC1 -valued predictable self-
financing trading strategies .wt /tD1;:::;T parametrized in terms of proportions of
wealth invested in the N C 1 available assets (i.e., over all self-financing trading-
consumption strategies, in the sense of Definition 6.1, with a null consumption
process before date T) satisfying
!
X
N
WtC1 D Wt wntC1 .rtC1
n
rf / C rf ; for all t D 0; : : : ; T 1; (6.31)
nD1
in line with equation (6.19) (recall that WT D CT ).

Similarly as in Brandt [280], we suppose that there exists an Rm -valued (for
m 2 N) adapted stochastic process .zt /tD1;:::;T which can be thought of as a vector
of economic state variables. For t D 1; : : : ; T, let us denote yt WD .rt ; zt / 2 RNCm ,
where rt is the return vector at date t of the N risky assets available in the market.
We assume that the process .yt /tD1;:::;T is a Markov process, in the sense that, for all
t D 1; : : : ; T 1, the Ft -conditional distribution of fytC1 ; : : : ; yT g does only depend
on yt and not on the previous values fy1 ; : : : ; yt1 g of the process. In other words,
for any t D 1; : : : ; T and for any integrable function f , it holds that
E Œf .ytC1 ; : : : ; yT /jFt D E Œ f .ytC1 ; : : : ; yT /jyt :
In this Markovian setting, using equation (6.31), the value function associated
to the optimal investment Problem (6.30) can be written as follows, for x > 0 and
t D 0; 1; : : : ; T 1:

CT ˇˇ

V.x; t/ D sup ı Tt E ˇFt
c2CtC .x/

" !! ˇ #
1 Y
T1 X
N
ˇ
D sup ı E
Tt
x wnsC1 .rsC1
n
rf / C rf ˇFt
ˇ
.ws /sDtC1;:::;T sDt nD1
" !! ˇ #
1 Y
T1 X
N
ˇ
D sup ı E
Tt
x wnsC1 .rsC1
n
rf / C rf ˇy t
ˇ
.ws /sDtC1;:::;T sDt nD1
N t; yt /;
DW V.x;
where VN W RC f0; 1; : : : ; Tg RNCm ! RC . In particular, note that VN is a deter-

ministic function, unlike the general value function V introduced in (6.9). Indeed,
in this Markovian context, the dependence on Ft appearing in representation (6.9)
can be replaced by a dependence on the vector yt containing the current realization
of the N C m state variables. This is a consequence of the Markov property of the
process .yt /tD1;:::;T .
As shown in Proposition 6.3, the Bellman equation (6.11) holds in correspon-

dence of the optimal portfolio strategy .wt /tD1;:::;T , with associated optimal wealth
process .Wt /tD0;1;:::;T , where W0 denotes a given initial wealth. Hence,
N t ; t; yt / D ıEŒV.W
V.W N tC1

; t C 1; ytC1 /jFt ; for all t D 0; 1; : : : ; T 1;
(6.32)
and the characterization (6.17) of the optimal portfolio strategy .wt /tD1;:::;T can be
rewritten as
EŒVN 0 .WtC1

; t C 1; ytC1 /.rtC1
n
rf /jFt D 0; (6.33)
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, where VN 0 denotes the derivative

of the function VN with respect to its first argument. Note that, due to the Markov
property of the process .yt /tD1;:::;T , the conditioning with respect to Ft appearing
in (6.32) and (6.33) can be replaced by a conditioning with respect to yt . In the
present context, the following proposition gives a more explicit representation
of the Bellman equation and a characterization of the optimal portfolio process
.wt /tD1;:::;T .
Proposition 6.10 In the context of Problem (6.30), if the process y D .yt /tD1;:::;T
has the Markov property, then the value function VN introduced above satisfies
" ˇ #
x w> .rtC1 rf / C rf ˇ
N t; yt / D sup ıE
V.x; tC1
˚.t C 1; ytC1 /ˇˇyt ;
wtC1 2RN
(6.34)
for all x > 0 and t D 0; 1; : : : ; T 1, where ˚ W f1; : : : ; Tg RNCm !
RC is a deterministic function satisfying ˚.T; / D 1. Moreover, the condition
characterizing the optimal portfolio process .wt /tD1;:::;T is given by
2 ! 1 3
X ˇ
N
ˇ
E4 wn
tC1 .rtC1
n
rf / C rf .rtC1
n
rf /˚.t C 1; ytC1 /ˇˇyt 5 D 0; (6.35)
nD1

Proof Equation (6.32), together with the definition of the value function V, N the
self-financing condition (6.31) and the Markov property of the process .yt /tD1;:::;T ,
implies the following representation of the Bellman optimality condition, for all
t D 0; 1; : : : ; T 2:
N t; yt /
V.x;
" " ! ˇ #! ˇ #
1 Y
T1
ˇ ˇ
D sup ıE ı E w> ˇ ˇyt
sC1 .rsC1 rf / C rf
T.tC1/
sup x ˇytC1 ˇ
wtC1 .ws /sDtC2;:::;T sDt
6.2 Equilibrium and Pareto Optimality 279
2
>
6 x wtC1 .r tC1 r f / C r f
D sup ıE 4
wtC1
0 20 1 31 3
ˇ ˇ
Y
T1
ˇ ˇ
sup @ı T.tC1/ E 4@ w> .r r / C r A ˇytC1 5A ˇyt 5 :
sC1 sC1 f f ˇ ˇ
.ws /sDtC2;:::;T sDtC1
Defining the function ˚ W f1; : : : ; Tg RNCm ! RC by

0 20 1 31
Y
T1 ˇ
ˇ
˚.t C 1; ytC1 / WD sup @ı T.tC1/
E 4@ wsC1 .rsC1 rf / C rf A
> ˇytC1 5A ;
ˇ
.ws /sDtC2;:::;T sDtC1
for all t D 0; 1; : : : ; T 2 and ˚.T; / D 1, we have thus proved (6.34).

Condition (6.35) then easily follows. t
u
N t; yt / at
In particular, the above proposition shows that the value function V.x;
date t can be represented in terms of the maximization of the expected utility of
wealth at the following date t C 1 multiplied by the deterministic function ˚ which
depends on the value ytC1 of the Markov process y at date t C 1. Moreover, since
N t; yt / D x V.1;
V.x; N t; yt /, for all t D 0; 1; : : : ; T, it holds that
h ˇ i
N t; yt / D ˇ
V.1; sup ıE w>
tC1 .rtC1 rf / C rf ˚.t C 1; ytC1 /ˇyt D ˚.t; yt /;
wtC1 2RN
for all t D 0; 1; : : : ; T. We have thus shown that

N t; yt / D x ˚.t; yt /;
V.x; for all t D 0; 1; : : : ; T:

As a special case, if the processes .rt /tD1;:::;T and .zt /tD1;:::;T are independent and
asset returns at different dates are independent among themselves, then the function
˚ can be reduced to a deterministic function of time (i.e., there is no need to keep
track of the conditioning information) and the optimality condition (6.35) reduces
to the optimality condition characterizing the optimal portfolio in a simple single-
period portfolio choice problem (myopic behavior). In the absence of independence,
the multi-period optimal policy differs from a sequence of single-period optimal
policies, due to the intertemporal hedging effect captured by the interaction with the
term ˚.t C 1; ytC1 /.
6.2 Equilibrium and Pareto Optimality
In this section, we extend to a multi-period setting the equilibrium analysis

developed in Chap. 4 in the context of a single-period economy. In particular, we
shall focus on the relation between the Pareto optimality of an equilibrium allocation
and the completeness of the financial market. Unlike in a single-period economy,
the notion of completeness in a multi-period economy will not depend only on the
terminal dividends of the available securities but also on their (endogenous) price
processes. The presentation of this section is inspired from Huang & Litzenberger
[971, Chapter 7].
In a multi-period setting with dates t 2 f0; 1; : : : ; Tg, we consider an economy
populated by I 2 N agents i 2 f1; : : : ; Ig, with homogeneous beliefs and preferences
characterized by time additive state independent utility functions. For simplicity, we
assume that all the I agents exhibit the same discount factor ı 2 .0; 1/. Every agent
i 2 f1; : : : ; Ig possesses an endowment process ei D .eit /tD0;1;:::;T , which can also
be represented as ei D feit .At /I At 2 Ft ; t D 0; 1; : : : ; Tg, with eit .At / denoting
the endowment of agent i (in units of the reference consumption good) at time t in
correspondence of the event At 2 Ft , with ei0 .A0 / WD ei0 . In this way, the agents’
endowment is allowed to be contingent on the events realized over time.5 We assume
that, for all i D 1; : : : ; I, the endowment process ei is non-negative and, to avoid
trivial situations, that there exists at least one date t 2 f0; 1; : : : ; Tg and one event
At 2 Ft such that eit .At / > 0. Similarly, a consumption process ci D .cit /tD0;1;:::;T
can be represented as ci D fcit .At /I At 2 Ft ; t D 0; 1; : : : ; Tg, with cit .At / denoting
the non-negative consumption of agent i at time t in correspondence of the event
At 2 Ft , with ci0 .A0 / WD ci0 . This represents the fact that the endowment and the
consumption processes are adapted to the information flow F.
For every agent i 2 f1; : : : ; Ig and a consumption process ci , the associated
expected utility is given by
X T X
ui ci0 C ıt At ui cit .At / ; (6.36)
tD1 At 2Ft
where ui W RC ! R is a strictly increasing and strictly concave differentiable utility

function. Observe that the preference functional (6.36) implicitly assumes that the
preferences of the agents do not change over time and depend on time only through
a geometric discount factor.
5
Note that the present structure of endowment process generalizes the case where the endowment
is expressed in terms of traded securities. Indeed, if the economy comprises N traded securities and
agent i 2 f1; : : : ; Ig is endowed at the initial date t D 0 with 0i 2 RN units of the traded assets,
then the corresponding endowment process is given by ei D feit .At /I At 2 Ft ; t D 0; 1; : : : ; Tg,
PN i;n n
with eit .At / D nD1 0 dt .At /, with dt .At / denoting the dividend of security n at time t in
n
correspondence of the event At 2 Ft .

Pareto Optimal Allocations
In the present context, the notion of (ex-ante) Pareto optimal allocation can be
defined as a straightforward generalization of Definition 4.1 to a multi-period
setting. The feasible allocations are given by the collection of non-negative con-
sumption plans fci I i D 1; : : : ; Ig satisfying the feasibility constraint
X
I X
I
cit .At / eit .At /; for all At 2 Ft and t D 0; 1; : : : ; T: (6.37)
iD1 iD1
As discussed in Sect. 4.1, Pareto optimal allocations can be characterized in terms

of the no-trade equilibrium in a representative agent economy. Indeed, an allocation
fci I i D 1; : : : ; Ig is Pareto optimal if and only if it is a solution to the following
optimal consumption problem of a representative agent, for some strictly positive
weights fai I i D 1; : : : ; Ig:
0 1
X
I
X
T X
max ai @ui ci0 C ıt At ui cit .At / A ; (6.38)
fci IiD1;:::;Ig
iD1 tD1 At 2Ft
where the maximization is over all feasible allocations. Due to the strict monotonic-
ity of the utility functions ui , the feasibility constraint (6.37) can be equivalently
expressed as an equality (meaning that the available resources are fully consumed in
correspondence of a Pareto optimal allocation). Interior ex-ante Pareto optimal allo-
cations fci I i D 1; : : : ; Ig can then be characterized by the complete allocation of
the available resources together with the optimality conditions (compare with (4.5)–
(4.6))
0
ai ui .ci
0 / D '0 ; (6.39)
0
ai At ı t ui ci t .At / D 'At ; for all At 2 Ft and t D 1; : : : ; T; (6.40)
for all i D 1; : : : ; I, where f'0 ; 'At I At 2 Ft ; t D 1; : : : ; Tg is a family of

strictly positive Lagrange multipliers associated to problem (6.38)–(6.37). The
above optimality conditions can also be rewritten as
0
At ui ci
t .At / 'A
ı ts
D t ; for all At 2 Ft ; As 2 Ft and s; t D 0; 1; : : : ; T;
As ui0 ci
s .A s / ' As
(6.41)
with A0 WD 1 and 'A0 WD '0 (recall that we always assume that all the events
have a strictly positive probability of occurrence). Observe that the right-hand side
of (6.41) does not depend on i. This implies that, as already discussed in Chap. 4,
the (probability weighted) marginal rates of substitution between consumption in
different dates/states of the world are equal among all the agents in correspondence
of a Pareto optimal allocation. In particular, similarly as in Theorem 4.3, this implies
the co-monotonicity property of Pareto optimal allocations, which means that in
correspondence of a Pareto optimal allocation the optimal consumption of every
agent only depends on the aggregate endowment and only aggregate risk matters.
Note also that, in general, there is a dependence between a Pareto optimal allocation
and the set of weights fai I i D 1; : : : ; Ig defining (6.38) (i.e., the aggregation
property does not necessarily hold in a strong sense).
Arrow Securities and Static Completeness
In Chap. 4, in the context of a single-period economy, we have seen that market

completeness implies the Pareto optimality of equilibrium allocations. As we are
going to show, the same result holds true in a general multi-period setting, if we
assume that a complete set of Arrow securities is available at the initial date t D 0.
In the present multi-period context, an Arrow security for the event At 2 Ft is a
security which pays a dividend equal to one at date t in correspondence of the event
At and zero otherwise. Referring to the information flow introduced at the beginning
of this chapter, it is clear that the total number of Arrow securities is equal to the
total number of knots P of the event tree (excluding the knot corresponding to the
initial date t D 0), i.e., TtD1 t , with t D jFt j.
PT
Let us first assume that all the N WD tD1 t Arrow securities are available
for trade at the initial date t D 0. Under this assumption, it is evident that every
consumption plan can be attained by holding a suitable quantity of the N Arrow
securities. We also assume that markets are only open at the initial date t D 0 and we
denote by qAt the price (at date t D 0) of the Arrow security associated to the event
At at date t (with q0 being the price of one unit of the consumption good at the initial
date t D 0). The expected utility (6.36) can be maximized by simply determining
at the initial date t D 0 the quantities to be bought/sold of the N Arrow securities,
under the budget constraint that the endowment process ei suffices to finance such an
investment in the N Arrow securities. The optimal consumption-investment problem
of agent i 2 f1; : : : ; Ig can then be written as
0 1
X
T X
max @ui .ci0 / C ıt At ui cit .At / A ; (6.42)
fci0 ;cit .At /I At 2Ft ;tD1;:::;Tg
tD1 At 2Ft
under the budget constraint
X
T X X
T X
q0 ci0 C qAt cit .At / D q0 ei0 C qAt eit .At /: (6.43)
tD1 At 2Ft tD1 At 2Ft
Since all the events At are assumed to have a strictly positive probability of
realization, in equilibrium the prices qAt are strictly positive. Indeed, it is easy
to show that, in correspondence of an equilibrium, the prices qAt must be strictly
positive, otherwise arbitrage opportunities would appear (see also Sect. 6.3).
A feasible allocation fci I i D 1; : : : ; Ig together with a collection of Arrow
security prices fq0 ; qAt I At 2 Ft ; t D 1; : : : ; Tg is said to be an equilibrium for the
economy if, for every i D 1; : : : ; I, the consumption plan ci solves Problem (6.42)
for agent i, with the budget constraint (6.43) being satisfied in correspondence of
the price system fq0 ; qAt I At 2 Ft ; t D 1; : : : ; Tg, and markets clear, i.e.,
X
I X
I
t .At / D
ci eit .At /; for all At 2 Ft and t D 0; 1; : : : ; T:
iD1 iD1
This definition can be regarded as an extension to the present context of Defini-

tion 4.6 of an Arrow-Debreu equilibrium. As shown in the following proposition,
it is easy to prove that an equilibrium allocation is Pareto optimal if all N Arrow
securities are available (see also Huang & Litzenberger [971, Section 7.3]).
P
Proposition 6.11 Suppose that all the N D TtD1 t Arrow securities are available
for trade at t D 0. Then any feasible strictly positive allocation fci I i D 1; : : : ; Ig
corresponding to an equilibrium is Pareto optimal.
Proof Let fci I i D 1; : : : ; Ig be an equilibrium allocation with associated prices
fq0 ; qAt I At 2 Ft ; t D 1; : : : ; Tg for the N Arrow securities. For every i D 1; : : : ; I,
the consumption plan ci is a solution to problem (6.42)–(6.43) and, hence, by the
optimality condition, it holds that
0 i
ui .ci
0 / D q0 (6.44)
0
ı t At ui ci .At / D i qAt ; for all At 2 Ft and t D 1; : : : ; T; (6.45)
where i > 0 is the Lagrange multiplier for agent i in correspondence of the

optimum ci . Letting ai WD 1=i , the above optimality conditions reduce to (6.39)–
(6.40). Moreover, since markets clear, condition (6.37) is satisfied as an equality, for
all At 2 Ft and t D 0; 1; : : : ; T. This suffices to prove the Pareto optimality of the
allocation fci I i D 1; : : : ; Ig. t
u
It is important to observe that, if all the N Arrow securities are available for trade
and markets are only open at the initial date t D 0, then the optimal consumption
problem (6.42) reduces to a static problem, where the solution is simply given by
a suitable investment at the initial date t D 0 in the N Arrow securities, which
allow to span all possible consumption plans. We express this property by saying
that, under the present assumptions, static completeness holds. The dynamic feature
of the optimal investment-consumption problem considered in Sect. 6.1 has totally
disappeared.
Actually, under the assumptions of Proposition 6.11, we can prove a stronger

result. Indeed, if all the N Arrow securities are available for trade at the initial
date t D 0 and markets stay open at the subsequent dates t D 1; : : : ; T, then the
same equilibrium allocation will be reached if agents form rational expectations. In
particular, as we are now going to show, there is no need for the markets to remain
open at the future dates t D 1; : : : ; T and, if markets are reopened, then there will
be no trade in equilibrium.
If markets remain open at the dates t D 1; : : : ; T, then the optimal investment-
consumption Problem (6.42) becomes a dynamic problem, similarly to prob-
lem (6.6). In particular, in order to formulate their decisions, agents have to form
expectations about the future prices of the N Arrow securities. In line with the
analysis developed in Chap. 4, we assume that agents exhibit rational expectations
and, therefore, in equilibrium they will perfectly forecast future prices (perfect
foresight or Radner equilibrium, see Sect. 4.2). We assume that the N Arrow
securities are available for trade at the initial date t D 0 with strictly positive prices
fqAt I At 2 Ft ; t D 1; : : : ; Tg. For all As 2 Fs , At 2 Ft and s; t 2 f1; : : : ; Tg, define
the quantities qAs jAt by
8
< qAs ; if s > t and As At I
qAsjAt WD qAt
(6.46)
:0; otherwise:
We can prove that the quantities introduced in (6.46) correspond to the prices
associated with a Radner equilibrium generating the same equilibrium allocation
considered in the setting of Proposition 6.11. In the next proposition, qAs jAt will
represent the price at date t in correspondence of the event At of the Arrow security
paying a unitary amount in correspondence of the event As at the future date s.
Proposition 6.12 Suppose that, for every t D 0; 1; : : : ; T 1 and At 2 Ft ,
all the Arrow securities paying in correspondence of the events As At , for
s D t C 1; : : : ; T, are available for trade. Let fci I i D 1; : : : ; Ig be a strictly positive
feasible allocation such that ci solves (6.42)–(6.43), for every i D 1; : : : ; I. Then
the allocation fci I i D 1; : : : ; Ig and the price system fq0 ; qAs jAt I As 2 Fs ; At 2
Ft ; s; t D 1; : : : ; Tg constitute a Radner equilibrium.
Proof Let fci I i D 1; : : : ; Ig be a feasible allocation such that ci is a solution
to (6.42)–(6.43), for every i D 1; : : : ; I. To prove the proposition, it suffices to show
that, for all i D 1; : : : ; I and for every date t 2 f0; 1; : : : ; Tg and event At 2 Ft ,
the consumption plan fci s .As /I As At ; As 2 Fs ; s D t; : : : ; Tg is a solution to the
following problem:
0 1
X
T X
max @u ct .At / C
i
ı st As jAt ui cs .As / A ;
fcs .As /IAs 2Fs ;sDt;:::;Tg
sDtC1 As 2Fs
under the budget constraint
X
T X X
T X
ct .At / C qAs jAt cs .As / D ci
t .At / C qAs jAt ci
s .As /:
sDtC1 As 2Fs sDtC1 As 2Fs
The consumption plan fci s .As /I As At ; As 2 Fs ; s D t; : : : ; Tg solves the above

problem if (and only if) there exists a strictly positive multiplier iAt such that
0
ui .ci
t .At // D At ;
i
0 i
ı st As jAt ui ci
s .As / D At qAs jAt ; for all As 2 Fs ;
for all s D t; : : : ; T. Due to the optimality conditions (6.44)–(6.45) together

with (6.46) (and recalling that As jAt D As =At , for all As 2 Fs with As At ),
it can be easily checked that the above conditions are satisfied by setting
qAt
iAt WD i :
ı t At
Since At 2 Ft and t 2 f0; 1; : : : ; Tg are arbitrary, the proposition is proved. t

u
Proposition 6.12 shows that, if the complete set of Arrow securities is available
for trade at the initial date t D 0, then there is no trade if markets remain open at
the subsequent dates t D 1; : : : ; T. Indeed, if an equilibrium allocation is reached at
t D 0, then the same allocation will be an equilibrium when agents are allowed
to trade at the subsequent dates t D 1; : : : ; T, with corresponding prices given
by (6.46). Actually, agents will not trade at the dates t D 1; : : : ; T. The quantity
qAs jAt represents the price at date t in event At of the Arrow security paying a unitary
dividend in correspondence of the event As and qAs jAt D 0 for all events As which
are incompatible with the event At (see also Exercise 6.10).
Long-Lived Securities and Dynamic Completeness
We have so far considered economies where the whole set of Arrow securities
is available for trade at the initial date t D 0. In this case, the market is
(statically) complete and equilibrium allocations are Pareto optimal allocations
(see Proposition 6.11). In particular, Pareto optimality is reached by trading in the
optimal way at the initial date t D 0, without any trading activity taking place at the
subsequent dates t D 1; : : : ; T, even if markets remain open (see Proposition 6.12).
We now consider the case where the complete set of Arrow securities is not
necessarily available at t D 0. In this case, unlike the situation considered in
Proposition 6.12, the possibility of trading at the following dates t D 1; : : : ; T
can become valuable, since it allows agents to reallocate wealth depending on the
resolution of uncertainty. As we shall see, under suitable conditions and if markets
are dynamically complete, Pareto optimality P can be reached in equilibrium with a
number of traded securities smaller than TtD1 t .
We adopt the setting of Sect. 6.1. The economy is assumed to comprise N
securities paying non-negative (but possibly null) random dividends at the dates
t D 0; 1; : : : ; T and available for trade at every date t D 0; 1; : : : ; T (long-lived
securities, following the terminology of Huang & Litzenberger [971]). For each
n D 1; : : : ; N, the n-th security is characterized by its (ex-dividend) price process
sn D .snt /tD0;1;:::;T and by its dividend process dn D .dtn /tD0;1;:::;T , both assumed to be
adapted to the information flow F. The cum-dividend price process is then given by
pn D sn Cdn , for all n D 1; : : : ; N. Similarly as at the beginning of Sect. 6.1, we shall
use the notation St D .s1t ; : : : ; sNt /> , Dt D .dt1 ; : : : ; dtN /> and Pt D .p1t ; : : : ; pNt /> .
Since the N securities are available for trade at all dates t D 0; 1; : : : ; T, it is
natural to allow agents to dynamically reallocate their wealth by means of self-
financing trading-consumption strategies, in the sense of Definition 6.1. Each agent
i D 1; : : : ; I is assumed to be endowed with a given quantity 0i D .0i;1 ; : : : ; 0i;N />
of the N traded securities at date t D 0 and, without loss of generality, P we assume
that the total supply of each security is normalized to 1 (i.e., IiD1 0i;n D 1, for all
n D 1; : : : ; N). If there is not a complete set of Arrow securities available at t D 0
and, hence, the investment-consumption problem (6.6) cannot be reduced to a static
optimization problem at the initial date t D 0 (as considered in the first part of this
section), then problem (6.6) is a truly dynamic problem, which can be solved by
relying on the dynamic programming approach developed in Sect. 6.1.
In the context of an economy with N securities traded at the dates t D
0; 1; : : : ; T 1, the notion of Radner equilibrium can be formulated as follows,
extending in a natural way Definition 4.7 (recall that, since the economy is assumed
to end at date T, it always holds that ST D 0).
Definition 6.13 Let D D .Dt /tD0;1;:::;T be a given dividend process, i.e., a non-
negative non-null adapted process. A tuple fS ; DI . i ; ci /; i D 1; : : : ; Ig, with
S D .St /tD0;1;:::;T being a non-negative non-null adapted process and . i ; ci /
a self-financing trading-consumption strategy, for all i D 1; : : : ; I, constitutes a
Radner equilibrium if
(i) for every i D 1; : : : ; I, the strategy . i ; ci / solves problem (6.6) for agent i in
correspondence of the price-dividend couple .S ; D/;
P I i;n
(ii) iD1 t 1, for all n D 1; : : : ; N and t D 0; 1; : : : ; T.
It is easy to show that in equilibrium, if the inequality appearing in part (ii) of
Definition 6.13 holds as an equality, then the aggregate consumption equals the
aggregate
P dividends PNdelivered by the N available securities. In other words, it holds
that IiD1 ci
t D d
nD1 t
n
, for all t D 0; 1; : : : ; T (see Exercise 6.11).
As we are going to show, Pareto optimality can be reached in equilibrium if the
market comprising N long-lived securities is complete in the sense of the following
definition (see also Kreps [1135]), which is formulated with respect to a given price-
dividend couple of processes .S; D/ (recall that, following the notation introduced
in Sect. 6.1, the set C0C .x/ contains all consumption streams which can be financed
starting from a given initial wealth x at date t D 0).
Definition 6.14 A consumption stream c D .ct /tD0;1;:::;T is said to be attainable (or
replicable) if there exists an initial wealth x 2 RC such that c 2 C0C .x/. We say
that the market is dynamically complete if every non-negative adapted process (i.e.,
every consumption stream) is attainable.
According to Definition 6.14, a consumption stream is attainable if there exists
some initial wealth from which one can construct a trading strategy financing the
consumption stream. Clearly, the initial wealth x can be considered as the (market)
value of the consumption stream c 2 C0C .x/ at the initial date t D 0. Note that, if
P
all the TtD1 t Arrow securities are available at t D 0, then the market is trivially
complete in the sense of Definition 6.14, since every consumption stream can be
attained by means of a suitable static portfolio of the Arrow securities. However, as
we are now going to show, dynamic completeness can be more realistically reached
by allowing agents to trade on a much smaller number of long-lived securities.
As a preliminary, following Dothan [581], let us define the so-called bifurcation
index t .At / WD jfAtC1 2 FtC1 W AtC1 At gj, for t D 0; 1; : : : ; T 1. In other words,
the quantity t .At / represents the number of distinct events that can occur at the next
date t C 1 given the realization of the event At at date t. Referring to the event tree
described at the beginning of this chapter, the bifurcation index simply counts the
number of branches leaving from each node of the tree, so that t .At / is the number
of knots at t C 1 which can be reached from the knot corresponding to the event
At at date t. Let us also define WD maxtD0;1;:::;T1 maxAt 2Ft t .At /, representing
the maximum number of branches leaving from any node of the tree. For every date
t D 0; 1; : : : ; T 1 and every event At 2 Ft , we collect in the following matrix the
cum-dividend prices which can be realized at the subsequent date t C 1:
0 1
p1tC1 .A1tC1 / p2tC1 .A1tC1 / : : : pNtC1 .A1tC1 /
B p1 .A2 / p2 .A2 / : : : pN .A2 / C
B tC1 tC1 tC1 tC1 tC1 tC1 C
PtC1 .At / D B :: :: : :: C: (6.47)
@ : : : : : A
1 t .At / 2 t .At / t .At /
ptC1 .AtC1 / ptC1 .AtC1 / : : : ptC1 .AtC1 /
N
For every n D 1; : : : ; N, the n-th column of the above matrix represents the possible
prices of the n-th security in correspondence of all the events which might become
true at date t C 1, given that event At is realized at the preceding date t. Similarly,
for every s D 1; : : : ; t .At /, the s-th row coincides with the vector PtC1 .AstC1 / of the
prices of the N securities at date t C 1 in correspondence of the event AstC1 At .
The following proposition (the proof of which is given in Exercise 6.12) provides
a characterization of dynamic completeness. For any t D 1; : : : ; T and At 2 Ft , we
say that the Arrow security paying a unitary dividend in correspondence of event At
at date t (and zero otherwise) is attainable if there exists an attainable consumption

stream which coincides with the dividend process of the Arrow security.
Proposition 6.15 Given the price-dividend couple of processes .S; D/, the follow-
ing are equivalent:
(i) the market is dynamically complete;
(ii) for every t D 0; 1; : : : ; T 1 and At 2 Ft , it holds that rank.PtC1 .At // D
t .At /; P
(iii) all the TtD1 t Arrow securities are attainable.
Part (ii) of this proposition makes clear that market completeness is strongly
related to the temporal resolution of uncertainty. Indeed, if the uncertainty is
gradually resolved over time (i.e., the number of distinct branches leaving each node
of the tree is small), then it suffices to trade in a small number of long-lived securities
in order to achieve market completeness. On the contrary, if the uncertainty is
resolved in a more abrupt way (i.e., there exist nodes from which many distinct
branches of the tree originate), then many securities are needed to reach market
completeness. This also reflects the fact that in a dynamic setting agents learn over
time that the true state of the world belongs to a refining sequence of events and,
hence, adjust their strategies according to a learning process. However, unlike in the
case of the single-period economy considered in Chap. 4, dynamic completeness
does not only depend on the asset dividends but also on their (endogenous) price
processes.
Part (ii) of Proposition 6.15 implies that a necessary condition for the market
to be dynamically complete is that N , since N t .At / has to hold for
every t D 0; 1; : : : ; T 1 and At 2 Ft . In other words, a necessary condition
for dynamic completeness is that, in correspondence of every node of the event
tree, the number of traded securities is greater or equal than the number of distinct
branches leaving from that node. In view of part (iii) of Proposition 6.15, if markets
are dynamically complete, then every Arrow security can be replicated by trading
dynamically in the available long-lived securities. As a consequence (if there are
no arbitrage opportunities, see Sect. 6.3), the market value at the initial date t D 0
of an Arrow security paying a unitary dividend in correspondence of event At at
date t (and zero otherwise) corresponds to the initial wealth needed to finance a
consumption plan c such that cs .As / D 1As DAt if s D t and zero otherwise.
We are now in a position to prove that, if the market is dynamically complete,
then an equilibrium allocation is Pareto optimal. In view of the equivalence .i/ ,
.iii/ in Proposition 6.15, it is easy to see that, if the market is dynamically complete,
then the equilibrium allocation obtained in the presence of N long-lived securities
(Radner equilibrium, see Definition 6.13) coincides with the equilibrium PT allocation
obtained by allowing agents to trade at the initial date all the tD1 t Arrow
securities at the prices implicit in the price-dividend couple .S ; D/ corresponding
to the Radner equilibrium. This simply follows from the fact that, if the market is
dynamically complete, then the complete set of Arrow securities can be attained
by dynamic trading in the N long-lived securities. We have thus established the

following result.
Proposition 6.16 Let D D .Dt /tD0;1;:::;T be a given dividend process, i.e., a non-
negative non-null adapted process. Let the tuple fS ; DI . i ; ci /; i D 1; : : : ; Ig
correspond to a Radner equilibrium, in the sense of Definition 6.13, and suppose
that the market is dynamically complete in correspondence of the price-dividend
couple .S ; D/. Then the allocation fci I i D 1; : : : ; Ig is Pareto optimal.
If markets are incomplete, then only a subset of the Arrow securities can be
replicated by dynamic trading in the N long-lived securities. In this case, an
allocation corresponding to a Radner equilibrium is not necessarily Pareto optimal.
Note also that the constrained Pareto optimality result obtained in Diamond [571]
under market incompleteness in a two-period economy (see Definition 4.11 in
Sect. 4.2) does not extend to a multi-period setting. Indeed, equilibrium allocations
in an incomplete multi-period market are typically Pareto inefficient, in the sense
that feasible reallocations of wealth can lead to Pareto improvements of the
equilibrium allocation (see also Magill & Quinzii [1286, Chapter 25]).
The condition N is also sufficient for the market to be dynamically
complete in a generic sense (see Kreps [1135] for full details). Indeed, fixing
the information flow, the state space and agents’ preferences, an Arrow-Debreu
equilibrium allocation fcPTI i D 1; : : : ; Ig (i.e., an equilibrium allocation in the
i
economy where all the tD1 t Arrow securities are traded) can be implemented
as a Radner equilibrium (see Definition 6.13) with N long-lived securities
for almost every dividend process of those N securities (i.e., for every dividend
process belonging to a set such that the closure of its complement has Lebesgue
measure zero). In correspondence of such a Radner equilibrium, the market will
be dynamically complete and prices determined according to the agents’ optimality
conditions.
Representative Agent Analysis
Having discussed the notions of equilibrium and market completeness, let us now
generalize to a multi-period setting the representative agent analysis introduced in a
two-period economy in Sect. 4.1, referring the reader to Constantinides [488], Milne
[1344], Huang & Litzenberger [971], Duffie [593] for a more detailed presentation.
As above, we restrict ourselves to an economy populated by I agents charac-
terized by homogeneous beliefs, the same discount factor ı and time additive and
state independent preferences represented by a family of strictly increasing and
strictly concave utility functions fui I i D 1; : : : ; Ig. Letting ei D feit .At /I At 2
Ft ; t D 0; 1; : : : ; Tg be the endowment process of agent i, for i D 1; : : : ; I, the
aggregate endowment process P can be represented as e D fet .At /I At 2 Ft ; t D

0; 1; : : : ; Tg, with et .At / WD IiD1 eit .At /. Under the present assumptions, the result
of Proposition 1.6 can be extended to a multi-period risky setting. More specifically,
the same arguments used in the proof of Proposition 4.4 allow to show that the
representative agent’s utility function can be represented in the form (6.36):
X
T X
u.c0 / C ıt At u ct .At / ; (6.48)
tD1 At 2Ft
where the function u W RC ! R is given by

!
X
I
u.x/ D max a u .x / ;
i i i
(6.49)
.x1 ;:::;xI /2RIC
PI iD1
iD1 x x
i
where fai I i D 1; : : : ; Ig is a collection of strictly positive weights characterizing

the representative agent’s utility function. In particular, observe that the utility
function (6.48) preserves the time additivity and state independence of the individual
preference structures, with the same discount factor ı. It can also be shown that the
function u is strictly increasing and concave (and the inequality constraint appearing
in (6.49) can be equivalently replaced by an equality constraint).
As explained in Sect. 1.3, given the weights fai I i D 1; : : : ; Ig and the aggregate
endowment process e, a feasible allocation fci I i D 1; : : : ; Ig which defines (in
the sense explained before Proposition 1.6) the representative agent’s P utility func-
tion (6.48) is Pareto optimal. Given a complete market where all the TtD1 t Arrow
securities are traded (as considered in the first part of Sect. 6.2) and an aggregate
endowment process e, a feasible allocation fci I i D 1; : : : ; Ig corresponding to an
equilibrium of the economy defines the utility function of a representative agent,
where the weight ai is given by the reciprocal of the Lagrange multiplier i of
the optimal investment-consumption problem of agent i at the optimum, for every
i D 1; : : : ; I (see Exercise 6.13). In this case, the representative agent will simply
consume the aggregate endowment process (no-trade equilibrium).
Relaxing the assumption that the whole set of Arrow securities is available for
trade at the initial date t D 0, the same result can be obtained under the assumption
that there are N long-lived securities which make the market dynamically complete
(in the sense of Definition 6.14). Indeed, as shown in Proposition 6.15, in this
case the whole set of Arrow securities can be replicated by means of long-lived
securities and an allocation corresponding to a Radner equilibrium coincides with an
equilibrium allocation (see Proposition 6.16 and the preceding discussion). In this
case, in correspondence of the no-trade equilibrium, the representative agent will
simply hold the aggregate supply of long-lived securities and consume the aggregate
dividends generated by the N securities.
The optimality conditions corresponding to a no-trade equilibrium of the rep-

resentative agent economy (given by the first order conditions of the optimization
Problem (6.48) in correspondence of the aggregate endowment process e) allow to
characterize the prices of the whole set of Arrow securities in terms of the marginal
rate of substitution of the representative agent’s utility function defined in (6.48)–
(6.49). Indeed, normalizing q0 D 1, it holds that

At u0 e.At /
qAt D ıt ; (6.50)
u0 e0
and

As u0 e.As /
qAs jAt D ı st ; (6.51)
At u0 e.At /
for all As 2 Fs , At 2 Ft and s; t 2 f1; : : : ; Tg with s > t and As At . It is important

to remark that, in general, the economy with I agents and the representative agent
economy only share the same equilibrium prices of the Arrow securities, meaning
that the representative agent economy is only locally representative of the original
economy. Indeed, as pointed out in Sect. 4.3, the representative agent’s utility
function depends on the weights fai I i D 1; : : : ; Ig and, therefore, on the distribution
of the aggregate endowment among the I agents.
As in a two-period economy (see Sect. 4.3 and compare also with Rubinstein
[1485] and Milne [1344]), the aggregation property holds in a strong sense if
agents exhibit homogeneous beliefs, the same discount factor ı, time separable
and state independent preferences with utility functions belonging to the HARA
class (i.e., generalized power, exponential or logarithmic utility functions with the
same cautiousness coefficient, compare with Propositions 4.5 and 4.15). In this case,
the representative agent’s utility function u will also belong to the HARA class
(with the same cautiousness coefficient of the individual utility functions) and there
exists a representative agent in a strong sense, meaning that the economy with I
agents and the single agent economy are observationally equivalent and equilibrium
prices do not depend on the initial resources distribution, since the preferences of
the representative agent will not depend on the latter. Moreover, if a sufficiently
rich set of claims on the aggregate endowment process can be traded, then markets
can be effectively complete (compare with the discussion at the end of Sect. 4.3)
and an equilibrium allocation can be Pareto optimal. In this sense, the validity of
formulae (6.50)–(6.51) can be extended to incomplete markets.
6.3 The Fundamental Theorem of Asset Pricing
In this section, we generalize to a multi-period setting the fundamental results

presented in Sect. 4.4 in the context of a two-period economy. As we shall see, most
of the results can be extended to a dynamic economy in a rather straightforward
way and the interpretation of the major findings remains unchanged. In particular,
the equivalence between the absence of arbitrage opportunities, the existence of
a strictly positive linear pricing functional (and, equivalently, of a risk neutral
probability measure) and the existence of an equilibrium for an economy continues
to hold in a dynamic setting. In line with the structure of Sect. 4.4, we first
characterize the absence of arbitrage opportunities via the existence of a risk neutral
probability measure and then we study the relation between the absence of arbitrage
opportunities and the existence of an equilibrium.
In this section, we assume that the economy comprises N C 1 long-lived
securities, characterized by their price and dividend processes S D .St /tD0;1;:::;T and
D D .Dt /tD0;1;:::;T , respectively. As before, the processes S and D are supposed to
be adapted to the information flow F and the cum-dividend price process is defined
as P WD S C D. Security 0 is supposed to be a risk free asset paying a constant
rate rf > 0, so that p0t D s0t D rft , for all t D 0; 1; : : : ; T 1, and p0T D dT0 D rfT
(without loss of generality, the price of the risk free security at the initial date t D 0
is normalized to p00 D 1).6 The remaining N securities represent risky assets paying
non-negative random dividends at each date t D 0; 1; : : : ; T. In this environment, we
assume that the agents’ trading activity can be described by means of self-financing
trading-consumption strategies, in the sense of Definition 6.1.
No-Arbitrage and Risk Neutral Probability Measures
The notion of arbitrage opportunity introduced in Definition 4.17 can be formulated

as follows in a multi-period setting.
Definition 6.17 For a given couple .S; D/ of price-dividend processes, a non-
negative consumption process c D .ct /tD0;1;:::;T such that c 2 C0C .0/ is said to
be
(i) an arbitrage opportunity of the first kind if there exist a date t 2 f1; : : : ; Tg and
an event At 2 Ft such that ct .At / > 0;
(ii) an arbitrage opportunity of the second kind if c0 > 0.
6
We want to point out that the results presented in this section can be extended to the case where
the assumption of a constant risk free rate rf is relaxed by assuming a deterministic time-varying
rate or a stochastic rate given by a predictable stochastic process.
A non-negative consumption process c D .ct /tD0;1;:::;T is simply said to be an

arbitrage opportunity if it is an arbitrage opportunity of the first kind and/or an
arbitrage opportunity of the second kind.
In other words, an arbitrage opportunity is a non-negative consumption stream c
which is strictly positive in correspondence of some event (at the initial date or at
some future date, with strictly positive probability) and which can be financed from
zero initial wealth, i.e., c 2 C0C .0/. More specifically, an arbitrage opportunity of
the first kind is a consumption stream which does not require any initial wealth at
the initial date t D 0 and yields a non-negative and non-null stream of payoffs at the
dates t 2 f1; : : : ; Tg. An arbitrage opportunity of the second kind is a consumption
stream which can be financed by a strictly negative amount of wealth at t D 0 (in
the sense that 1> S0 D c0 < 0) and leads to a non-negative (but possibly null)
stream of payoffs at dates t D 1; : : : ; T. Clearly, there are no arbitrage opportunities
if and only if C0C .0/ D f0g, with 0 denoting the process which is identically null
at all dates t D 0; 1; : : : ; T (recall that the set C0C consists of non-negative adapted
processes).
As in the case of a two-period economy (see Proposition 4.18), the absence of
arbitrage opportunities implies the validity of the Law of One Price, meaning that
identical consumption streams are necessarily financed by identical initial wealths.
The proof of the next proposition is given in Exercise 6.15.
Proposition 6.18 Let c D .ct /tD0;1;:::;T and cQ D .Qct /tD0;1;:::;T be two consumption
processes financed by the strategies D .t /tD0;1;:::;T and Q D .Qt /tD0;1;:::;T ,
respectively, and such that P.ct D cQ t / D 1 for all t D 0; 1; : : : ; T. Then the following
hold:
Q
(i) if there are no arbitrage opportunities of the second kind, then W0 ./ D W0 ./;
(ii) if there are no arbitrage opportunities of the first kind, then Wt ./ D Wt ./ Q
with probability one, for all t D 1; : : : ; T.
For any consumption process c D .ct /tD0;1;:::;T 2 C0C .x/, the initial wealth x
represents the cost at t D 0 of financing the consumption stream. More generally,
for a self-financing trading-consumption strategy .; c/ 2 A .x0 /, the quantity Wt ./
represents the cost (or market value) at time t of the future consumption stream
.cs /sDt;:::;T , for t 2 f0; 1; : : : ; Tg, i.e., the amount of wealth at time t that allows
to finance the future consumption stream. In this sense, the Law of One Price
(Proposition 6.18) implies that if two consumption processes can be financed by
trading in the available securities and deliver identical payment streams, then they
are financed by the same wealth. The absence of arbitrage opportunities also implies
that the wealth associated to a self-financing trading-consumption strategy must
always be greater (or equal) than the current consumption, as shown in the following
lemma (see Exercises 6.16 and 6.17 for the proofs).
Lemma 6.19 Suppose that the price-dividend couple .S; D/ does not admit arbi-
trage opportunities and let .; c/ 2 A .x0 /. Then the following hold:
(i) P.Wt ./ ct / D 1, for all t D 0; 1; : : : ; T;
(ii) for all t D 0; 1; : : : ; T and At 2 Ft , if there exist s 2 ft C 1; : : : ; Tg and

As At such that cs .As / > 0, then it holds that Wt ./.At / > ct .At /, with
Wt ./.At / denoting the value of the process W./ in correspondence of event
At and date t.
Observe that the strategy N WD tC1 1At considered in the proof of Lemma 6.19
(see Exercise 6.16) realizes an arbitrage opportunity (in the sense of Definition 4.17)
in the single trading period Œt; t C 1 with respect to the couple .St ; PtC1 /. Indeed,
it holds that N > St 0, with strict inequality holding in correspondence of the
event At , and N > .StC1 C DtC1 / D WtC1 ./1At ctC1 1At 0. This remark will
turn out to be useful in the proof of Lemma 6.20 below. Indeed, the absence of
arbitrage opportunities in a multi-period economy can be characterized in terms of
the absence of arbitrage opportunities in each single trading interval Œt; t C 1, for
t D 0; 1; : : : ; T 1. The proof of the following lemma is given in Exercise 6.18.
Lemma 6.20 Let .S; D/ be a couple of price-dividend processes. The following are
equivalent:
(i) the couple .S; D/ admits an arbitrage opportunity;
(ii) there exists t 2 f0; 1; : : : ; T 1g such that the couple .St ; PtC1 / admits an
arbitrage opportunity in the sense of Definition 4.17 in the period Œt; t C 1.
According to the above lemma and recalling the event tree structure described
at the beginning of the present chapter, there are no arbitrage opportunities in the
multi-period economy if and only if there is no arbitrage opportunity (in the sense of
Definition 4.17) in every elementary single-period sub-tree. This observation turns
out to be useful for characterizing the absence of arbitrage in a multi-period setting,
as shown in Theorem 6.23 below. As a preliminary, let us define the notion of risk
neutral probability measure, extending Definition 4.21 to the present setting.
Definition 6.21 A probability measure P on .˝; F / such that P .!/ > 0 for all
! 2 ˝ is a risk neutral probability measure for a couple .S; D/ of price-dividend
processes if, for all t 2 f0; 1; : : : ; T 1g, it holds that
1 n
E stC1 C dtC1
n
jFt D snt ; for all n D 0; 1; : : : ; N; (6.52)
rf
where E ŒjFt denotes the Ft -conditional expectation under the probability mea-
sure P .
Note that, by definition, a risk neutral probability measure P is equivalent to the
original probability measure P, in the sense that P and P share the same null sets.
Let us first derive some useful properties of a risk neutral probability measure.
The proof of the following proposition is based on Definition 6.21 together with
some simple computations and is given in detail in Exercise 6.19.
Proposition 6.22 Let P be a risk neutral probability measure for .S; D/. Then, for
all t 2 f0; 1; : : : ; T 1g, it holds that
" ˇ #
X
T
dsn ˇˇ

snt D E Ft ; for all n D 0; 1; : : : ; N: (6.53)
sDtC1
rfst ˇ
As a consequence, the process

!
snt X dn
t
t C
s
s
rf sD0
rf tD0;1;:::;T
is a martingale under the probability measure P , for all n D 0; 1; : : : ; N. Moreover,

for any trading-consumption strategy .; c/ 2 A .x0 / and for all s; t 2 f0; 1; : : : ; Tg
with s < t, it holds that
" ˇ #
Wt ./ X cr ˇˇ
t1
Ws ./
E C r ˇFs D : (6.54)
rft r
rDs f
rfs
The above proposition shows that, under a risk neutral probability measure P ,
the (ex-dividend) price of a traded security is simply given by the conditional
expectation of the future dividend stream, discounted with respect to the risk free
rate rf , thus explaining the risk neutral terminology. Equivalently, recalling that the
return rtC1n n
of the n-the security is defined as rtC1 D pntC1 =snt (under the implicit
assumption that all prices are strictly positive at every date t D 0; 1; : : : ; T 1), it
holds that
E ŒrtC1
n
jFt D rf ; for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N; (6.55)
meaning that the expected return of each security under a risk neutral probability
measure is equal to the risk free rate.
Proposition 6.22 shows an important martingale property of self-financing
trading-consumption strategies. In fact, equation (6.54) means that, for every
trading-consumption strategy .; c/ 2 A .x0 /, the process
!
Wt ./ X cr
t1
C
rft rr
rD0 f tD0;1;:::;T
is a martingale under a risk neutral probability measure P . This property has inter-
esting consequences for the analysis of optimal investment-consumption problems
of the type (6.6) (see Corollary 6.28 below and compare also with Exercise 6.23).
In particular, the martingale property (6.54) holds for any self-financing trading-
consumption strategy .; 0/.
The following result represents the counterpart to Proposition 4.22 for the case of
a multi-period economy and shows that the absence of arbitrage opportunities can
be characterized in terms of the existence of a risk neutral probability measure.
Theorem 6.23 Let .S; D/ be a couple of price-dividend processes. The following
are equivalent:
(ii) there exists a risk neutral probability measure P , in the sense of Defini-
tion 6.21.
Proof .i/ ) .ii/: by Lemma 6.20, the absence of arbitrage opportunities in the
sense of Definition 6.17 implies that each single-period sub-tree is arbitrage free in
the sense of Definition 4.17. More specifically, for every t D 0; 1; : : : ; T 1 and At 2
Ft , the couple .St .At /; PtC1 .At // is arbitrage free in the sense of Definition 4.17,
with PtC1 .At / being defined as in (6.47) (here extended to the case of N C 1
securities). Hence, in view of Proposition 4.22, for every t D 0; 1; : : : ; T 1 and
t .At / t .At /
At 2 Ft , there exists a vector qtC1 .At / D .q.A1tC1 jAt /; : : : ; q.AtC1 jAt //> 2 RCC
such that
1
PtC1 .At /> qtC1 .At / D St .At /; (6.56)
rf
where St .At / D .s0t .At /; s1t .At /; : : : ; sNt .At //> are the prices of the N C 1 securities
t .At /
at date t in correspondence of event At . Since qtC1 .At / 2 RCC and it holds that
Pt .At /
sD1 q.A s
jA
tC1 t / D 1, the elements of q .A
tC1 t / can be taken as the quantities
characterizing a probability measure (conditional on the event At ) over the events in
FtC1 that are subsets of At .
We now construct a risk neutral probability measure P by aggregating all the
qtC1 .At /, recalling that the information flow F D .Ft /tD0;1;:::;T is a refining sequence
of partitions of ˝ and that the final partition FT consists of all the elementary
events. For any fixed ! 2 ˝ and for every date t D 1; : : : ; T, let At .!/ denote
the event of Ft which contains !. For t D 0; 1; : : : ; T 1, let us also define
q.AtC1 .!/jAt .!// as the element of the vector qtC1 .At .!// corresponding to the
event AtC1 .!/ containing !. Define then
Y
T1

! WD q AtC1 .!/jAt .!/ :
tD0
The probability measure P defined by .1 ; : : : ; j˝j

/ is a risk neutral probability
measure, in the sense of Definition 6.21. Indeed, it holds that ! > 0 for all ! 2 ˝.
Moreover,
X X T1
Y
! D q AtC1 .!/jAt .!/
!2˝ !2˝ tD0
X X X
D q.A1 jA0 / q.A2 jA1 / : : : q.!jAT1 / D 1;
A1 2F1 A2 2F2 !2FT
with A0 D ˝. According to the measure P , the probability of an event At 2 Ft ,

for t D 1; : : : ; T, is given by
X XY
t1

P .At / D ! D q AsC1 .!/jAs .!/ ;
!2At !2At sD0
while the vector qtC1 .At / represents the conditional probabilities associated with
P . In fact:
P .AtC1 \ At / AtC1 \At AtC1

P .AtC1 jAt / D D D D q.AtC1 jAt /;
P .At / At At
for any AtC1 2 FtC1 such that AtC1 At . Together with (6.56), this implies
the validity of condition (6.52), thus showing that P is a risk neutral probability
measure.
.ii/ ) .i/: suppose that there exists a risk neutral probability measure P and let
c 2 C0C .0/. Then, due to equation (6.54), it holds that
XT
1
0 D c0 C E Œct ;
rt
tD1 f
so that the process c is identically null. This shows that C0C .0/ D f0g, meaning that
there are no arbitrage opportunities. t
u
The proof of Theorem 6.23 is constructive, in the sense that it explicitly
constructs a risk neutral probability measure P (in the sense of Definition 6.21)
by decomposing the multi-period economy into a family of elementary single-
period models and by aggregating all the risk neutral probability measures obtained
for each single-period sub-tree. However, Theorem 6.23 can also be proved in a
more abstract way by relying on a separating hyperplane argument, analogously to
Theorem 4.20. Note that the risk neutral probability measure P is not unique in
general (see Theorem 6.26 below).
As in Sect. 4.4, the absence of arbitrage opportunities can be equivalently
characterized in terms of (the properties of) pricing functionals. Let us first introduce
the notion of pricing functional in a dynamic setting. Let c D .ct /tD0;1;:::;T 2 C0C .x/
be a consumption process, for some initial wealth x 2 RC . As discussed after
Proposition 6.18, the quantity x can be regarded as the initial cost (or market value)
at t D 0 for the consumption stream .ct /tD0;1;:::;T . Moreover, due to the Law of One
Price, such initial cost is uniquely defined if there are no arbitrage opportunities.
Therefore, if c 2 C0C .x/, we can write V.c0 ; c1 ; : : : ; cT / WD x, where V is the
function which gives the market value of an attainable consumption plan. Note
also that, if .S; D/ is a couple of price-dividend processes not admitting arbitrage
opportunities, then relation (6.54) implies that V can be represented as
XT
1
V.c0 ; c1 ; : : : ; cT / D x D E Œcs ;
rs
sD0 f
where P is a risk neutral probability measure. This representation shows that the
initial wealth needed to finance an attainable consumption plan can be expressed as
the risk neutral expectation of future discounted consumption.
The following definition is a natural extension of Definition 4.19 to the present
setting. We let K denote the set of all real-valued stochastic processes adapted to
the information flow F.
Definition 6.24 A mapping Q W K ! R is said to be a pricing functional for
.S; D/ if, for every x 2 RC and for every process c D .ct /tD0;1;:::;T 2 C0C .x/, it holds
that
Q.c/ WD Q.c0 ; c1 ; : : : ; cT / D V.c0 ; c1 ; : : : ; cT / D x:
A pricing functional Q W K ! R is said to be linear if
Q.˛c C ˇc0 / D ˛Q.c/ C ˇQ.c0 /;
for all c; c0 2 K and ˛; ˇ 2 R. Moreover, the functional Q W K ! R is said to

be strictly positive if, for every non-negative and non-null process c 2 K , it holds
that Q.c/ > 0. Similarly as in Sect. 4.4, a pricing functional extends the notion of
market value to consumption plans that are not necessarily financed by some trading
strategy. Note that the pricing functional is uniquely defined for every attainable
consumption plan.
We are now in a position to prove that the absence of arbitrage opportunities is
equivalent to the existence of a strictly positive linear pricing functional, as shown
in the following simple corollary.
Corollary 6.25 Let .S; D/ be a couple of price-dividend processes. The following
are equivalent:
(ii) there exists a risk neutral probability measure P ;
(iii) there exists a strictly positive linear pricing functional Q W K ! R.
Proof .i/ , .ii/: this equivalence is the content of Theorem 6.23.
.ii/ ) .iii/: if P is a risk neutral probability measure for .S; D/, then a strictly
positive linear pricing functional Q W K ! R for .S; D/ can be defined as
XT
1
Q.c/ D Q.c0 ; c1 ; : : : ; cT / WD E Œct :
rt
tD0 f
The linearity of Q is evident by construction, while the strict positivity follows from
the fact that the probability measure P is equivalent to P.
.iii/ ) .i/: suppose that there exists a strictly positive linear pricing functional
Q W K ! R and let c 2 C0C .0/ be an arbitrage opportunity. Then, in view of
Definition 6.24, it holds that
Q.c0 ; c1 ; : : : ; cT / D 0:
Since the pricing functional Q is strictly positive, this implies that the process c is
identically zero and, hence, cannot be an arbitrage opportunity. t
u
The proof of the above corollary shows that, similarly to the case of a two-
period economy, there is a one-to-one relation between strictly positive linear
pricing functionals and risk neutral probability measures. Recalling the event-tree
structure described at the beginning of the present chapter, a consumption process
c D .ct /tD0;1;:::;T can be represented as c D fc0 ; ct .At /I At 2 Ft ; t D 1; : : : ; Tg, with
ct .At / denoting the level of consumption at date t in correspondence of event At . By
the Riesz representation theorem, if Q W K ! R is a linear pricing functional, then
it can be represented in terms of a collection fm0 ; mt .At /I At 2 Ft ; t D 1; : : : ; Tg, so
that, for any c 2 K ,
X
T X
Q.c/ D m0 c0 C mt .At /ct .At /: (6.57)
tD1 At 2Ft
Similarly, if P is a risk neutral probability measure, the proof of Theorem 6.23

shows that it can be represented by the collection fAt I At 2 Ft ; t D 1; : : : ; Tg, so
that
XT
1 XT
1 X
E Œc t D c 0 C At ct .At /: (6.58)
rt
tD0 f
rt
tD1 f At 2Ft
As discussed after Theorem 4.20, the quantities fm0 ; mt .At /I At 2 Ft ; t D 1; : : : ; Tg

appearing in (6.57) admit the interpretation of state prices associated to one unit of
consumption in correspondence of the different dates/events (Arrow securities). In
particular, consider the consumption plan 1 D .0; : : : ; 0; 1/ which is identically null
before the final date T and always equal to one at T. Of course, the consumption
plan 1 can be attained by simply holding a quantity 1=rfT of the risk free security.
Hence, in view of Definition 6.24, it holds that

X 1
mT .!/ D Q.1/ D V.1/ D :
!2˝
rfT
The same argument can be applied for t D 0; 1; : : : ; T 1, thus showing that

X 1
mt .At / D ; for all t D 0; 1; : : : ; T 1:
rft
At 2Ft
These observations imply that, if Q is a strictly positive linear pricing functional,

then the quantities
mt .At /
At D P ; for At 2 Ft and t D 0; 1; : : : ; T;
A2Ft mt .A/
define a risk neutral probability measure. Conversely, similarly as in the proof of

Corollary 6.25, if P is a risk neutral probability measure then the quantities
At
mt .At / D ; for At 2 Ft and t D 0; 1; : : : ; T;
rft
define a strictly positive linear pricing functional. This last relation shows how state
prices can be generated from a risk neutral probability measure P .
The following theorem generalizes Theorem 4.26 and characterizes dynamic
market completeness in terms of risk neutral probability measures (this result is
typically referred to as the second Fundamental Theorem of Asset Pricing).
Theorem 6.26 Let .S; D/ be a couple of price-dividend processes not admitting
arbitrage opportunities. The following are equivalent:
(i) the market is dynamically complete, in the sense of Definition 6.14;
(ii) there exists a unique risk neutral probability measure P .
Proof In view of Proposition 6.15, the market is dynamically complete if and only
if, for all At 2 Ft and t D 0; 1; : : : ; T 1, it holds that rank.PtC1 .At // D t .At /.
Hence, the equivalence .i/ , .ii/ follows by noting that equation (6.56) in the proof
of Theorem 6.23 has a unique solution for all At 2 Ft and t D 0; 1; : : : ; T 1 if
only if condition .ii/ of Proposition 6.15 is satisfied. t
u
In view of the above result, if the market is free of arbitrage opportunities and
dynamically complete then there exists a unique risk neutral probability measure
or, equivalently, a unique strictly positive linear pricing functional. Furthermore, the
market is dynamically complete if and only if there exists a unique set of strictly
positive prices for the Arrow securities (state prices), which fully characterize the
strictly positive linear pricing functional. Note that, in a dynamically complete
market, the whole set of prices of the Arrow securities is fully determined by the
couple .S; D/. Of course, this is no longer true in the case of incomplete markets.
Note, however, that there is no general implication between the absence of arbitrage
opportunities and market completeness, in the sense that one can construct examples
of dynamically complete markets admitting arbitrage opportunities.
The discussion made in Sect. 4.4 after Proposition 4.22 can be extended to
the present multi-period context, with the same interpretation. In particular, if the
market is dynamically complete, in the sense of Definition 6.14, then S every Cnon-
negative adapted process can be attained, meaning that K C D x2RC C0 .x/,
where K C denotes the set of non-negative adapted stochastic processes. In other
words, given an appropriate initial wealth, every non-negative consumption plan can
be financed by trading in the N C 1 available securities. Moreover, it can be easily
verified that, if a market is dynamically complete in the sense of Definition 6.14,
then every real-valued process c D .ct /tD0;1;:::;T 2 K can be financed by a trading
strategy. This simply follows by considering the decomposition c D cC c of c into
its positive and negative parts and by the linearity of the self-financing
S condition.
Hence, in a dynamically complete market it holds that K D x2R C0 .x/, where we
denote by C0 .x/ the set of real-valued adapted processes c D .ct /tD0;1;:::;T such
that there exists a trading strategy D .t /tD1;:::;T satisfying the self-financing
condition (6.7) with respect to x. In a dynamically complete market, the arbitrage
free price (market value) of a process c 2 K is uniquely given by
XT
1
Q.c/ D V.c/ D c0 C t E Œct :
r
tD1 f
By following the same arguments given in Exercise 4.20, it can be shown that, if it
was possible to attain a consumption plan c starting from an initial wealth x ¤ Q.c/,
then the market would admit arbitrage opportunities.
Under the assumption of a complete market, the above discussion leads us to the
risk neutral valuation paradigm. Indeed, a contingent claim X can be represented
by an adapted process X D .xt /tD1;:::;T 2 K , with xt denoting the random payoff
of the contingent claim at date t. We say that a contingent claim is European if the
process X is identically null before the terminal date T, i.e., the claim only delivers a
non-null payoff at its expiration date (maturity). The previous arguments imply the
following simple but important result.
Proposition 6.27 Suppose that the couple .S; D/ of price-dividend processes does
not admit arbitrage opportunities and is such that the market is dynamically
complete. Let X D .xt /tD1;:::;T be a contingent claim. Then, the unique arbitrage
free price q0 .X/ of X at date t D 0 is given by
XT
1
q0 .X/ D Q.X/ D E Œxt : (6.59)
rt
tD1 f
In particular, in the case of a European contingent claim X, it holds that
1
q0 .X/ D Q.X/ D E ŒxT :
rfT
Proof Since the market is dynamically complete, there exists a trading strategy D
.t /tD0;1;:::;T such that the couple .; .0; x1 ; : : : ; xT // is self-financing, starting from
some initial wealth x0 2 R (i.e., .; .0; x1 ; : : : ; xT // 2 A .X0 /). Hence, in view of
equation (6.54) together with Definition 6.24, it holds that
XT
1
Q.X/ D V.0; x1 ; : : : ; xT / D W0 ./ D t E Œxt D x0 :
r
tD1 f
t
u
As observed above (compare also with Exercise 4.20), q0 .X/ D Q.X/ is the
unique price such that, if the contingent claim X is introduced in the market with
price q0 .X/, then no arbitrage opportunity arises.
If the market is incomplete then the situation is in general more complicated and
there exist elements of K that cannot be attained. In this case, one has to distinguish
between the elements of K that can be attained and the unattainable consumption
streams/contingent
S claims. In the first case, i.e., for the elements belonging to
x2R C0 .x/, the arbitrage free price is still uniquely defined as in Proposition 6.27
above, regardless of the choice of the risk neutral probability measure (see Exer-
cise 6.21). In the case Sof unattainable consumption plans/contingent claims, i.e., for
the elements of K n . x2R C0 .x//, there exist infinitely many arbitrage free prices.
Similarly as in Definition 4.23, one can prove the existence of an interval of arbitrage
free prices by introducing the super-replication price and the sub-replication price.
In particular, analogously to Proposition 4.25, this interval of arbitrage free prices
can be shown to be generated by the family of discounted risk neutral expectations
of the consumption stream, with respect to all risk neutral probability measures. Any
candidate price not belonging to such an interval would inevitably lead to arbitrage
opportunities (compare with Exercise 4.23).
The existence of a risk neutral probability measure allows to give an alternative
representation of the budget constraint in the optimal investment-consumption
problem (6.6), as shown in the following corollary, under the assumption of
complete markets (see Exercise 6.23 for a related result in the case of possibly
incomplete markets). Recall that, in the original formulation of problem (6.6), an
agent maximizes the expected utility of consumption over all self-financing trading-
consumption strategies .; c/ starting from some given initial wealth x > 0.
Corollary 6.28 Let .S; D/ be a couple of price-dividend processes not admitting
arbitrage opportunities. Suppose that the market is dynamically complete and let
P be the (unique) risk neutral probability measure. Then, for any x > 0, the
maximization in problem (6.6) can be taken over all non-negative adapted processes
c D .ct /tD0;1;:::;T satisfying
XT
1
c0 C t E Œct D x: (6.60)
r
tD1 f
Proof Observe that the maximization in problem (6.6) is over all c 2 C0C .x/. Since
the market is complete and P is the risk neutral probability measure, the claim
follows since
( )
XT
1
C C
C0 .x/ D c 2 K W c0 C E Œct D x :
rt
tD1 f
PT
Indeed, if c 2 C0C .x/ K C
, Proposition 6.22 implies that c0 C 1
tD1 rft E Œct D x.
On the contrary, if c 2 K C and the market is complete, then there exists a trading
strategy .t /tD0;1;:::;T which finances the consumption process c, so that, in view of
Proposition 6.22,
XT
1
W0 ./ D c0 C t E Œct D x;
r
tD1 f
for some x > 0, thus showing that c 2 C0C .x/. t

u
Corollary 6.28 provides an important insight to address optimal investment-
consumption problems of the type (6.6) and leads to the so-called martingale
approach (see, e.g., Cox & Huang [504]). In a nutshell, in the context of complete
markets, the martingale approach corresponds to a two-step approach to the solution
of problem (6.6): in the first step the preference functional in (6.6) is simply
maximized over all non-negative adapted processes c 2 K C satisfying the budget
constraint in the form (6.60), with respect to the risk neutral probability measure
P . In a second step, by relying on the completeness of the market, the optimal
trading strategy is characterized as the trading strategy which finances the optimal
consumption process.
No-Arbitrage and Equilibrium
After having characterized the absence of arbitrage opportunities in terms of

risk neutral probability measures or, equivalently, strictly positive linear pricing
functionals, we now address the relation between the absence of arbitrage and the
existence of an equilibrium. As we shall see, the following results can be regarded
as rather straightforward extensions to the present setting of the results obtained in
Sect. 4.4 in the case of an economy with two dates.
We consider an economy comprising N C 1 traded securities (long-lived secu-

rities) characterized by the couple .S; D/ of price-dividend processes, with security
0 being a risk free asset with rate of return rf > 0. As in Sect. 6.2, we suppose
that the economy is populated by I agents, with state independent and time
additive preferences described by the utility functions fui I i D 1; : : : ; Ig. Each agent
i D 1; : : : ; I is endowed with N0i 2 RNC1 shares of the N C 1 traded securities (as
P
in Sect. 6.2, without loss of generality, we assume that IiD1 N0i;n D 1, for all n D
0; 1; : : : ; N). A first and basic result shows that the absence of arbitrage opportunities
is a necessary condition for the existence of a solution to the optimal investment-
consumption problem (6.6) and, hence, for the existence of an equilibrium of the
economy. This generalizes to a multi-period economy the result of Proposition 4.27
(see also Corollary 4.28).
Proposition 6.29 For any i D 1; : : : ; I, suppose that ui W RC ! R is strictly
increasing and that there exists an optimal solution . i ; ci / 2 A .x/ to prob-
lem (6.6), for some x 2 RC , in correspondence of the couple .S; D/ of price-dividend
processes. Then there are no arbitrage opportunities. As a consequence, if the
couple .S; D/ corresponds to a Radner equilibrium, in the sense of Definition 6.13,
then there are no arbitrage opportunities.
Proof Arguing by contradiction, let . i ; ci / 2 A .x/ be a solution to problem (6.6)
for agent i and let cQ 2 C0C .0/ be an arbitrage opportunity, financed by some strategy
Q D .Qt /tD0;1;:::;T . Consider then the strategy .O i ; cO i / defined by O i WD i C Q and
Q cQ / 2 A .0/, it follows that .O i ; cO i / 2 A .x/ and
cO i WD ci C cQ . Since .;
W0 .O i / D W0 . i / C W0 ./
Q D W0 . i / D x
implies that cO i 2 C0C .x/. Moreover, in view of Definition 6.17, there exist a date
t 2 f0; 1; : : : ; Tg and an event At 2 Ft such that cO it D ci t C c Q t > ci
t on At .
Since u is strictly increasing, this contradicts the optimality of . ; c /. In view of
i i i
Definition 6.13, the last claim of the proposition is obvious. t

u
Of course, since an arbitrage opportunity corresponds to the possibility of financing
a non-null consumption stream out of nothing, the above result is hardly surprising.
We now study the converse and more interesting implication, namely whether
the absence of arbitrage opportunities implies the existence of an equilibrium of the
economy. As a first step, allowing for risk neutral agents, the following version of
Proposition 4.29 can be established by extending to the present multi-period setting
the proof of Proposition 4.29 (see Exercise 6.25).
Proposition 6.30 Let .S; D/ be a couple of price-dividend processes not admitting
arbitrage opportunities. Then there exists an economy populated by I risk neutral
agents endowed with N0i 2 RNC1 shares of the N C 1 securities, for all i D 1; : : : ; I,
with beliefs given by risk neutral probabilities and a risk free discount factor
ı WD 1=rf , such that the couple .S; D/ corresponds to a Radner equilibrium of this
economy.
As explained in the case of an economy with two dates, the above proposition
also gives an explanation for the risk neutral terminology (compare also with the
discussion following Proposition 4.29).
Let us now move to the more interesting case of an economy populated by risk
averse agents. As a preliminary, similarly as in the case of a two-period economy
(compare with Proposition 4.30 and see Exercise 6.27 for more details), under
suitable assumptions on the utility function ui , it can be shown that the absence of
arbitrage opportunities is not only necessary (as shown in Proposition 6.29 above)
but also sufficient in order to ensure the existence of a solution to the optimal
investment-consumption problem (6.6).
Proposition 6.31 For any i D 1; : : : ; I, suppose that ui W RC ! R is continuous
and strictly increasing and concave. Then, there exists an optimal solution to the
investment-consumption problem (6.6) in correspondence of the couple .S; D/ of
price-dividend processes if and only if .S; D/ does not admit arbitrage opportunities.
The last proposition implies that, given an arbitrage free couple .S; D/ of
price-dividend processes, there exists a single-agent economy such that .S; D/ is
associated to a (no-trade) equilibrium of such an economy. This can be proved as
a direct extension of Proposition 4.31, by endowing the single agent at date t D 0
with the optimal consumption plan fci .At /I At 2 Ft ; t D 0; 1; : : : ; Tg, recalling the
notation introduced at the beginning of Sect. 6.2. The interesting implication of this
observation is given by the fact that it allows to construct a risk neutral probability
measure in terms of the marginal rates of substitution at the optimum, as shown in
the following proposition.
Proposition 6.32 For any i D 1; : : : ; I, let ui W RC ! R be strictly increasing and
concave and denote by ci D .cit /tD0;1;:::;T the corresponding consumption process
solving problem (6.6) with respect to the couple .S; D/ of price-dividend processes.
Then a risk neutral probability measure P can be defined by letting
0
T .!/
ui ci
! WD .ırf / T
! ; for all ! 2 ˝: (6.61)
ui0 .ci
0 /
Furthermore, the conditional probabilities associated to P are given by

i0 i

ts u ct .At /
At jAs
WD P .At jAs / D .ırf / At jAs ; (6.62)
ui0 ci
s .As /
for all s; t 2 f0; 1; : : : ; Tg with t > s and As 2 Fs , At 2 Ft such that At As .

Proof If ci is an optimal consumption process associated to the utility function
ui , then Corollary 6.5 (in the case where a risk free security is traded) together
with (6.14) implies that
0 0
ui .ci
t / D ırf EŒu .ctC1 /jFt ;
i i
for all t D 0; 1; : : : ; T 1: (6.63)
By iterating and using the tower property of the conditional expectation, this implies
0 i0 i
that ui .ci
0 / D .ırf / EŒu .cT /, so that
T
0
X X T .!/
0
.ırf /T EŒui .ci
ui ci T /
! D .ırf / T
! D D 1:
!2˝ !2˝
ui0 .ci
0 /
i0
u .c0 /
i
Since ui is assumed to be strictly increasing, this shows that f! I ! 2 ˝g describes

a probability measure on .˝; F /. Moreover, in view of (6.15) and using repeatedly
equation (6.63), it holds that, for any n D 0; 1; : : : ; N and t D 0; 1; : : : ; T 1,

n ˇ
i0 i0 i0 i ptC1 ˇ
u .ci
t /st
n
D ıEŒu .ci
tC1 /ptC1 jFt
n
D .ırf / Tt
E u .cT / ˇFt ;
rf
so that, using again the fact that ui is strictly increasing and equation (6.63),
h 0 ˇ i h 0 ˇ i
pntC1 ˇ pntC1 ˇ
.ırf /T E ui .ci
T / rf ˇFt .ırf /T E ui .ci
T / rf ˇFt
snt D D
.ırf /t ui0 .ci t / .ırf /T E ui0 .ci
T /jFt

0 ˇ
T / tC1 ˇ
.ırf /T ui .ci pn
E 0 rf ˇFt

n ˇ
ui .ci 0 / ptC1 ˇ
D h ˇ i DE ˇFt ;
T /ˇ
0
.ırf /T ui .ci rf
E 0 i
u .c /
i ˇF t
0
where the last equality follows from the conditional version of the Bayes rule.7 This
shows that P is a risk neutral probability measure, in the sense of Definition 6.21.
The conditional probabilities given in (6.62) can be obtained as shown in Exer-
cise 6.20. t
u
In particular, representation (6.62) (by taking s D 0 and As D ˝) yields
i
i0

tu ct .At /
At D .ırf / At DW `t .At /At ; for all At 2 Ft and t D 0; 1; : : : ; T;
ui0 ci
0
(6.64)
where f`t .At /I At 2 Ft ; t D 0; 1; : : : ; Tg is defined by
0
ui .ci
t .At //
`t .At / WD .ırf /t ; for At 2 Ft and t D 0; 1; : : : ; T;
ui0 .ci
0 /
7
Recall that the conditional version of the Bayes rule can be stated as follows. Let P and P be two
probability measures on the space .˝; F / such that P
P (i.e., P and P are equivalent). Then,

for any bounded random variable X, it holds that E ŒXjFt D EŒX dP dP
jFt =EŒ dP
dP
jFt , for each
t D 0; 1; : : : ; T.
and represents the likelihood ratio between the risk neutral probability measure P
and the original probability measure P. The process .Mt /tD0;1;:::;T defined by
i0
tu .ci
t /
Mt WD ı i0
; for t D 0; 1; : : : ; T; (6.65)
u .ci
0 /
represents the stochastic discount factor (or pricing kernel). Similarly as in the basic
case of a two-period economy (see Sect. 4.4), this shows that the likelihood ratio and
the stochastic discount factor are determined by the marginal utility of an agent in
correspondence of the optimal consumption stream. In Sect. 6.4, we shall derive
several asset pricing relations
P by relying on expression (6.64).
Note also that, since At 2Ft At D 1 for all t D 0; 1; : : : ; T, relation (6.64)
implies that
0
1 EŒui .ci
t /
D 0 ; for all t D 0; 1; : : : ; T;
.ırf / t u .c0 /
i i
which, in turn implies that

0
t .At /
ui ci
At D At ; for all At 2 Ft and t D 0; 1; : : : ; T:
EŒui0 .ci
t /
Furthermore, if the subjective discount factor ı coincides with 1=rf , relation (6.62)
can be expressed as
0
ui cit .At /
At jAs D i0 i At jAs ;
u cs .As /
for all s; t 2 f0; 1; : : : ; Tg with t > s and As 2 Fs , At 2 Ft such that At As .

Proposition 6.32 as well as the above relations concern the optimal investment-
consumption problem of an individual agent. Under suitable assumptions, we can
extend these results to the case of an economy populated by I risk averse agents by
relying on the existence of a representative agent, as discussed in Sect. 6.2. More
specifically, if the I agents exhibit state independent time homogeneous preferences
characterized by the utility functions fui I i D 1; : : : ; Ig and by the same discount
factor ı and if the market is complete, then the equilibrium prices can be supported
by the no-trade equilibrium of a single agent (representative agent) economy, with
the representative agent’s utility function u being constructed as in (6.48). Under
these assumptions, Proposition 6.32 and, hence, relation (6.64) can be established
with respect to the representative agent’s utility function u, so that
0

tu et .At /
At D .ırf / At DW `t .At /At ; for all At 2 Ft and t D 0; 1; : : : ; T;
u0 e0
(6.66)
where e D fet .At /I At 2 Ft ; t D 0; 1; : : : ; Tg denotes the aggregate endowment

process. Hence, the representative agent’s marginal rate of substitution in cor-
respondence of the aggregate endowment generates a stochastic discount factor.
Under additional hypotheses on the agents’ utility functions (see the discussion at
the end of Sect. 6.2), the aggregation property can be shown to hold in a strong
sense, meaning that the representative agent’s utility function only depends on the
aggregate endowment and not on the specific distribution among the I agents. In
this case, the stochastic discount factor (6.66) constructed from the representative
agent’s utility function u will also be independent on how the aggregate endowment
is distributed among the agents (see also Sect. 4.4 for a related discussion).
The Multi-Period Binomial Model
We close this section by briefly extending to a multi-period context the simple

binomial model introduced at the end of Sect. 4.4. Following the seminal paper Cox
et al. [507], we assume that the market comprises two securities: a risk free security
(bond), whose price evolves deterministically as s0t D rft , for all t D 0; 1; : : : ; T,
with rf being the risk free interest rate, and a risky security (stock) which does not
pay dividends and such that its price process .s1t /tD0;1;:::;T evolves according to
s1t D s1t1 "t ; for all t D 1; : : : ; T;
with s10 > 0 and where ."t /tD1;:::;T is a sequence of random variables taking values
in fd; ug, for some 0 < d < u, and such that
P."t D u/ D pt and P."t D d/ D 1 pt ;
with pt 2 .0; 1/, for all t D 1; : : : ; T. We assume that the information flow of the
economy corresponds to the information generated by the stochastic evolution of
the stock price. Note that the risk free asset does not generate any information, since
its price process is deterministic.
The following proposition summarizes the properties of the multi-period bino-
mial model.
Proposition 6.33 In the context of the multi-period binomial model described
above, the market is dynamically complete. Moreover, there are no arbitrage
opportunities if and only if d < rf < u. In this case, the risk neutral probability
measure P is characterized by
rf d u rf
P ."t D ujFt / D u WD and P ."t D djFt / D d WD ;
ud ud
(6.67)
for all t D 1; : : : ; T.
Proof Similarly as in the proof of Proposition 4.32, since rf > 0 and u > d, it holds
that rank.PtC1 .At // D 2 D t .At / for every t D 0; 1; : : : ; T 1. Therefore, in view
of Proposition 6.15, the market is dynamically complete. By Theorem 6.23, there
are no arbitrage opportunities if and only if there exists a risk neutral probability
measure P . In the present case of the multi-period binomial model, condition (6.52)
reads
1
uP ."tC1 D ujFt / C dP ."tC1 D djFt / D 1;
rf
for all t D 0; 1; : : : ; T 1. It is easy to see that the above equation admits a unique
strictly positive solution .P ."tC1 D ujFt /; P ."tC1 D djFt // such that
P ."tC1 D ujFt / C P ."tC1 D djFt / D 1
if and only d < rf < u. In that case, the probability measure P is characterized
by (6.67), for all t D 1; : : : ; T. t
u
According to the above proposition, if there are no arbitrage opportunities, then
the risk neutral probabilities of an upward movement and of a downward movement
in each period Œt; tC1 are given by .rf d/=.ud/ and .urf /=.ud/, respectively.
In particular, observe that, under the risk neutral probability measure P , the
sequence ."t /tD1;:::;T is composed of independent and identically distributed random
variables, regardless of their distribution under the original probability measure P.
Note also that the condition d < rf < u characterizing the absence of arbitrage
opportunities in the multi-period binomial model coincides with the condition
obtained in the case of the single-period binomial model (see Proposition 4.32).
Indeed, in view of Lemma 6.20, there is an arbitrage opportunity in the multi-period
model if and only if there is an arbitrage opportunity in some elementary single-
period sub-tree of the model.
Let us conclude by considering the problem of valuing a European contingent
claim paying the random payoff X D f .s1T / at the maturity T, for some deterministic
function f W RC ! RC . For instance, in the case of an European Call option,
the payoff f .s1T / is given by maxfs1T KI 0g, for some fixed strike price K > 0
(see also Exercises 6.24 and 6.26 for explicit examples of binomial models). As
explained above, since the market is dynamically complete, any contingent claim
can be attained by trading in the two available securities.
Corollary 6.34 In the context of the multi-period binomial model described above,
the arbitrage free price q0 .f .s1T // of an European contingent claim paying the
random payoff f .s1T / at the expiration date T can be explicitly computed as
T
1 1 1 1 X T
q0 f .sT / D T E f .sT / D T .u /j .d /Tj f s10 uj dTj ;
rf rf jD0 j

T
where D TŠ
jŠ.Tj/Š is the binomial coefficient.
j
Proof The claim directly follows from Proposition 6.27 by noting that the random
variables ."t /tD1;:::;T are independent and identically distributed as Bernoulli random
variables under the risk neutral probability measure P . Moreover, the number
of realizations that have exactly j upwards movements is given by the binomial
coefficient TŠ=.jŠ.T j/Š/. t
u
6.4 Asset Pricing Relations
In the present section, we extend to a multi-period economy the asset pricing

relations presented in Chap. 5 in the context of a single-period economy. As outlined
at the beginning of the present chapter, one of the fundamental features of a multi-
period economy is represented by the dynamic evolution of information. Hence, in
this section we shall derive conditional asset pricing relations, meaning that, at each
date t 2 f0; 1; : : : ; T 1g, the relation will be expressed in terms of expectations and
covariances conditioned on the current information Ft available at date t. Of course,
by the tower property of conditional expectations, conditional asset pricing relations
also lead to unconditional asset pricing relations. However, we want to point out that
the validity of a conditional linear factor pricing model does not necessarily imply
the validity of an unconditional linear factor pricing model (see Cochrane [465,
Chapter 8] for a discussion on this point).
By relying on the equilibrium analysis developed in the preceding section, we
aim at explaining the conditional expected excess returns of the traded securities
in terms of the aggregate consumption process. Note that, unlike in a two-period
economy, in a multi-period economy the aggregate wealth does not coincide with
the aggregate consumption (except, of course, at the terminal date t D T). Indeed, as
can be deduced from relation (6.3), the aggregate wealth at date t D 0; 1; : : : ; T 1,
is equal to the aggregate consumption plus the aggregate ex-dividend prices of the
existing securities, so that the aggregate consumption only represents a fraction of
the aggregate wealth. Hence, unlike in Sect. 5.1, we can no longer freely interchange
aggregate wealth and aggregate consumption.
In this section, we assume that the economy is populated by I 2 N agents
with state independent and time homogeneous preferences, homogeneous beliefs
and with identical discount factor ı 2 .0; 1/. We also assume that the market
6.4 Asset Pricing Relations 311
is (effectively) complete, so that equilibrium prices can be supported by the no-

trade equilibrium of a single agent (representative agent) economy, where the
representative agent’s utility function u./ can be constructed as in (6.48). As
explained at the beginning of Sect. 6.3, we suppose that there are N C 1 traded
securities, with security 0 being a risk free security paying the constant rate rf .
Without loss of generality, the total supply of each security is normalized to one.
As above, we denote by e D .et /tD0;1;:::;T the aggregate endowment process,
which Pis entirely determined by the dividend processes of the traded securities, i.e.,
N
et D nD0 dt , for all t D 0; 1; : : : ; T. As explained in Sect. 6.2, the aggregate
n
consumption process coincides with the aggregate dividends paid by the traded
securities. In correspondence of a no-trade equilibrium, the representative agent
simply consumes at each date t D 0; 1; : : : ; T the aggregate dividends paid at date t
by the N C 1 available securities.
As shown in the previous section (see Proposition 6.32 and equation (6.66)),
under the present assumptions the following fundamental asset pricing relation
holds, for all t D 0; 1; : : : ; T 1, with respect to the representative agent’s marginal
rate of substitution8 between consumption at date t C 1 and consumption at date t:

0 ˇ
u .etC1 / n ˇ
snt D E ı 0 .dtC1 C sntC1 /ˇFt ; for all n D 0; 1; : : : ; N: (6.68)
u .et /
In turn, by applying the tower property of conditional expectation and recalling that
snT D 0, relation (6.68) implies that, for all t D 0; 1; : : : ; T 1,

0
X u .etCs / n ˇˇ
Tt
snt D E ıs 0 dtCs ˇFt ; for all n D 0; 1; : : : ; N: (6.69)
sD1
u .et /
In terms of returns, equation (6.68) can be rewritten as follows:

u0 .etC1 / n ˇˇ
1DE ı 0 r ˇFt
u .et / tC1

0 0 (6.70)
u .etC1 / ˇˇ u .etC1 / n ˇˇ
DE ı 0 ˇFt EŒrtC1
n
jFt C Cov ı 0 ; rtC1 ˇFt ;
u .et / u .et /
for all n D 0; 1; : : : ; N and t D 0; 1; : : : ; T 1. Moreover, since the representative

agent’s utility function u is assumed to be strictly increasing (so that u0 ./ > 0),
8
In view of the optimality conditions of Proposition 6.4, the asset pricing relations presented in this
section can also be obtained in an incomplete market with respect to an individual agent’s utility
function evaluated in correspondence of the individual agent’s optimal consumption plan.
relation (6.70) can be further rewritten as

0
u .e / n ˇˇ
1 Cov ı u0 .etC1/
; rtC1 Ft
h 0 i
t
EŒrtC1
n
jFt D ;
u .etC1 / ˇˇ
E ı u0 .et / Ft
for all n D 0; 1; : : : ; N and t D 0; 1; : : : ; T 1. In particular, for the risk free security
n D 0, this gives
1 ıEŒu0 .etC1 /jFt

D ; for all t D 0; 1; : : : ; T 1; (6.71)
rf u0 .et /
where we have used the fact that the aggregate endowment process e is adapted to
the information flow F, so that et is Ft -measurable. In particular, note that the right-
hand side of relation (6.71) determines the equilibrium risk free interest rate for the
time interval Œt; tC1, which equals the expected rate of intertemporal substitution of
consumption of the representative agent. In particular, the equilibrium risk free rate
will be constant if the representative agent is risk neutral or if there is no variation
over time in the aggregate endowment process. Furthermore, relation (6.71) implies
that the equilibrium risk free rate follows a predictable stochastic process, since
the risk free rate for the interval Œt; t C 1 is measurable with respect to Ft , for
all t D 0; 1; : : : ; T 1. See also Exercise 6.28 for a related representation of the
equilibrium prices of traded securities.
The following simple proposition gives the general asset pricing relation explain-
ing conditional expected excess returns in terms of the conditional covariance with
the representative agent’s intertemporal rate of substitution. This represents the
result at the basis of the Consumption Capital Asset Pricing Model (CCAPM).
Proposition 6.35 Under the assumptions of the present section, the following
holds, for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1:
0
u .etC1 / n ˇˇ
EŒrtC1
n
jFt rf D rf Cov ı 0 ; r ˇFt (6.72)
u .et / tC1
Cov.u0 .etC1 /; rtC1
n
jFt /
D : (6.73)
EŒu0 .etC1 /jFt
Proof Relation (6.72) immediately follows by (6.70) and (6.71). To prove rela-
tion (6.73), it suffices to notice that, in view of equation (6.71), it holds that
u0 .et /=.ırf / D EŒu0 .etC1 /jFt , for all t D 0; 1; : : : ; T 1. t
u
The interpretation of the above result is analogous to that of relation (5.3)
obtained in a two-period economy. Indeed, relation (6.72) shows that the conditional
expected excess return of a traded security is determined by the conditional covari-
ance of its return with the representative agent’s intertemporal rate of substitution
in correspondence of the aggregate endowment. The risk premium of a security will
be positive if and only if the return is negatively correlated with the representative
agent’s intertemporal rate of substitution. As a consequence, all traded securities
will exhibit a null risk premium (meaning that EŒrtn D rf , for all n D 1; : : : ; N and
t D 1; : : : ; T) if the representative agent is risk neutral or if there is no variation
over time in the aggregate endowment process or if the aggregate endowment
follows a predictable process (so that etC1 is Ft -measurable, for every t 2 N, and
the conditional covariance appearing in (6.72)–(6.73) becomes null). As explained
in Sect. 5.1, relation (6.72) can be interpreted in terms of the diversification
and insurance principles and shows that only systematic risk will be priced in
equilibrium. However, unlike in Sect. 5.1, in general we cannot replace aggregate
endowment/consumption with aggregate wealth for t D 0; 1; : : : ; T 1. Similarly,
we cannot guarantee the existence of a portfolio (market portfolio) whose return is
perfectly correlated with the aggregate endowment/consumption.
If the aggregate endowment and the asset returns are jointly distributed according
to a multivariate normal distribution, at every date t D 1; : : : ; T, then we can
make relation (6.73) more explicit by relying on Stein’s lemma. In this case, the
conditional risk premium takes the form
EŒu00 .etC1 /jFt

EŒrtC1
n
jFt rf D 0
Cov etC1 ; rtC1
n
jFt ;
EŒu .etC1 /jFt
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, thus showing that the conditional risk

premium of a security is proportional to the conditional covariance of its return with
the aggregate endowment.
As shown in Proposition 6.32 (applied with respect to the representative agent’s
utility function u), the quantity
ıu0 .et /
mt WD ; for t D 1; : : : ; T;
u0 .et1 /
can be interpreted as the one-period stochastic discount factor, in the sense that, in
view of (6.68), it holds that

E mtC1 rtC1
n
jFt D 1; for all n D 0; : : : ; N and t D 0; 1; : : : ; T 1:
In view of (6.65), the sequence .mt /tD1;:::;T of one-period stochastic

Q discount factors
generates the stochastic discount factor .Mt /tD0;1;:::;T via Mt WD tsD1 ms , with M0 D
1. Letting zntC1 WD rtC1
n
rf be the excess return of security n, for n D 1; : : : ; N at
date t C 1, relation (6.72) can be equivalently rewritten in terms of mtC1 as
p
EŒzntC1 jFt Cov mtC1 ; zntC1 jFt Var.mtC1 jFt /
p D p p ;
Var.ztC1 jFt /
n
Var.ztC1 jFt / Var.mtC1 jFt / EŒmtC1 jFt
n
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, where we have used the fact that

EŒmtC1 jFt D 1=rf , due to (6.71). The last relation has the same interpretation
of (5.4) and immediately leads to the following result, which extends to the present
multi-period setting the Hansen-Jagannathan bounds obtained in Proposition 5.1 for
a two-period economy.
Proposition 6.36 Under the assumptions of the present section, letting mt WD
ıu0 .et /
u0 .et1 / , for all t D 1; : : : ; T 1, the following holds:
ˇ ˇ
ˇ n ˇ p
ˇEŒrtC1 jFt rf ˇ Var.mtC1 jFt /
q ; for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1:
Var.rtC1 jFt /
n EŒm tC1 jFt
As explained in Sect. 5.1, Proposition 6.36 shows that in equilibrium the conditional
Sharpe ratio of each traded security at every date t is bounded from above by the
conditional volatility-expectation ratio of the one-period stochastic discount factor
mtC1 .
Similarly as in Sect. 5.1, since the asset pricing relations presented so far are
linear with respect to the asset returns, they also hold for any portfolio composed
of the N C 1 traded securities. More specifically, if .wt /tD1;:::;T is an RNC1 -
valued predictable stochastic process, with wt representing the proportions of
wealthP invested in the N C 1 traded assets at date t (compare with (6.18)), and if
rtw WD NnD0 wnt rtn denotes the return of the portfolio wt on the interval Œt 1; t, then

w
jFt /
EŒrtC1
w
jFt rf D : (6.74)
EŒu0 .etC1 /jFt
In the case of a multi-period economy, there does not exist an analogous to the
market portfolio appearing in relation (5.6). Nevertheless, we can still find a special
portfolio with respect to which an analogue of relation (5.9) can be established, as
shown in the following proposition.
Proposition 6.37 Under the assumptions of the present section, let .wO t /tD1;:::;T be
an RNC1 -valued predictable stochastic process such that, for all t D 0; 1; : : : ; T 1,
the quantity wO tC1 satisfies 1> wO tC1 D 1 and
> ˇ >
Corr rtC1 wO tC1 ; u0 .etC1 /ˇFt D min Corr rtC1 w; u0 .etC1 /jFt :
w2RNC1
w> 1D1
Then, for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, it holds that

ˇ
O ˇ
EŒrtC1
n
jFt rf D ˇne;t E rtC1
w
Ft rf ; (6.75)
O O O >
with ˇne;t WD Cov.rtC1
w
; rtC1
n
jFt /= Var.rtC1
w
jFt / and rtC1
w
WD rtC1 wO tC1 .
Proof As shown in Exercise 6.29, the portfolio process .wO t /tD1;:::;T satisfies

O
; rtC1 jFt ˇ
w n
Cov rtC1
O ˇ
Cov u0 .etC1 /; rtC1
n
jFt D Cov u0 .etC1 /; rtC1
w
Ft ;
O
w
Var rtC1 jFt
(6.76)
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1. Equations (6.73) and (6.74) (applied

O imply that
to the portfolio w)

n
jFt / wO
EŒrtC1
n
jFt rf D O
E rtC1 jFt rf ;
w
jFt /
so that, in view of equation (6.76),
wO
Cov.rtC1 ; rtC1
n
jFt / wO
EŒrtC1
n
jFt rf D O
E rtC1 jFt rf ;
w
Var.rtC1 jFt /
thus proving relation (6.75). t

u
The asset pricing relation (6.75) has the typical ˇ-form, with respect to a risk
factor represented by the return of the portfolio having minimal conditional corre-
lation with the representative agent’s marginal utility of aggregate consumption.
This makes more explicit the general CCAPM relations (6.72)–(6.73) (see also
Rubinstein [1487] and Breeden & Litzenberger [287]).
We have so far presented general asset pricing relations, without making specific
assumptions on the form of the utility function u or on the distribution of the asset
returns or of the aggregate endowment. In the remaining part of this section, we
derive more explicit asset pricing relations under specific assumptions on the utility
functions and/or on the distributions. In particular, we shall see that, under suitable
assumptions, the risk premium of an asset is determined by the covariance of the
asset return with the growth rate of the aggregate consumption process.
Let us suppose that the representative agent’s utility function has a quadratic form,
i.e., u.x/ D ax b2 x2 , with a; b > 0. Since the function u is increasing only on
the domain Œ0; a=b (compare with the discussion in Sect. 2.2), we assume that the
aggregate endowment process is such that et 2 Œ0; a=b, for all t D 0; 1; : : : ; T. In
this case, the asset pricing relations obtained in Proposition 6.35 take the form
b
EŒrtC1
n
jFt rf D Cov etC1 ; rtC1
n
jFt
a bEŒetC1 jFt
b
D Cov etC1 ; zntC1 jFt ;
a bEŒetC1 jFt
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1. This shows that, in the case of a

quadratic utility function, the conditional expected risk premium is determined by
the conditional covariance of the excess return with the aggregate endowment. Sim-
ilarly, one can establish a version of Proposition 6.37 with respect to the portfolio
process .wO t /tD1;:::;T having maximal correlation with the aggregate endowment, i.e.,
such that the quantity wO tC1 satisfies, for all t D 0; 1; : : : ; T 1,
> >
Corr rtC1 wO tC1 ; etC1 jFt D max Corr rtC1 w; etC1 jFt :
w2RNC1
w> 1D1
Power Utility
Assume now that the representative agent’s utility function is of the power form,
i.e., u.x/ D x = , with 2 .0; 1/. In this case, the asset pricing relations obtained
in Proposition 6.35 take the form
1 ˇ 1 ˇ
etC1 n ˇ etC1 n ˇ
Cov et
; rtC1 ˇFt Cov et
; ztC1 ˇFt
EŒrtC1
n
jFt rf D
1 ˇ D
1 ˇ ;
etC1 ˇ etC1 ˇ
E et ˇFt E et ˇFt
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1. This shows that, in the case of a power

utility function, the conditional risk premium is positive if and only if the excess
return is negatively correlated with the growth rate of the aggregate consumption
process. Moreover, the equilibrium risk free rate rf is determined by the conditional
expected growth rate of aggregate consumption. Indeed, due to equation (6.71), it
holds that
" #
1 etC1 1 ˇˇ
D ıE ˇFt : (6.77)
rf et
Log-Normal Distribution
Under the assumption that the joint conditional distribution of the asset returns
and the representative agent’s intertemporal rate of substitution is log-normal and
homoskedastic, a simple formula can been derived for the asset risk premia. More
specifically, we have the following proposition, the proof of which is given in
Exercise 6.30.
0
Proposition 6.38 Under the assumptions of the present section, let mt WD uıu 0 .e
.et /
t1 /
,
for all t D 1; : : : ; T, and suppose that .log mt ; log rtn / conditionally on Ft1 follows
a bivariate normal distribution with mean n 2 R2 and covariance ˙ n 2 R22 , for
all n D 1; : : : ; N and t D 1; : : : ; T. Then the following holds:
˙22
n
EŒlog rtC1
n
jFt log rf D ˙12
n
2

n
Var log rtC1 jFt
D Cov log mtC1 ; log rtC1
n
jFt ;
2
(6.78)
A similar result has been obtained in Hansen & Singleton [896] (see also
Singleton [1549]) by assuming that the representative agent’s utility function is
of the power form u.x/ D x = and that the aggregate consumption growth
rate and the return of each asset are jointly distributed according to a log-normal
distribution. Indeed, if .log.et =et1 /; log rtn / conditionally on Ft1 has a bivariate
normal distribution with mean n 2 R2 and covariance ˙ n 2 R22 , for all
n D 1; : : : ; N and t D 1; : : : ; T, then it holds that
˙22
n
EŒlog rtC1
n
jFt log rf D . 1/˙12 n
2

n
Var log rtC1 jFt
D
2

. 1/ Cov log.etC1 =et /; log rtC1
n
jFt ;
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1. This relation follows by the same

arguments used in Exercise 6.30 to prove (6.78). The equilibrium risk free rate rf is
determined by
. 1/2
log rf D log ı . 1/E log.etC1 =et /jFt Var log.etC1 =et /jFt :
2
In the case of a power utility function and a log-normal distribution, the above
formulae relate the conditional expected return to the conditional expectation and
variance of the aggregate consumption growth rate. In particular, the logarithmic
risk free return is linear in the expected logarithmic consumption growth rate,
with a slope equal to the coefficient of relative risk aversion of the representative
agent (1 ). In this sense, a large consumption growth rate leads to a high
risk free rate or to a low risk aversion coefficient (high elasticity of intertemporal
substitution) in order to induce agents to transfer money intertemporally smoothing
the consumption stream. The variance term in the equilibrium risk free rate is
due to a demand for precautionary saving. All the returns are decreasing in the
representative agent’s discount factor ı. The conditional risk premium of an asset
is linear in the conditional variance of the logarithmic return and in the conditional
covariance of the logarithmic return with the logarithmic consumption growth rate
with a slope equal to the coefficient of relative risk aversion of the representative
agent.
By relying on the Hansen-Jagannathan bound established in Proposition 6.36
and on the properties of the log-normal distribution, one can also obtain an upper
bound for the conditional Sharpe ratio of a traded security in terms of the conditional
variance of the aggregate consumption growth rate, similarly as in the case of a two-
period economy (see formula (5.12) in Sect. 5.1).
Multi-Factor Models
In view of Propositions 6.35 and 6.37, the CCAPM can be regarded as a one-factor
asset pricing model, where the risk premia of the traded securities are described in
terms of their beta coefficients with respect to the representative agent’s marginal
rate of intertemporal substitution of consumption (or, in view of Proposition 6.37,
with respect to the portfolio minimally correlated with the representative agent’s
rate of intertemporal substitution).
A multi-factor model has been proposed through the conditional Intertemporal
Capital Asset Pricing Model (ICAPM) (originally proposed in continuous time by
Merton [1330]; we base our presentation on Constantinides [492]). Suppose that
there exists an adapted stochastic factor process .yt /tD0;1;:::;T such that yt is RK -
valued, for all t D 0; 1; : : : ; T, for some K 2 N, and such that the aggregate
endowment et at date t can be written as an Ft1 -measurable function ft1 ./ W
RK ! RC of yt , i.e., et D ft1 .yt /, for all t D 1; : : : ; T. In other words, yt can
be thought of as a vector of economic variables which determine the state of the
economy. Furthermore, assume that the .K C N C 1/-dimensional vector .yt ; rt /
is distributed according to a multivariate normal distribution. Under the present
assumptions, relation (6.73) together with an application of Stein’s lemma implies
that
X
K
EŒu00 .et /fk;t .ytC1 /jFt
EŒrtC1
n
jFt rf D Cov.yktC1 ; rtC1
n
jFt /;
kD1
EŒu0 .etC1 /jFt
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, where fk;t ./ denotes the first derivative
of the function ft ./ with respect to its k-th argument. If the utility function is
logarithmic, then the conditional ICAPM reduces to the conditional CAPM (see
Constantinides [492, Section V] for more details).
A multi-factor asset pricing model can also be derived under the assumption
that the conditional one-period stochastic discount factor can be expressed as a
linear function of K factors (see also Campbell [335], Ferson & Jagannathan
[696], Ferson [688]). More specifically, suppose that there exists a K-dimensional
adapted stochastic process .yt /tD0;1;:::;T , for some K 2 N, such that
X
K
mt D a t bkt ykt ; for all t D 1; : : : ; T;
kD1
for some predictable real-valued and RK -valued stochastic processes .at /tD1;:::;T and
.bt /tD1;:::;T , respectively. Then formula (6.72) implies that
X
K
EŒrtC1
n
jFt rf D rf Cov mtC1 ; rtC1
n
jFt D rf bktC1 Cov.yktC1 ; rtC1
n
jFt /
kD1
X
K
Cov.yktC1 ; rtC1
n
jFt / X
K
D rf bktC1 Var.yktC1 jFt / DW rf kt ˇtnk ;
kD1
Var.yktC1 jFt / kD1
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1, with

Cov yktC1 ; rtC1
n
jFt
kt WD bktC1 Var.yktC1 jFt / and ˇtnk WD ;
Var.yktC1 jFt /
where we have used the predictability of the processes a and .b1 ; : : : ; bK /.

An intertemporal discrete time competitive equilibrium version of the Arbitrage
Pricing Theory (APT) has been obtained in Connor & Korajczyk [479] and
Bossaerts & Green [270] assuming a factor model for asset dividends. In these
models, the beta coefficients and the intertemporal risk premia vary over time.
Similarly as above, a multi-factor representation of risk premia can be obtained
assuming a linear factor model for dividends and normality of the relevant random
variables by expanding (6.72) be means of Stein’s lemma (see Constantinides
[492]). A non-linear no-arbitrage multi-factor model has been proposed in Bansal
& Viswanathan [138].
The Term Structure of Interest Rates
The equilibrium analysis and the asset pricing relations presented so far provide a
natural framework for studying the term structure of interest rates. Indeed, unlike
in the case of a simple two-period economy, in a multi-period economy the risk
free interest rate is allowed to vary through time (as long as it follows a predictable
stochastic process), depending on the equilibrium price of intertemporal consump-
tion. In particular, as we have already seen in relation (6.71), the equilibrium risk
free rate is related to the conditional expected growth of the aggregate consumption.
The study of the term structure of interest rates takes explicitly into account how
discount factors behave in correspondence of different time horizons.
For t 2 f0; 1; : : : ; T 1g and 2 f1; : : : ; T tg, let B.t; t C / denote the price at
date t of a zero-coupon risk free bond maturing at the future date t C . This means
that B.t; tC/ is the price at date t of a security delivering a riskless unitary dividend
at (and only at) the future date t C . Adopting the representative agent’s valuation,
the equilibrium pricing relation (6.69) implies that

u0 .etC / ˇˇ

B.t; t C / D E ı ˇFt (6.79)
u0 .et /
and the yield implicit in the bond price B.t; t C /, i.e., the rate i.t; t C / which
satisfies 1=B.t; t C / D .1 C i.t; t C // , can then be expressed as

0 1=
1 u .etC / ˇˇ
i.t; t C / D E ˇFt 1: (6.80)
ı u0 .et /
For every date t 2 f0; 1; : : : ; T 1g, the collection fi.t; t C /I D 1; : : : ; T tg

represents the (spot) term structure of interest rates at date t. The above relations
show that the term structure of interest rates encodes the conditional expectations
about the growth of the aggregate endowment over time, as measured through the
representative agent’s marginal utility.
By no arbitrage arguments, the interest rates i.t; t C / defined in (6.80) allow
to recover the forward interest rates. For any t 2 f0; 1; : : : ; T 1g and 1 ; 2 2
f0; 1; : : : ; T tg with 1 < 2 , the forward rate f .tI t C 1 ; t C 2 / is defined as the
interest rate set at date t which satisfies the relation
2
1 C i.t; t C 2 / D 1 C i.t; t C 1 / 1 1 C f .tI t C 1 ; t C 2 / 2 1 : (6.81)
In other words, f .tI tC1 ; tC2 / is the interest rate set at date t for lending/borrowing
over the future time interval Œt C 1 ; t C 2 which is coherent with the absence of
arbitrage opportunities if all zero-coupon risk free bonds are available in the market
(see also Exercise 6.33). As a matter of fact, the same amount of wealth is obtained
by investing over the horizon Œt; t C 2 (yielding the amount of wealth .1 C i.t; t C
2 //2 , corresponding to the left-hand side of (6.81)) or by investing over the horizon
Œt; t C1 and then reinvesting at date t C1 the wealth over the horizon Œt C1 ; t C2
(thus yielding the amount of wealth .1 C i.t; t C 1 //1 .1 C f .tI t C 1 ; t C 2 //2 1 ,
corresponding to the right-hand side of (6.81)).
Consider a fixed date t 2 f0; 1; : : : ; T 2g and suppose that at date t it is possible
to trade a zero-coupon risk free bond with maturity t C 1 having price B.t; t C 1/ and
a zero-coupon bond with maturity tC2 having price B.t; tC2/. Suppose furthermore
that at the subsequent date t C 1 it will be possible to trade a zero-coupon bond with
maturity t C 2 for the price B.t C 1; t C 2/ (which is measurable with respect to the
information at date t C 1). Let P be a risk neutral probability measure. Then, as
shown in Exercise 6.31, a no-arbitrage argument implies that (see also Pascucci &
Runggaldier [1404, Proposition 4.5])
B.t; t C 2/ D B.t; t C 1/E ŒB.t C 1; t C 2/jFt ; (6.82)
with E ŒjFt denoting the Ft -conditional expectation under P . However, this

relation does not hold in general under the physical/statistical probability measure
P. We say that the expectation hypothesis holds if relation (6.82) holds under the
physical/statistical probability measure P, i.e., if
B.t; t C / D B.t; t C 1/EŒB.t C 1; t C 2/ : : : B.t C 1; t C /jFt :
As follows from the next proposition, the expectation hypothesis is satisfied only
in rather special cases under the physical probability measure P.
Proposition 6.39 Under the assumptions of the present section, the following
hold:
(i) if the representative agent is risk neutral, then i.t; t C / D 1=ı 1 for every
t D 0; 1; : : : ; T 1 and 2 f1; : : : ; T tg;
(ii) if the representative agent is risk averse and the aggregate endowment process
is constant over time, then i.t; t C / D 1=ı 1 for all t D 0; 1; : : : ; T 1 and
every 2 f1; : : : ; T tg.
Moreover, for all t 2 f0; 1; : : : ; T 2g, it holds that
0
u .etC1 / u0 .etC2 / ˇˇ
B.t; t C 2/ D B.t; t C 1/EŒB.t C 1; t C 2/jFt C Cov ı 0 ;ı 0 ˇFt :
u .et / u .etC1 /
(6.83)
Proof The first two claims follow immediately from relation (6.80), while equa-
tion (6.83) can be obtained by using the tower property of the conditional expecta-
tion together with relation (6.79):

0
u0 .etC2 / ˇˇ u .etC1 / u0 .etC2 / ˇˇ
B.t; t C 2/ D E ı 2 0 ˇFt D E ı 0 ı 0 ˇFt
u .et / u .et / u .etC1 /

0 ˇ
0
u .etC1 / ˇ u .etC2 / ˇˇ
DE ı 0 ˇFt E ı 0 ˇFt
u .et / u .etC1 /
0
C Cov ı 0 ;ı 0 ˇFt
u .et / u .etC1 /
0
D B.t; t C 1/EŒB.t C 1; t C 2/jFt C Cov ı 0 ;ı 0 ˇFt :
u .et / u .etC1 /
t
u
The above proposition shows that in the special case where the representative
agent is risk neutral or when there is no variation in the aggregate consumption
process, then the term structure is flat. In the general case of a risk averse
representative agent, formula (6.83) shows that the price B.t; t C 2/ differs from the
price obtained under the expectation hypothesis due to the presence of an additional
covariance term which corresponds to B.t; t C 2/ B.t; t C 1/EŒB.t C 1; t C 2/jFt
and is called expected term premium. In turn, this implies that the expectation
hypothesis will be satisfied if, given the information available at the previous date,
the one-period stochastic discount factor (the intertemporal rate of substitution of
the representative agent) is not serially correlated. In other words, the expectation
hypothesis holds if the current growth rate of aggregate consumption has no
predictive power on the future growth rate of aggregate consumption. Under suitable
assumptions (see Exercise 6.32), we can also obtain an approximate linear relation
between the equilibrium yield i.t; t C / and the conditional expected logarithmic
growth rate of aggregate consumption.
A Markov Chain Model in Infinite Horizon
We close this section by briefly presenting some asset pricing relations in the context
of an economy with an infinite time horizon (i.e., T D 1) where the aggregate
endowment process follows a Markov chain (we refer to Ljungqvist & Sargent
[1231, Chapter 13] for full details). This model has been used as the basis for several
important studies, notably in the analysis of the equity premium puzzle in Mehra &
Prescott [1319].
Suppose that the aggregate endowment process .et /t2N evolves as a Markov chain
taking values in the K-dimensional state space fe.1/; : : : ; e.K/g, for some K 2 N,
so that

P etC1 D e.j/jet D e.i/ D ij ; for all i; j 2 f1; : : : ; Kg;
P
with KjD1 ij D 1, for all i D 1; : : : ; K. We denote by ˘ 2 RKK the associated
transition matrix composed of the elements fij I i; j D 1; : : : ; Kg. As above in the
present section, we adopt the representative agent’s perspective and assume that
his utility function is given by u. We also assume that the market only contains a
single traded security which, at every date t 2 N, pays the aggregate endowment as
dividend. We denote by s.k/t the price of such a security at date t if the aggregate
endowment at date t is in state k, for k 2 f1; : : : ; Kg. The pricing relation (6.69)
(here extended to an infinite time horizon) yields the following equilibrium price
for the aggregate consumption stream in correspondence of the event fet D e.k/g:
1
X
0 ˇ X1
u .etCs / ˇ 1
s.k/t D E ıs 0 etCs ˇFt D 0 ı s E u0 .etCs /etCs jet D e.k/ ;
sD1
u .et / u .e.k// sD1
where the second equality follows from the Markov property of the aggregate
endowment process .et /t2N . In turn, the right-hand side of the above relation
implies that the equilibrium price of the aggregate consumption stream does not
depend on time t but only on the current state of the aggregate endowment, so
that we can simply write s.k/ for the price (at any date t 2 N) of the security
in correspondence of state k of the aggregate endowment. Together with the asset
pricing relation (6.68), this implies that
XK
u0 .e. j//
s.k/ D ı 0
e. j/ C s. j/ kj ;
jD1
u .e.k//
for all k DP1; : : : ; K. Let us define %.k/ WD s.k/u0 .e.k//, for all k D 1; : : : ; K, and
!.k/ WD ı KjD1 u0 .e.j//e.j/kj , so that
X
K
%.k/ D !.k/ C ı kj %.j/; for all k D 1; : : : ; K:
jD1
In vector notation, this last equation can be rewritten as % D ! Cı˘%, which admits
the unique solution
% D .I ı˘ /1 !;
where I 2 RKK is the identity matrix (see Ljungqvist & Sargent [1231, Exam-
ple 13.7.2]).
Let us now assume that the representative agent’s utility function is of the power
type, i.e., u.x/ D x = and that the growth rate of the aggregate endowment (rather
than the aggregate endowment itself, as considered above) follows a finite state
Markov chain (under the standing assumption that the aggregate endowment is
always strictly positive). More precisely, assume that the quantity xt WD et =et1 ,
for t 2 N, satisfies

P xtC1 D .j/jxt D .k/ D kj ; for all j; k 2 f1; : : : ; Kg;
for all t 2 N, where f.1/; : : : ; .K/g represents the K-dimensional state space of
the Markov chain .xt /t2N . As above, in view of equation (6.69), the equilibrium
price st at date t of the traded security which delivers the aggregate endowment as
dividend satisfies
1 1 1
1 X
X
X
st D 1
ı s EŒetCs jFt D et ı s EŒxtCs jFt D et ı s EŒxtCs jxt ;
et sD1 sD1 sD1
where the last equality uses the Markov property of the process .xt /t2N . Defining the
process .
t /t2N by
t WD st =et , for all t 2 N, the above equation shows that
t does
only depend on the state of xt and not on all information contained in Ft at date t.
Hence, we can simply denote by
.k/ the value of
t if xt D .k/, for some k 2
f1; : : : ; Kg, for every t 2 N. In other words,
.k/ represents the price-dividend ratio
of the security in correspondence of state k of the aggregate endowment growth. By
relying on this observation and using equation (6.68) (dividing by et and rearranging
terms), this gives
X
K

.k/ D ı .j/ 1 C
.j/ kj ; for all k 2 f1; : : : ; Kg: (6.84)
jD1
This last equation has been used in Mehra & Prescott [1319] to compute equilibrium
prices. In this setting, at every date t 2 N and in state k, the price of a zero-coupon
bond which matures at the subsequent date t C 1 is determined by
X
K
B.t; t C 1/.k/ D ı .j/ 1 kj ; for all k D 1; : : : ; K:
jD1
6.5 Infinite Horizon Economies and Rational Bubbles
As we have seen in the previous section, in an economy with a finite time horizon
T < 1, the equilibrium prices of the traded securities are determined by the condi-
tional expectation of the future discounted dividends, weighted by the representative
6.5 Infinite Horizon Economies and Rational Bubbles 325
agent’s marginal rate of substitution in correspondence of the aggregate endowment.

In this section, we remove the assumption of a finite time horizon and allow for
infinitely lived economies. In this context, we will see that asset prices can exhibit
a bubble component, which represents a deviation from the conditional expectation
of future discounted dividends. This is an interesting possibility, taking into account
the numerous historical episodes where asset prices appeared to be incompatible
with plausible values of the underlying dividends/fundamentals (think for instance
of the Dutch tulip bubble at the beginning of the 16th century).
For simplicity, we consider an economy with two traded securities (the analysis
can be easily extended to economies with multiple securities): a risk free asset
paying the constant rate rf and a risky asset described by the couple of non-negative
price-dividend processes .s; d/. We assume that markets are open at every date
t 2 N (infinite time horizon). As in the previous section, we adopt the representative
agent’s paradigm and suppose that the representative agent’s utility function and
discount factor are given by u and ı 2 .0; 1/, respectively. In the present context,
the asset pricing relation (6.68) implies that

u0 .et /st D ıE u0 .etC1 /.stC1 C dtC1 /jFt ; for all t 2 N; (6.85)
where .et /t2N denotes the aggregate endowment/consumption process. Given the
processes .et /t2N and .dt /t2N , equation (6.85) does not fully determine the equilib-
rium price process .st /t2N because it is defined backward in time and involves a
conditional expectation (see also Pesaran [1414] on the solution of models under
rational expectations). Indeed, as shown in the following proposition, there does not
exist in general a unique solution .st /t2N to equation (6.85) in an infinitely lived
economy.
P
Proposition 6.40 Suppose that 1 0
sD1 ı EŒu .es /ds < 1. Then, under the assump-
s
tions of the present section, an F-adapted non-negative stochastic process .st /t2N
satisfies equation (6.85) if and only if st D st C ˇt , for all t 2 N, where
1
1 X s
st WD ı EŒu0 .etCs /dtCs jFt ; for all t 2 N; (6.86)
u0 .et / sD1
and .ˇt /t2N is an F-adapted stochastic process satisfying
u0 .et /ˇt D ıEŒu0 .etC1 /ˇtC1 jFt ; for all t 2 N:
Proof Let .st /t2N be a non-negative F-adapted process satisfying (6.85). Then, for
any t 2 N, the tower property of the conditional expectation implies that

u0 .et /st D ıE u0 .etC1 /.stC1 C dtC1 /jFt

D ıE u0 .etC1 /dtC1 C ıE u0 .etC2 /.stC2 C dtC2 /jFtC1 jFt
D ıEŒu0 .etC1 /dtC1 jFt C ı 2 EŒu0 .etC2 /dtC2 jFt C ı 2 EŒu0 .etC2 /stC2 jFt :
By iterating the same argument, we get that, for any t; T 2 N,
X
T
u0 .et /st D ı s EŒu0 .etCs /dtCs jFt C ı T EŒu0 .etCT /stCT jFt :
sD1
Taking the limit for T ! 1 in the last equation, we get

1
X
u0 .et /st D ı s EŒu0 .etCs /dtCs jFt C lim ı T EŒu0 .etCT /stCT jFt
T!1
sD1 (6.87)
DW u0 .et /st C u0 .et /ˇt ;
where we define ˇt WD limT!1 ı T EŒu0 .etCT /stCT jFP 0

t =u .et /, for all t 2 N. Note
1
that st is well defined due to the assumption that sD1 ı s EŒu0 .es /ds < 1. It

remains to show that u0 .et /ˇt D ıEŒu0 .etC1 /ˇtC1 jFt , for all t 2 N. This easily
follows by replacing st D st C ˇt into equation (6.85), so that
1
X
ı s EŒu0 .etCs /dtCs jFt C u0 .et /ˇt
sD1
" #
1
X 0 ˇˇ
D ıE ı E u .etC1Cs /dtC1Cs jFtC1 ˇFt C ıE u0 .etC1 /dtC1 jFt
s
sD1
C ıEŒu0 .etC1 /ˇtC1 jFt

1
X
D ı s EŒu0 .etCs /dtCs jFt C ıEŒu0 .etC1 /ˇtC1 jFt ;
sD1
where the first equality follows from the hypothesis that st C ˇt satisfies (6.85)
and where in the last equality we have used the tower property of the conditional
expectation. Conversely, suppose that .ˇt /t2N is an F-adapted stochastic process
such that u0 .et /ˇt D ıEŒu0 .etC1 /ˇtC1 jFt , for all t 2 N. Then it is immediate to
verify that the process .st /t2N defined by st WD st C ˇt , for all t 2 N, satisfies
equation (6.85). t
u
According to the above proposition, any price process .st /t2N which solves
equation (6.85) can be decomposed into two components: the fundamental value
.st /t2N , which is completely determined by the conditional expectation of the future
discounted dividends, and the bubble component .ˇt /t2N , which is an F-adapted
process such that .ı t u0 .et /ˇt /t2N is a martingale. Given a dividend stream, the
fundamental value of a security can be computed by relying on equation (6.86) (see
also Exercises 6.36–6.37 for some explicit computations of the fundamental value
of a security).
As we have seen above, provided that the martingale property of the process
.ı t u0 .et /ˇt /t2N is satisfied, then any adapted process .ˇt /t2N can represent a bubble.
In particular, we can have a deterministic bubble if ˇt D ˇ.t/, for some deterministic
function ˇ W N ! RC . For instance, a simple deterministic bubble can be defined as
ˇt D ˛ˇt1 , for all t 2 N, thus implying that ˇ.t/ D ˛ t ˇ0 . This represents a bubble
which will continue to expand forever and never bursts. More interesting bubbles
can be defined by letting .ˇt /t2N be a stochastic process. For instance, in Blanchard
& Watson [248] a speculative bubble has been introduced by letting
(
1 0
0
u .etC1 /ˇtC1 D ı u .et /ˇt C "tC1 ; with probability ;
"tC1 ; with probability1 ;
with 2 .0; 1/ and where ."t /t2N is a sequence of random variables such that
EŒ"tC1 jFt D 0, for all t 2 N. A stochastic process .ˇt /t2N of this form does not
depend on the fundamental value of a security and represents a bursting bubble,
which can grow to higher and higher values and then suddenly collapse (with
probability 1 ). When the bubble collapses, its expected value is zero and a new
bubble can start again. Another type of bubble process is represented by intrinsic
bubbles, which depend on the dividend process itself (see Exercise 6.39 and Froot
& Obstfeld [745]).
In general, a process .ˇt /t2N can be thought of as the bubble component because
it explodes in expectation as time goes to infinity. Indeed, by the martingale property
of the process .ı t u0 .et /ˇt /t2N and recalling that ı 2 .0; 1/,
(
1 if ˇt < 0I
lim EŒu0 .etCT /ˇtCT jFt D lim ı T u0 .et /ˇt D
T!1 T!1 C1 if ˇt > 0:
This exploding behavior illustrates why the process .ˇt /t2N represents the bubble
component of an asset price. It is important to remark that, as we have seen, asset
prices exhibiting a non-null bubble component are rational in the sense that they
are coherent with no-arbitrage and equilibrium, being solutions to the fundamental
equation (6.85).
Observe that, in view of equation (6.87), a price process .st /t2N satisfying
equation (6.85) coincides with the fundamental value if and only if the following
transversality condition holds:
lim ı T EŒu0 .etCT /stCT jFt D 0; for all t 2 N: (6.88)

T!1
Note also that, if the time horizon T is finite, then the standing assumption that prices
are null at the terminal date (i.e., sT D 0) implies that the transversality condition is
always satisfied and, hence, the bubble component is null. This shows that rational
bubbles can only exist in infinite horizon economies.
In equilibrium, the bubble component of any security in strictly positive supply

is always non-negative. Indeed, since u0 .et /ˇt D limT!1 ı T EŒu0 .etCT /stCT jFt , if
ˇt < 0, then there exists some future date T at which P.sT < 0/ > 0. However,
a negative price for an asset delivering a non-negative stream of dividends is not
consistent with market equilibrium, because the demand for such an asset would be
infinite at date T in correspondence of the event fsT < 0g. This shows that ˇt 0
for all t 2 N. Note, however, that this observation does not automatically imply that
the price of an asset characterized by a speculative bubble is always larger than the
corresponding fundamental value. This holds if the dividend stream and the discount
factor are not affected by the presence of a non-null bubble component. Instead,
if the presence of a bubble has an impact on the future dividend stream and on
the discount factor, then there are situations where the price st can be smaller than
the fundamental value st (see, e.g., Weil [1649]). For instance, in an overlapping
generations model, a positive bubble can induce an increase of the discount factor
and, consequently, a decrease of the fundamental value.
It is interesting to note that, under the present assumptions, a rational bubble
cannot restart once it bursts (see also Diba & Grossman [573, 575]). In other words,
a non-null rational bubble must be present from the initial date t D 0 of the economy
and cannot appear at some later date, meaning that ˇt D 0 implies that ˇtCs D 0 for
all s 2 N (see Exercise 6.38 for a proof). Moreover, as shown in Exercise 6.38, an
asset having a price process which is uniformly bounded from above (i.e., when it
holds that 0 st K, for all t 2 N for some constant K > 0; think for instance of
a European Put option written on a stock) cannot exhibit a bubble.
It is important to emphasize that the absence of arbitrage opportunities and the
existence of an equilibrium of the economy do not suffice to exclude the existence of
bubble solutions to the fundamental equation (6.85). However, suitable theoretical
considerations restrict the behavior of bubble solutions. In particular, as we have
seen above, bubbles are restricted to be non-negative, must be present from the
beginning and cannot restart and can only exist if the price process is unbounded
from above. Under appropriate assumptions, we can exclude the existence of non-
trivial bubble components. For instance, if we assume that the market (in an infinite
time horizon) is complete and that the economy is populated by infinitely lived
agents, then the absence of arbitrage opportunities implies that there cannot exist
a bubble solution. Indeed, assuming that the presence of a bubble does not affect
the fundamental value of the asset, suppose on the contrary that there exists a date
t 2 N such that P.ˇt > 0/ D P.st > st / > 0. Then, at date t on the event
fst > st g, an agent could realize an arbitrage opportunity by selling short the
security (thus receiving the amount st ) and investing in the portfolio of the Arrow
securities replicating the future dividend stream .ds /sDt;tC1;::: at the cost of st < st ,
thus realizing an arbitrage profit at date t. Note, however, that this arbitrage argument
relies on the assumption of infinitely lived agents. If agents are finitely lived, then
the absence of arbitrage opportunities does not suffice to exclude the presence of
speculative bubbles.
Since speculative bubbles are non-negative, the presence of a bubble component
implies that the asset price st will diverge as time goes to infinity (indeed, the
transversality condition (6.88) cannot be satisfied by the price process .st /t2N if
the bubble component is non-null). As shown in Brock [305], in a complete market
economy with an infinitely lived representative agent, speculative bubbles cannot
exist since the transversality condition (6.88) necessarily holds in correspondence
of the representative agent’s optimum. In the seminal paper Tirole [1594], it has been
shown that speculative bubbles cannot exist in correspondence of a dynamic rational
expectations equilibrium in an economy populated by a finite number of infinitely
lived risk neutral agents for an asset delivering an exogenously specified dividend
stream. This non-existence result is robust with respect to the introduction of short-
sale restrictions and heterogeneous information among the agents. However, asset
prices can exhibit a bubble if agents have a myopic behavior (in the sense that, at
each date t 2 N, they base their portfolio choice only on the comparison between
the price at date t and the distribution of the price at the next date t C 1).
The introduction of short-sales and borrowing constraints has important conse-
quences for the (non-)existence of speculative bubbles. For instance, Montrucchio
& Privileggi [1351] have shown that, in a general economy (without restrictions on
the stochastic process describing the dividends) populated by infinitely lived homo-
geneous agents with a zero short-sales constraint and a uniformly bounded relative
risk aversion coefficient, speculative bubbles cannot occur regardless of the specific
form of the agents’ utility functions. Always assuming a zero short-sales constraint,
the non-existence of speculative bubbles for unbounded utility functions (including
logarithmic and power utilities) has been established in Kamihigashi [1062], with
conditions depending only on the asymptotic behavior of the marginal utility at zero
and infinity. In Santos & Woodford [1496], the authors have established that typical
examples of speculative bubbles found in the literature represent rather exceptional
cases. More specifically, they establish that in an economy populated by a finite
number of infinitely lived agents with potentially incomplete markets, incomplete
participation (thus allowing for overlapping generation models) and (price/time
dependent) borrowing constraints, speculative bubbles cannot exist for an asset in
positive net supply, under the assumption that the present value of the aggregate
endowment is finite. However, bubbles can exist for an asset having infinite maturity
(or having finite maturity but in zero net supply; on this point see also Magill &
Quinzii [1285] and Huang & Werner [972, 973]).
Concerning the possible existence of speculative bubbles, Kocherlakota [1111]
has shown that, in the context of a deterministic economy with a finite number
of infinitely lived agents, the presence of constraints on debt accumulation makes
bubbles possible in equilibrium. Such constraints prevent agents from engaging in
Ponzi schemes9 and ensure the existence of an equilibrium. At the same time, the
same constraints prevent agents from transforming the presence of a bubble into an
arbitrage opportunity and, hence, in some cases bubbles can exist in equilibrium. In
9
In an economy with an infinite time horizon, a Ponzi scheme represents a strategy where the
debt is rolled over indefinitely and never repaid (see, e.g., Lengwiler [1182, Section 6.2.2]), taking
advantage of the infinite horizon of the economy.
particular, if agents face a wealth constraint, then an asset can exhibit a bubble if
and only if it is in zero net supply. Conditions are also obtained for the existence of
a bubble in the presence of exogenous short-sales constraints.
In Tirole [1595], the author considers a deterministic overlapping generations
economy populated by finitely lived agents and provides necessary and sufficient
conditions for the existence of a speculative bubble in this setting. It is shown that
a bubble can exist if the growth rate of the economy is larger than the long run
equilibrium bubble-free interest rate (dynamically inefficient economy). However,
the equilibrium is asymptotically bubble-free (see Tirole [1595, Proposition 1]). On
the other hand, if the interest rate exceeds the growth rate of the economy, then no
bubble can exist in equilibrium. A similar result was obtained in Weil [1647] in the
context of a stochastic economy. In an overlapping generation economy with agents
living more than two periods, we can observe a non-divergent bubble similar to that
proposed in Blanchard & Watson [248] (see Leach [1171]).
More generally, it is difficult to prove the non-existence of speculative bubbles
in the case of assets for which the fundamentals are not clearly defined. In an
asymmetric information environment with common priors, a condition for a bubble
to exist is that the bubble is not common knowledge among agents (see Allen et al.
[49] and Conlon [474]).
Summing up, one can affirm that there are many theoretical arguments against
the presence of speculative bubbles in asset prices. We also want to point out that
the hypothesis of common priors among the agents is crucial for the above results to
hold. In Chap. 8 we shall discuss the existence of speculative bubbles in an economy
with asymmetrically informed agents.
In Sect. 6.1, we have analysed the optimal investment-consumption problem

by means of the dynamic programming approach and we refer the reader to
Pascucci & Runggaldier [1404, Section 2.3] for an illustration of the dynamic
programming approach in the context of several examples of discrete time models.
The continuous time versions of the investment problems considered in Sect. 6.1
have been first considered in the seminal papers Merton [1327, 1328]. As we have
discussed after Corollary 6.28, an alternative approach to the solution of optimal
investment-consumption problems is represented by the martingale approach,
which is especially simple in the case of a complete markets. According to the
martingale approach, the solution of the optimal investment-consumption problem
is decomposed into two successive steps: first, the optimization problem is solved
over an abstract space of random variables and, as a second step, the optimal strategy
is determined as the hedging strategy for the optimal random variable obtained in
the first step. This approach was first developed for complete markets in Cox &
Huang [504, 505], Pliska [1421] and then extended to an incomplete market setting
by He & Pearson [919]. A detailed presentation of the martingale approach in the
context of general continuous time models based on Itô-processes can be found in

Karatzas & Shreve [1073] and further generalizations to the semimartingale setting
have been developed in Kramkov & Schachermayer [1128, 1129]. In the context
of discrete time models, the martingale approach is also discussed in Pascucci &
Runggaldier [1404, Chapter 2].
On the basis of the theoretical implications of the solutions to optimal portfolio
problems, during the last three decades the financial literature has paid significant
attention to the comparison of the theoretical results with the empirical findings.
In particular, there have been empirical studies on household portfolio choices (see
Guiso et al. [857], Campbell & Viceira [356], Campbell [337]) reporting evidence
of an asset allocation puzzle. Canner et al. [360], analysing stocks, bonds and cash
allocations recommended by four financial advisors in the U.S., find that young
people (i.e., investors with a long horizon) as well as less risk averse investors
are advised to invest in risky assets (high stocks-bonds ratio) more than old or
strongly risk averse investors. In Jagannathan & Kocherlakota [1009], the opposite
phenomenon has been reported for investment in bonds (horizon effect). Moreover,
it is shown that such asset allocation recommendations are inconsistent with the
two mutual funds separation theorem and with the optimal trading-consumption
strategy obtained under the assumption of identically and independently distributed
asset returns and of a power or logarithmic utility function.
The empirical analysis on portfolio holdings provides mixed results on the
relationship between age and investment in stocks: in many cases an increasing or
a hump-shaped pattern in age is observed (see Ameriks & Zeldes [55], Bertaut &
Starr-McCluer [210], Guiso et al. [860], Poterba & Samwick [1429]). In particular,
as reported in Ameriks & Zeldes [55], the fraction of investment in risky assets
reaches a peak in the interval 50–59 years. Moreover, even though there is a hump-
shaped pattern in the ownership of risky assets with respect to age, the risky asset
holding is almost constant with respect to age conditionally on shares ownership.
It is also observed that rich people save more and hold more risky assets with
respect to what the theory would predict (see Carroll [371], Dynan et al. [614]). It
is empirically observed that the ownership of risky assets and the participation rate
in the financial market tend to be increasing with respect to wealth, contradicting
the optimal portfolios suggested in the seminal papers Samuelson [1493], Merton
[1327]. Moreover, Heaton & Lucas [929] report that households with high and
variable business income invest less wealth in stocks than other similarly wealthy
households, thus showing that entrepreneurial income risk has a significant impact
on portfolio choices.
In the empirical literature analysing households’ portfolios, limited diversifica-
tion and a high degree of heterogeneity have been reported. Moreover, there exists a
participation puzzle, meaning that a rather low participation rate in the stock market
is detected. Mankiw & Zeldes [1298] have shown that at the time of their study
only one quarter of the U.S. families owned stocks. Since that time, the fraction of
population trading in the market has increased, but still nowadays a large part of
the population does not own stocks (about one half of the population in the U.S.,
according to Bertaut & Starr-McCluer [210], Ameriks & Zeldes [55], Haliassos &
Bertaut [876]). In other countries, the fraction of population active in the financial
market is even lower. Moreover, this figure does not increase significantly by
including in the analysis also investors holding stocks through pension funds.
Typically, active participation to the stock market occurs rather late in the life span
of an agent. Grinblatt et al. [832] show that stock market participation is positively
related to education and cognitive ability. It is difficult to explain the participation
puzzle in the context of the models presented in this chapter, based on classical asset
pricing theory.
Further puzzles have emerged when testing the implications of the theory in the
context of international asset pricing models (i.e., models with several financial
markets corresponding to different countries). In particular, a home bias puzzle
has been detected both in stock holdings as well as in consumption: on the one
hand, the proportion of foreign assets held by investors is too limited (see French
& Poterba [740]) and, on the other hand, domestic investors do not share risk
with foreign investors, with the consequence that consumption growth rates do
not move together across different countries as much as international risk sharing
would suggest (see Lewis [1216]). Huberman [981] reports a geographical bias,
showing that shareholders of a Regional Bell Operating Company tend to live in
the area which it serves. In a related direction, Benartzi [190] shows that agents
tend to invest significantly in their employer’s stock and this behavior is not due
to the presence of inside information. Coval & Moskowitz [501, 502] show that
mutual funds and stockholders invest locally, thus providing further evidence on the
existence of a home bias. Note that, as shown in DeMarzo et al. [553], a general
equilibrium model where agents care about their consumption relative to the per
capita consumption in their community can explain this portfolio bias, since agents
bias their portfolios towards the (undiversified) portfolios held by the other members
of the community (see also Sect. 9.2 for a discussion of similar models). Moskowitz
& Vissing-Jorgensen [1356] show that the investment in private equity is extremely
concentrated: even though private equity returns are generally not higher than the
market return on publicly traded stocks, there is a substantial investment in a single
privately held firm.
From a theoretical point of view, several papers have studied the impact on
optimal portfolio choices of specific features of the investment opportunity set. For
instance, optimal portfolio problems in the presence of a stochastic risk free interest
rate have been considered in Brennan & Xia [300, 301], Wachter [1634], Campbell
& Viceira [355]. In a continuous time model with a constant risk free rate, Kim
& Omberg [1099] analyse the optimal consumption-investment problem in the
presence of a single risky asset whose expected return follows a mean reverting
process. Under the assumption of a HARA utility function defined with respect to
consumption at the terminal date, a closed form solution to the optimal portfolio
problem is obtained. Moreover, it is shown that, if risk premia innovations are
negatively correlated with asset price movements and the risk premium is positive,
then the hedging demand is positive if the agent is more risk averse than in the case
of a logarithmic utility function. This implies that, in the presence of mean reversion,
we expect an agent with a long horizon to hold a portfolio larger than an agent
with a short horizon. The optimal portfolio is increasing in the investment horizon
because mean reversion in stock returns reduces the variance of cumulative returns
over long horizons. Other studies analysing the implications of mean reversion on
consumption and optimal portfolios include Barberis [146], Campbell & Viceira
[354], Campbell et al. [340, 342], Wachter [1636], Brennan [290], Brennan et al.
[296], Brandt [279], Lynch & Balduzzi [1259], Jurek & Viceira [1055]. Analysing
the optimal investment-consumption problem in the presence of a stochastic
volatility, Liu [1226] and Chacko & Viceira [384] obtain results similar to those
obtained under mean reversion: if volatility innovations are negatively correlated
with asset price movements, the risk premium is positive and the agent is more risk
averse than in the logarithmic case, then the hedging demand is positive and the
risky asset demand is increasing with respect to the investment horizon.
When the investment horizon is long, long term bonds with a stochastic interest
rate present similar characteristics to a non-risky investment. As a consequence, the
investment in long term bonds should be increasing with respect to an investor’s
risk aversion. A similar effect is obtained also in the presence of stochastic
inflation. Assuming a coefficient of relative risk aversion greater than one, these
effects contribute to explain the asset allocation puzzle (see Campbell & Viceira
[355], Campbell et al. [340], Wachter [1634], Brennan & Xia [300, 301]). It is
optimal for conservative long term investors to buy (inflation indexed) long term
bonds and agents should invest significantly in risky assets when they are young,
reducing progressively their exposure when their age increases (see Campbell et al.
[339], Bajeux-Besnainou et al. [182, 181]).
The possible predictability of asset returns also has important implications on
portfolio choices (see Campbell & Viceira [354], Balduzzi & Lynch [118], Han
[883], Wachter & Warusawitharana [1638], Handa & Tiwari [887], De Miguel
et al. [557]). In particular, Lynch [1258] has analysed portfolio choices based on
characteristics like size and the book-to-market ratio, when returns can be predicted
via the dividend yield. Compared with the investors’ allocation in their last period,
the presence of predictability leads the investor early in life to tilt his portfolio away
from high book-to-market stocks and from small stocks. Portfolio returns exhibit
both the size and the book-to-market effect. Moreover, it is shown that the utility
costs due to ignoring the predictability of returns are relevant.
If asset returns are distributed independently over time, the utility function
is of the logarithmic, exponential or power type and the investors are infinitely
lived, then it can be shown that consumption and portfolio weights are linear
with respect to wealth (see Hakansson [875]). A portfolio turnpike result for a
finite horizon economy with an agent maximizing expected utility of final wealth
is proved in Huberman & Ross [985]: the optimal policy converges to a time-
independent one as the investment horizon grows to infinity if the coefficient of
relative risk aversion converges as wealth increases. We also want to mention that
investment-consumption problems have been also extended by including housing
services (see Campbell & Cocco [341], Cocco [455], Yao & Zhang [1664]), illiquid
assets (see Ang et al. [70], Longstaff [1244], Tepla [1585]), luxury goods (see
Carroll [371], Wachter & Yogo [1639], Ait-Sahalia et al. [32]), health risk and
portfolio-saving decisions of the elders (see Palumbo [1399], Edwards [629], French
[737], French & Jones [738], De Nardi et al. [558], Yogo [1667]), annuities (see
Milevsky & Young [1340], Chai et al. [386], Horneff et al. [966]) and durable goods
(see Grossman & Laroque [844]). A different strand of literature has considered
the presence of non-diversifiable labor income and we refer to Sect. 9.5 for a
discussion on this point. Note that non-diversifiable labor income cannot be spanned
by the securities available for trading in the market and, therefore, leads to market
incompleteness.
In the context of a pure exchange economy, classical asset pricing theory has been
mainly developed in the seminal contributions of LeRoy [1183], Rubinstein [1487],
Lucas [1252], Harrison & Kreps [904]. Considering also a production economy,
equilibrium models have beed developed in Brock [305], Cox et al. [506], Cochrane
[460, 462], Rouwenhorst [1477], Jermann [1032]. In this case, instead of explaining
asset returns through marginal rates of substitution over intertemporal consumption,
asset returns are explained through marginal rates of transformation and, therefore,
through the firm’s investment demand and investment returns. In particular, these
models allow to relate asset returns to economic fluctuations and can explain the
variations in expected returns both cross-sectionally and over time. Empirically,
models based on a production economy are not rejected and have a performance
comparable to that of the CAPM and of the APT in the version proposed in Chen
et al. [424] and outperform the CCAPM.
In a dynamically complete market, the existence of an equilibrium is obtained
under standard regularity conditions. In an incomplete market setting, the existence
of an equilibrium has been studied in Duffie & Shafer [598, 599] (see also Magill
& Shafer [1288]). Duffie [592] has considered the existence of an equilibrium in
an economy comprising nominal assets. In the presence of an infinite horizon,
establishing the existence of an equilibrium is more challenging, as explained in
Duffie [593]. In that setting, Ponzi schemes have to be avoided (see also Hernandez
& Santos [939], Araujo et al. [75]). To this end, borrowing or short sale constraints
are introduced in order to ensure a suitable transversality condition. See also Huang
& Werner [973] on the implementation of an Arrow-Debreu equilibrium in an
infinite horizon economy.
In an economy with more than one good and real dividends, a generic result
has been obtained in Magill & Shafer [1287]. More specifically, they consider
equilibrium allocations corresponding to contingent markets (i.e., markets open at
the initial date for every good in correspondence of every time-event couple) and
a sequential system of spot and real asset markets, where each real asset delivers a
vector of goods (as considered in Sect. 4.2). They provide a necessary and sufficient
condition on the asset structure for the two equilibrium allocations to coincide in
a generic sense (i.e., on an open set of full measure). In particular, this condition
requires that, for every t D 0; 1; : : : ; T 1 and At 2 Ft , the number of traded assets
equals t .At /, using the notation introduced in Sect. 6.2.
As we have seen in Sect. 6.2, the notion of market completeness plays an

important role in the equilibrium analysis of multi-period economies. Guesnerie
& Jaffray [853] have shown that it is possible to complete the market through short-
lived securities. To this effect, it suffices to require that, for all events At 2 Ft , for
t D 0; 1; : : : ; T 1, and AtC1 At with AtC1 2 FtC1 , there exists a security
which is traded at date t in correspondence of the event At and pays a unitary
dividend at the subsequent date t C 1 in correspondence of the event AtC1 (and zero
otherwise). As in a two-period economy, contingent claims may help to complete
the market also in a multi-period setting. However, unlike in a two-period economy
(see Sect. 4.3), in a multi-period discrete time setting European Call options are
not always useful in order to complete the market (see Bajeux-Besnainou & Rochet
[183], but note that this result does not hold in continuous time). However, exotic
path-dependent options can allow to dynamically complete the market. The result
on dynamic market completeness in the presence of a sufficient number of traded
securities (see Proposition 6.15) has been extended in Duffie & Huang [596] to
a continuous time economy where the price processes are described by diffusion
processes. Loosely speaking, in this case the infinite dimensional state space is
matched by the possibility of infinitely frequent trading.
In the case of incomplete markets, an arbitrage free price is not univocally
determined. In this case, several different approaches have been proposed in the
literature in order to determine suitable arbitrage free prices (see Musiela &
Rutkowski [1364] for an overview). Recall that, if markets are incomplete, then
there exist contingent claims which cannot be perfectly replicated by trading
according to some self-financing strategy. Among the different approaches proposed
in the context of incomplete markets, several valuation techniques are based on the
idea of minimizing the risk of the unhedgeable part of a contingent claim, according
to some utility function or to a mean-variance criterion. Alternatively, one can insists
on the self-financing requirement, minimizing the quadratic cost of the replication
strategy (risk minimization, see Föllmer & Schweizer [722], Föllmer & Sondermann
[723] and the surveys by Pham [1418] and Schweizer [1508]). Another important
approach to determine an arbitrage free price in the case of incomplete markets is
represented by the utility indifference pricing, see Henderson & Hobson [936] for
an overview.
Similarly as in the case of Theorem 4.20 established in the context of a two-
period economy, one can prove a version of Theorem 6.23 by only assuming
the absence of arbitrage opportunities of the second kind. In that case, the same
arguments used in the proof of Theorem 6.23 allow to show the existence of a
probability measure P satisfying the martingale condition (6.52) but not necessarily
equivalent to P, in the sense that it may happen that some events have null
probability under the measure P . To this effect, see also LeRoy & Werner [1191,
Theorem 24.2.2].
We also mention that the CCAPM has been extended to a continuous time setting
in Breeden [285], Duffie & Zame [600] by assuming that asset returns follow
diffusion processes and a continuous time version of the CAPM has been developed
in Chamberlain [396].
We want to point out that, as argued in Shleifer & Summers [1540], the no-
arbitrage analysis has some intrinsic limits. For instance, it is typically assumed that
the agents’ investment horizon is equal to the horizon of the economy. However, if
the agents’ horizon is shorter than that of the economy, then a buy-and-hold strategy
may not work and an agent may not be able to exploit an arbitrage opportunity. In
this case, an agent will not only consider the underlying dividend stream paid by a
security, but also its future price when closing a position. In this setting, an arbitrage
opportunity may persist in the market. Dow & Gorton [584] show that a limited
horizon may induce an agent with private information not to trade on an arbitrage
opportunity. The informed trader trades if there is a chain of informed agents trading
in the future spanning the event date, i.e., at the end of his horizon there is an
informed trader who makes the price move in the direction of the information. This
effect induces an agent not to trade until the event date is close.
In Sect. 6.5, we have considered economies with an infinite horizon where
asset prices may exhibit a non-null bubble component. Assuming a finite number
of agents in the economy, a bubble can originate by a payoff stream that occurs
at infinity (lack of countable additivity in asset valuation; see LeRoy & Gilles
[1186], Gilles & LeRoy [781, 783, 784]). In Timmermann [1591], it is shown that
there is no bubble when there is a feedback of prices on dividends. The presence
of such a feedback effect is motivated by studies such as Lintner [1220], Marsh
& Merton [1306], showing that managers typically define dividends as proportions
of the fundamental value of the company which is reflected by the market price
of the company. A model with feedback effects admits multiple rational expecta-
tions equilibria and no speculative bubble. The existence of a bubble solution in
continuous time models has been addressed in Loewenstein & Willard [1242]. For
positive net supply assets, there are no bubbles under weak conditions. Bubbles
may exist for zero net supply assets. Heston et al. [940] study the existence of
a bubble in a market containing derivative assets. Intrinsic bubbles (fundamental
dependent bubbles) have been characterized in continuous time in Ikeda & Shibata
[997]. Bewley [219] suggests that the value of fiat money can be interpreted as a
bubble. On the interpretation of the dot-com bubble in terms of rational bubbles
see Ofek & Richardson [1382], LeRoy [1185], Pástor & Veronesi [1409], Griffin
et al. [828]. On the existence of speculative bubbles under endogenous borrowing
constraints see also Kocherlakota [1114], Hellwig & Lorenzoni [930], Werner
[1654]. On the existence of bubbles with heterogeneous opinions and short-sale
constraints see also Miller [1342], Varian [1611], Harris & Raviv [901], Kandel
& Pearson [1066], Harrison & Kreps [903], Chen et al. [418], Hong & Stein
[962], Scheinkman & Xiong [1502], Abreu & Brunnermeier [10], Kyle & Wang
[1150]. In mathematical finance, financial bubbles are often modeled by strict local
martingales (i.e., local martingales which are not true martingales) and we refer to
Protter [1435] for an overview of this approach.
6.7 Exercises 337
6.7 Exercises
Exercise 6.1 Let .; c/ be a self-financing trading-consumption strategy and let

W./ be the corresponding wealth process. Show that condition (6.5) holds.
Exercise 6.2 Suppose that the economy contains a risk free security paying a
constant and deterministic risk free rate rf > 0. Define the discounted quantities
SN t WD St =rft and D
N t WD Dt =rft , for all t D 0; 1; : : : ; T. For any trading-consumption
strategy .; c/, define the discounted portfolio value W N t ./ WD Wt ./=rft and the
discounted amount of consumption cN t WD ct =rf , for all t D 0; 1; : : : ; T. Show that
t
the self-financing condition (6.4) is stable with respect to discounting, in the sense
that, for any strategy .; c/ 2 A .x0 /:
N tC1 ./ D tC1

W > > N
SN tC1 C tC1 DtC1 cN t :
In particular, it holds that
X
T X
T1
N T ./ D 0> .SN 0 C D
W N 0/ C t> . SN t C D
N t/ cN t
tD1 tD0
X
T X
T1
DxC t> . SN t C D
N t/ cN t :
tD1 tD0
Exercise 6.3 Let the value function be defined as in (6.9) with respect to a utility
function u W RC ! R strictly increasing and concave. Prove that the value function
is strictly increasing and concave in its first argument (wealth).
Exercise 6.5 Prove Corollary 6.7.
Exercise 6.7 Prove Corollary 6.9.
Exercise 6.8 In the context of Proposition 6.8, suppose that, for all t D 1; : : : ; T 1,
the Ft -conditional distribution of the asset returns rtC1
n
, n D 1; : : : ; N, coincides
with the unconditional distribution of r1 , n D 1; : : : ; N. Prove that the optimal
n
portfolio process .wt /tD1;:::;T can be characterized as in equation (6.29), does not
depend on time and that the process .at /tD0;1;:::;T appearing in Proposition 6.8
reduces to a deterministic function of time.
Exercise 6.9 Let us consider a multi-period economy t D 0; 1; : : : ; T with two
assets: a risk free asset, whose price is given by Bt D rft , for all t D 0; 1; : : : ; T,
with rf > 0 being the risk free rate of return, and a risky asset with price process
.St /tD0;1;:::;T . We assume that, at any date t D 1; : : : ; T, given the price St1 at the
previous date t 1, the risky asset can take the two possible values St1 u and St1 d
with probabilities p and 1 p, respectively (this represents a special case of the

multi-period binomial model presented at the end of Sect. 6.3). Consider an optimal
investment-consumption problem of the type (6.6) with consumption only at the
terminal date T, i.e., the agent is interested in maximizing EŒu.WT / over all self-
financing trading strategies .t /tD1;:::;T . Prove the following claims:
p
(i) if u.x/ D x, then the value function introduced in (6.9) is given by
p
V.x; t/ D cTt x; for all t D 0; 1; : : : ; T and x > 0;
for a constant c > 0, and the optimal number of units of the risky asset is given
by
Wt1
t D K ; for all t D 1; : : : ; T;
St1
for a constant K 2 R which can be explicitly computed.

(ii) If u.x/ D log.x/, then the value function introduced in (6.9) is given by
V.x; t/ D c.T t/ C log.x/; for all t D 0; 1; : : : ; T and x > 0;
by
Wt1
t D K ; for all t D 1; : : : ; T;
St1

(iii) If u.x/ D x = , with 2 .0; 1/, then the value function introduced in (6.9) is
given by
x
V.x; t/ D cTt ; for all t D 0; 1; : : : ; T and x > 0;

by
Wt1
t D K ; for all t D 1; : : : ; T;
St1

Exercise 6.10 In the context of Proposition 6.12, let t 2 f0; 1; : : : ; T 1g and
At 2 Ft and suppose that, for some s 2 ft C 1; : : : ; Tg, there exists an event As 2 Fs
such that As At and qAs jAt ¤ qAs =qAt . Prove that such a price system admits
arbitrage opportunities and, hence, cannot correspond to an equilibrium.
6.7 Exercises 339
Exercise 6.11 Let fS ; DI . 1 ; c1 /; : : : ; . I ; cI /g be a Radner equilibrium, in

the sense of Definition
P 6.13. Suppose that the total supply of the N securities is
normalized to IiD1 0i;n D 1, for all n D 1; : : : ; N, and that all the agents have
strictly
PIincreasing
Putility functions ui W RC ! R. Show that in equilibrium it holds
N
that iD1 ct D nD1 dt , for all t D 0; 1; : : : ; T.
i n

Exercise 6.13 As in the first part of Sect. 6.2, consider an economy with I agents
(with homogeneous beliefs, the same discount factor ı, time additive and state
independent preferences represented by strictly increasing and strictly P concave
differentiable utility functions ui , i D 1; : : : ; I) and where all the TtD1 t Arrow
securities are traded. Let e D fet .At /I At 2 Ft ; t D 0; 1; : : : ; Tg be the aggregate
endowment process and fci I i D 1; : : : ; Ig a feasible allocation corresponding to an
(Arrow-Debreu) equilibrium, in correspondence of the prices fq0 ; qAt I At 2 Ft ; t D
1; : : : ; Tg for the Arrow securities. Show that the allocation fci I i D 1; : : : ; Ig
defines the utility function of a representative agent and that the prices fq0 ; qAt I At 2
Ft ; t D 1; : : : ; Tg are supported by a no-trade equilibrium in the representative
agent economy.
Exercise 6.14 Consider a representative agent economy where the representative
agent’s utility function is given by u.x/ D log.x/ and the aggregate endowment
process is e D fe0 ; e.At /I At 2 Ft ; t D 1; : : : ; Tg. Show that, in correspondence of
a no-trade equilibrium in the representative agent economy, the value at the initial
date t D 0 of the aggregate endowment process e is given by e0 .1 ı TC1 /=.1 ı/
(normalizing q0 D 1, as at the end of Sect. 6.2).
Exercise 6.16 Prove Lemma 6.19.
Exercise 6.17 Suppose that the price-dividend couple .S; D/ does not admit arbi-
trage opportunities and let .; c/ 2 A .x0 /. Use Proposition 6.22 to give an
alternative and simple proof of Lemma 6.19.
Exercise 6.18 Prove Lemma 6.20.
Exercise 6.20 In the setting of Proposition 6.32, prove the representation (6.62) for
the conditional probabilities associated to the risk neutral probability measure P
defined by (6.61) (compare also with Huang & Litzenberger [971, Section 8.5]).
Exercise 6.21 Suppose that the S market is incomplete, so that there exist elements of
K C that are not contained in x2RC C0C .x/. Show that the arbitrage free price of an
S
attainable consumption process c 2 x2RC C0C .x/ is uniquely defined, regardless
of the specific risk neutral probability measure P chosen.
Exercise 6.22 For any n D 1; : : : ; T, define the discounted gain process gn D

.gnt /tD1;:::;T by

1 1 n
gntC1 D .dtC1 C sntC1 / snt / ; for all t D 0; 1; : : : ; T 1:
rft rf
Suppose that .S; D/ is a couple of price-dividend processes not admitting arbitrage

opportunities and let P be a risk neutral probability measure. Let .t /tD1;:::;T be a
predictable RN -valued stochastic process. Show that, for every t D 1; : : : ; T, it holds
that
" N ˇ #
X ˇ
n nˇ
E t gt ˇFt1 D 0
nD1
and, as a consequence,
X
T X
N
E Œtn gnt D 0
tD1 nD1
i.e., the cumulative expected discounted gains of any trading strategy are null under
a risk neutral probability measure.
Exercise 6.23 Let .S; D/ be a couple of price-dividend processes not admitting
arbitrage opportunities and let P be a risk neutral probability measure (which exists
by Theorem 6.23). As in Exercise 6.2, define the discounted quantities SN t WD St =rft
and D N t WD Dt =rft , for all t D 0; 1; : : : ; T. Similarly, for any trading-consumption
strategy .; c/, define the discounted portfolio value W N t ./ WD Wt ./=rft and the
discounted consumption cN t WD ct =rf , for all t D 0; 1; : : : ; T. Prove that, for every
t
x 2 RC , a non-negative adapted process c D .ct /tD0;1;:::;T belongs to C0C .x/ if and

only if the following two requirements are satisfied:
P P
(i) cN T D x C TtD1 t> . SN t C D N t / T1 N t , for some predictable process
tD0 c
. /
PtT tD0;1;:::;T ;

(ii) tD0 E ŒN ct D x.
Exercise 6.24 Let us consider a model with dates t D 0; 1; 2 with four states of
nature f!1 ; !2 ; !3 ; !4 g and the following information flow
F0 D f!1 ; !2 ; !3 ; !4 g;
˚
F1 D F11 D f!1 ; !2 g; F12 D f!3 ; !4 g ;
˚
F2 D f!1 g; f!2 g; f!3 g; f!4 g :
6.7 Exercises 341
Two assets are traded in the market, with dividends at t D 2 given by

2 3
4 2
62 37
DD6
42
7
95
4 3
the prices of the two assets are .3; 2/ in F11 , .2; 3/ in F12 , and .1:1; 0:9/ in F0 .
(i) Is the market dynamically complete?
(ii) Are there arbitrage opportunities in the market?
(iii) If it exists, compute a risk neutral probability measure.
(iv) Determine the no-arbitrage price at the initial date t D 0 of a European Call
option written on the first asset with strike price K D 3 and maturity T D 2.
Exercise 6.25 Suppose that .S; D/ is a couple of price-dividend processes not
admitting arbitrage opportunities. Generalize to the multi-period case the proof of
Proposition 4.29 and prove Proposition 6.30.
Exercise 6.26 Consider a binomial model with three dates as introduced at the end
of Sect. 6.3 (with t 2 f0; 1; 2g) and a contingent claim with payoff f .s2 /, where
s2 denotes the price of the risky asset at the terminal date T D 2. By using a
backward induction procedure, compute the arbitrage free price of the claim as well
as the associated hedging strategy. Verify that the arbitrage free price computed at
the initial date t D 0 coincides with the discounted risk neutral expectation (see
Corollary 6.34).
Exercise 6.27 Let .S; D/ be a couple of price-dividend processes not admitting
arbitrage opportunities and let u W RC ! R be a continuous, strictly increasing
and concave utility function. Prove that, for all x > 0, the optimal investment-
consumption problem (6.6) admits a solution in correspondence of .S; D/.
Exercise 6.28 Under the assumptions of Sect. 6.4, prove that the following asset
pricing relation holds true, for all n D 0; 1; : : : ; N and t D 0; 1; : : : ; T 1:
X
Tt X
Tt 0 ˇ
EŒdtCs
n
jFt s u .etCs / n ˇ
snt D C Cov ı 0 ; dtCs ˇFt ;
sD1
rfs sD1
u .et /
Exercise 6.29 In the context of Proposition 6.37, let .wO t /tD1;:::;T be an RNC1 -valued
predictable process satisfying 1> wO t D 1 for all t D 1; : : : ; T and such that, for every
t D 0; 1; : : : ; T 1,
> >
Corr rtC1 wO tC1 ; u0 .etC1 /jFt D min Corr rtC1 w; u0 .etC1 /jFt :
w2RNC1 Ww> 1D1
Show that relation (6.76) holds true.

Exercise 6.30 Prove Proposition 6.38

Exercise 6.31 In the setting of Sect. 6.4, consider a fixed t 2 f0; 1; : : : ; T 2g
and suppose at date t it is possible to trade a zero-coupon bond with maturity t C 1
having price B.t; t C 1/ and a zero-coupon bond with maturity t C 2 having price
B.t; t C 2/. Suppose furthermore that at the future date t C 1 it will be possible to
trade a zero-coupon bond with maturity t C 2 for the price B.t C 1; t C 2/. Let P
be a risk neutral probability measure for the economy. Prove that
B.t; t C 2/ D B.t; t C 1/E ŒB.t C 1; t C 2/jFt :
Exercise 6.32 In the context of Sect. 6.4, let g.t; / WD .etC =et /1= and suppose
that the representative agent has a power utility function of the type
x1
u.x/ D :
1
Show that the following relation holds in an approximate form:

log 1 C i.t; t C / D log ı C EŒlog g.t; /jFt ;
for all t 2 f0; 1; : : : ; T 1g and 2 f0; 1; : : : ; T tg.

Exercise 6.33 Under the assumptions of Sect. 6.4, suppose thatPT D 2 and
consider a representative agent with logarithmic utility function 2tD0 ı t log.ct /.
Suppose furthermore that the aggregate endowment process .et /tD0;1;2 starts from
a deterministic value e0 at the initial date t D 0 and, at date t D 1, takes the
two possible values e1 .u/ and e1 .d/ with probabilities p and 1 p, respectively,
and then evolves deterministically on the period Œ1; 2, taking values e2 .u/ and
e2 .d/, respectively (the price of aggregate endowment is supposed to be always
normalized to one). In other words, the uncertainty of the economy is represented
by the uncertainty about two possible states of the world realized at the intermediate
date t D 1. Show that
h i
1
1 1 1 E x.0;1/
1Ci.0; 1/ D h i ; 1Ci.0; 2/ D r h i ; 1Cf .0I 1; 2/ D ı h 1 i ;
1
ıE x.0;1/ ı E x.0;2/1 E x.0;2/
where x.0; 1/ WD e1 =e0 and x.0; 2/ WD e2 =e0 denote the growth rates of the aggregate
endowment on the periods Œ0; 1 and Œ0; 2, respectively. Show also that on the period
Œ1; 2 it holds that
x.1; 2I u/
1 C i.1; 2I u/ D if e1 D e1 .u/;
ı
(6.89)
x.1; 2I d/
1 C i.1; 2I d/ D if e1 D e1 .d/;
ı
6.7 Exercises 343
with i.1; 2I u/ and x.1; 2I u/ (i.1; 2I d/ and x.1; 2I d/, resp.) denoting respectively the
interest rate and the aggregate endowment growth rate on the period Œ1; 2 if state u
(state d, resp.) is realized.
Exercise 6.34 Under the assumptions of Sect. 6.4, suppose that the economy is
infinitely-lived (i.e., T D 1) and let B.t; t C / denote the price at date t of a zero-
coupon risk free bond maturing at date t C , for t 2 N and 2 N. Suppose in
addition
P1 t that there exists a representative agent characterized by the utility function
tD0 ı log.ct / and that the aggregate endowment process .et /t2N evolves according
to
etC1 D et utC1 ; for all t 2 N;
where .ut /t2N is a sequence of independent and identically distributed positive

random variables and > 0. Show that, for each t 2 N, it holds that

2
ı 1 ı2 1
B.t; t C 1/ D E and B.t; t C 2/ D E :
u1 2 u1
Exercise 6.35 Under the assumptions of Sect. 6.4, suppose that the economy is
(i.e., T D 1), with a representative agent characterized by the
infinitely-lived P
utility function 1 t 1˛
tD0 ı ct =.1 ˛/ and that the growth rate process .xt /t2N of the
aggregate endowment (i.e., xt D et =et1 , for t 2 N) evolves as a two-state Markov
chain with values h and l (with h > l) and transition probabilities

ll lh
˘D ;
hl hh
where lh D P.xtC1 D hjxt D l/, for all t 2 N, and similarly for the other quantities.
Compute
(i) the risk free rates rf .l/ and rf .h/ corresponding to the two possible states of
aggregate endowment growth rate;
(ii) the price of the security delivering the aggregate endowment as dividend;
(iii) the expected return of the above security.
Exercise 6.36 Under the assumptions of Sect. 6.5, suppose that the representative
agent is risk neutral and consider the following securities in an infinite time
horizon:
(i) a security paying a constant dividend stream dt D dN at all dates t D 0; 1; 2; : : :
(in an infinite horizon). Show that the fundamental value of this security is given
ı N
by st D 1ı d, for all t 2 N;
(ii) a security paying a constant dividend stream dt D d.1/. N At some date t0 , a
new dividend policy is unexpectedly being announced to start at the future date
Nt > t0 . Such a new policy consists in the distribution of the constant dividend
N
dt D d.2/. Show that the fundamental value of this security is given by
8
ˆ
ˆ ı N
if t < t0 I
< 1ı d.1/;

st D 1ı
ı N
d.1/ N
C ıNtt .d.2/ N
d.1// ; if t0 t < NtI
ˆ
:̂ ı d.2/;
N if t Nt:
1ı
Exercise 6.37 Under the assumptions of Sect. 6.5, suppose that the representative
agent is risk neutral.
(i) Consider a security paying the following dividend stream:
dt D dN C %.dt1 d/
N C "t ; for all t 2 N;
with EŒ"t jFt1 D 0, for all t 2 N. Show that the fundamental value of this
security is given by

ı ı% ı%
st D dN C dt ; for all t 2 N:
1 ı 1 ı% 1 ı%
(ii) Consider a security paying the following dividend stream:
dt D ˇdt1 ; for all t 2 N;
for some ˇ > 1 such that ˇı < 1. Show that the fundamental value of this
security is given by
ıˇ
st D dt :
1 ıˇ
(iii) Consider a security paying the following dividend stream:
log dt D C log dt1 C "t ; for all t 2 N;
for some 2 R and where ."t /t2N is a sequence of independent and identically
distributed normal random variables with zero mean and variance 2 . Show
2
that, if log ı C C 2 < 0, the fundamental value of this security is given by
2
ıeC 2
st D 2
; for all t 2 N:
1 ıeC 2
6.7 Exercises 345
Exercise 6.38 Under the assumptions of Proposition 6.40, prove the following
claims:
(i) a rational bubble cannot restart once it bursts, i.e., ˇt D 0 implies that ˇtCs D 0
for all s 2 N;
(ii) suppose that a price process is uniformly bounded from above, i.e., there exists
a constant K > 0 such that P.st 2 Œ0; K/ D 1 holds for all t 2 N. Prove that
st D st holds for all t 2 N, i.e., there is no bubble component.
Exercise 6.39 Under the assumptions of Proposition 6.40, suppose that the divi-
dend process .dt /t2N is defined by
log dtC1 D C log dt C "tC1 ;
with > 0 and where ."t /t2N is a sequence of independent and identically
distributed normal random variables with zero mean and variance 2 . Show that
there exists an (intrinsic) bubble of the form u0 .et /ˇt D cdt where c is a constant
and is the solution of the equation 2 2 =2CClog ı D 0 (see Froot & Obstfeld
[745]).
Chapter 7
Multi-Period Models: Empirical Tests
I believe there is no other proposition in economics which has

more solid empirical evidence supporting it than the efficient
market hypothesis. That hypothesis has been tested and, with
very few exceptions, found consistent with the data in a wide
variety of markets.
Jensen (1978)
Differently from single-period asset pricing models, which can be empirically tested
only by assuming that they hold period by period, the implications of multi-period
models can be directly tested on financial time series. In the present chapter, we
aim at discussing some of the most important empirical findings on the properties
of asset prices and returns. We will refer to the asset pricing theory described in
Chap. 6 (in particular Sect. 6.4) as classical asset pricing theory.
Any formulation of the classical asset pricing theory necessarily consists of two
essential ingredients: an equilibrium model and the information available to market
participants, represented by some information flow F. Moreover, any specification
of a model requires some hypothesis on the agents’ preferences and on the
technology of the economy, in particular concerning the agents’ utility functions, the
agents’ endowment and the available investment opportunities. Of course, different
implications can be obtained from different specifications. Furthermore, the results
obtained depend on the choice of the reference probability measure. Indeed, one can
either work under the historical/statistical probability measure P (i.e., the probability
measure describing the real economic model) or under a risk neutral probability
measure P . Note that, as pointed out above, heterogeneous beliefs (i.e., “private”
probability measures) are allowed provided that they agree on null sets.
Under a risk neutral probability measure P , the expected return of every
security is equal to the risk free rate and, hence, all risk premia are null and the
discounted asset price follows a martingale in case of a constant risk free rate
(see Proposition 6.22). As a consequence, the discounted wealth of a self-financing
portfolio is a martingale (see equation (6.54)). Under the assumption of a constant
risk free rate, asset returns are not serially correlated and future returns cannot be
predicted on the basis of the information currently available in the market. Indeed,

348 7 Multi-Period Models: Empirical Tests
in view of equation (6.55), it holds that
Cov .rtC1 ; rt / D E ŒrtC1 rt E ŒrtC1 E Œrt

(7.1)
D E E ŒrtC1 jFt rt E ŒrtC1 E Œrt D 0;
since E ŒrtC1 jFt D rf , for all t, with Cov .; / denoting the covariance under a
risk neutral probability measure P . Note that (7.1) relies on the assumption of a
constant risk free rate rf . If the risk free rate is time varying (recall that the risk free
rate can also be a stochastic process, as long as it is predictable, see Sect. 6.4), then
future expected returns are not constant and might be predictable on the basis of the
current information.
Under the historical probability measure P, property (7.1) does not hold in
general. In the special case where agents are risk neutral, in equilibrium the
historical probability measure P is also a risk neutral probability measure, with the
equilibrium risk free rate being equal to the inverse of the agents’ discount factor,
so that the risk free rate can vary over time only if the agents’ discount factor does.
However, this case represents a rather special situation. In the more interesting case
where agents are risk averse, the equilibrium risk free rate and the asset risk premia
are typically time varying and depend on the agents’ preferences and on the structure
of the economy (in particular, on the endowment and on the available information).
In a representative agent economy, the equilibrium risk free rate and the asset risk
premia are determined by (6.71)–(6.72). It is important to remark that, under the
historical probability measure, future returns could be predictable on the basis of the
information available in the market and can be serially correlated. The fact that the
equilibrium risk free rate and the asset risk premia vary over time is due to changes
in the agents’ preferences and/or to changes in the structure of the economy (e.g.,
set of available investment opportunities).
According to the above observations, returns should be unpredictable in a risk
neutral economy or, under suitable assumptions, in an economy with risk averse
agents and constant preferences and investment opportunities. In view of the above
discussion, it is important to point out that a test providing evidence in favor of
predictability of asset returns is not necessarily to be interpreted as an evidence
against the validity of the classical asset pricing theory. Such an evidence would
only mean that agents are not risk neutral and that the set of investment opportunities
is changing over time.
The literature on the empirical tests of classical asset pricing theory can be clas-
sified in three main groups, depending on the type of implications investigated:
1. restrictions on the asset price-dividend-return time series;
2. restrictions on the asset risk premia, risk free rate and consumption process;
3. information in financial markets.
The first group of contributions aims at testing the implications of no-arbitrage
and equilibrium conditions on financial time series (prices, dividends, returns),
while the contributions of the second group mainly test the structural restrictions
7.1 Tests on the Price-Dividend Process: Bubbles and Excess Volatility 349
established in Sect. 6.4 on asset risk premia, consumption and risk free rate.
The third group of contributions concerns the role of information on asset price
restrictions obtained in equilibrium or under the hypothesis of absence of arbitrage
opportunities. Of course, a statistical test providing negative evidence of classical
asset pricing theory can be explained in two opposite ways: a) the specification
of the theory is not confirmed in the real world; b) the set of information and/or the
model on which the theory is being tested is not correctly specified. In the remaining
part of this chapter we shall discuss empirical studies that belong to the first and to
the second of the above three groups, referring to the next chapter for an analysis of
the role of information in financial markets.
This chapter is structured as follows. In Sect. 7.1, we survey the empirical
literature testing the existence of bubbles and the excess volatility of asset prices.
Section 7.2 reviews the empirical evidence on return predictability, also discussing
its relation with market efficiency. In Sect. 7.3, we provide a brief overview of the
empirical evidence on the validity of equilibrium asset pricing models and discuss
the equity premium puzzle as well as several related asset pricing puzzles. At the end
of the chapter, we provide a guide to further readings as well as a series of exercises.
In this chapter, we will mention several concepts related to the statistical analysis
of time series. We refer the reader to the textbooks Campbell et al. [348], Hamilton
[881], Tsay [1601] for a thorough presentation of the estimation methods and of the
econometric analysis of time series.
7.1 Tests on the Price-Dividend Process: Bubbles and Excess

Volatility
The first group of contributions mentioned above can be further classified into
three subgroups: studies on the presence of bubbles (testing whether prices are in
agreement with the fundamental solution (6.86) of the no arbitrage equation (6.85));
studies on the volatility of financial time series; studies on return predictability. In
many cases, the analysis concentrates on a specific asset pricing model, typically
represented by a risk neutral economy with constant investment opportunities
and discount factor. Again, we want to stress that empirical evidence against the
implications of a model (for instance, evidence of return predictability) does not
necessarily provide evidence against classical asset pricing theory as whole, but may
only indicate that risk neutrality is violated and/or that the structure of the economy
is changing over time.
Before starting our survey of the empirical literature, we want to mention that
the analysis of financial time series started even before the modern theory of
financial markets. Indeed, the conjecture that the no-arbitrage condition implies
unpredictability of price changes can be traced back to Bachelier (see Bachelier
[96]). Bachelier suggested that French government bonds were characterized
by an evolution which is well described by a random walk stochastic process
(see Campbell et al. [348] for a detailed presentation of the random walk model).
Bachelier’s contribution remained unrecognized for a long time, since the academic
and the financial community were skeptical about an asset pricing theory involving
random processes. For a long time, financial practitioners (well represented by Gra-
ham & Dodd [818]) identified the fundamental value of an asset as the discounted
flow of future dividends (fundamentalist approach). A stochastic approach to the
analysis of financial markets was also outside the economic theory paradigm: at
that time, neoclassical economics was looking for an equilibrium foundation of
market prices. According to that theory, prices are only related to the technology and
the preferences of the economy, without an explicit use of probabilistic tools. The
seminal contribution which marks the reconciliation between the fundamentalist
approach and Bachelier’s intuition is Samuelson [1491], where, assuming risk
neutral agents and using no-arbitrage arguments together with the law of iterated
expectations, it was shown that prices follow a martingale process.
Bubble Solutions vs. Fundamental Solution
The literature on the presence of speculative bubbles (in the sense of Sect. 6.5) in
financial time series is quite large (see West [1656], Flood & Hodrick [719], Gurkay-
nak [866] for a survey). First of all, the condition established in Tirole [1595] for
the existence of a bubble in a growing economy (dynamic inefficiency) has not
been confirmed empirically in developed countries (see Abel et al. [8]). Recalling
that a rational bubble is a non-negative process, a consequence of the presence
of a speculative bubble in financial time series is that prices grow at a rapid rate.
Consequently, if the dividend grows more slowly that the long run real interest
rate, then the dividend-price ratio should decrease. The empirical evidence in this
direction is weak.
If dividends are generated by a linear non-stationary stochastic process, then the
linearity of the no-arbitrage equation implies that, in the absence of a speculative
bubble, the stock price and the dividend processes are cointegrated (see Hamilton
[881, Chapter 19] for a description of cointegration). This feature of a financial time
series can be tested empirically. In particular, a stationarity test based on the non-
cointegration property of a bubble solution immune to the problems pointed out
in Hamilton & Whiteman [882] has been proposed in Diba & Grossman [574] in
a constant discount rate model, allowing for possible unobservable variables. The
empirical evidence obtained is against the presence of a bubble and this result is
confirmed by a large part of the subsequent literature. On the other hand, results in
favor of the presence of an intrinsic bubble in stock prices time series are obtained in
Froot & Obstfeld [745]. Note that stationarity tests are not able to detect periodically
collapsing bubbles of the type proposed in Blanchard & Watson [248] (see Evans
[656]).
The presence of a bubble has also been associated with the excess volatility
phenomenon, meaning that the asset price volatility is too high and is not compatible
with the no-arbitrage equation, as we are going to explain more precisely below.
In Blanchard & Watson [248] and Tirole [1595], depending on the correlation of
innovations of the bubble (i.e., the differences ˇt EŒˇt jFt1 ) with those of the
fundamental solution, it is claimed that a violation of volatility bounds as observed
for example in Shiller [1533] (see below) can be attributed to the presence of
a bubble. Moreover, if the innovations of the bubble component are positively
correlated with the innovations of the dividend process, then the presence of a
bubble leads to an increase of the variance. However, in Ikeda & Shibata [997],
considering an intrinsic bubble, it is shown that the volatility of a price process
exhibiting a speculative bubble can be both larger or smaller than the volatility
corresponding to the fundamental solution. Similarly, Drees & Eckwert [590] show
that, in the absence of correlation in asset returns, the volatility of a price process
exhibiting an intrinsic bubble can be either smaller or greater than the volatility
of the fundamental solution depending on agent’s risk attitude. Flood & Hodrick
[718] and Flood et al. [720] remark that many excess volatility tests effectively
embed bubbles into the null hypothesis, so that a failure of variance bounds test
cannot be attributed to the presence of a speculative bubble. A test on the presence
of speculative bubbles in a constant discount rate model based on the comparison of
two sets of estimates of the parameters needed to calculate the expected discounted
value of a given stock’s dividend stream has been proposed in West [1655]. The
author finds evidence in favor of the rejection of the hypothesis of absence of a
speculative bubble and claims that a bubble can be the origin of the excess volatility
phenomenon (see also Blanchard & Watson [248]).
A test on the presence of a bubble may be ambiguous, since it may be difficult to
distinguish empirically between bubbles and other phenomena such as irrationality,
noise, structural changes, non-stationarities and (rational) contribution to the asset
prices of factors observed by the market participants but not by the econometrician
(in this direction, see Hamilton & Whiteman [882], Flood & Hodrick [719], West
[1656]). Moreover, many tests are affected by the fact that the null hypothesis is
composite and the statistical power of the test is rather low. According to Hamilton
& Whiteman [882], the presence of a bubble is empirically untestable and can be
interpreted within the classical asset pricing framework through the presence of
unobservable factors.
Excess Volatility
The literature on asset price excess volatility is based on the observation that the
empirical volatility of asset prices is larger than what predicted on the basis of the
no-arbitrage equation (6.85). The literature on this topic is quite large and originated
from LeRoy & Porter [1189], Shiller [1533]. We refer the reader to LeRoy [1184],
Cochrane [458], Shiller [1538], West [1656], Gilles & LeRoy [782] for good surveys
on the topic.
Let us start by explaining the excess volatility phenomenon, adopting the setting
of Sect. 6.5. Besides the assumptions of Sect. 6.5, assume furthermore that the
representative agent is risk neutral and is characterized by a discount factor ı D
1=rf 2 .0; 1/. In this case, equation (6.86) implies that the fundamental value of a
security paying the dividend stream .dt /t2N can be expressed as
X1
1
st D EŒdtCs jFt ; for all t 2 N:
rs
sD1 f
We denote by .set /t2N the ex-post rational (or perfect foresight) price process of the
security, corresponding to the price of the security in the hypothetical case where
the dividend stream is fully known in advance:
X1
1
set WD dtCs ; for all t 2 N: (7.2)
rs
sD1 f
By comparing the representations of st and set above, we obtain the decomposition
set D st C ut ; for all t 2 N; (7.3)
where the process .ut /t2N represents the deviation of the present value of the future
dividends from its conditionalPexpectation based on the information Ft available at
time t and is defined by ut D 1 1
sD1 rfs "tCs , where the process ."t /t2N represents the
unexpected component of the variation in the stock price and is given by
"t WD st C dt rf st1 : (7.4)
Note that EŒ"t jFt1 D 0, for all t 2 N. Moreover, since r1f EŒst C dt jFt1 D st1 ,
for all t 2 N, the process ."t /t2N is serially uncorrelated. The decomposition (7.3)
of the perfect foresight price set follows by noting that, for any t 2 N,
1 1
!
X 1 X 1 1
s " tCs D se
t C s stCs s1 stCs1 D set st ;
r
sD1 f sD1
rf rf
as long as limT!1 stCT =rfT D 0, for all t 2 N.

Note that, in view of the definition of the process ."t /t2N , it holds that
X1
1
EŒut jFt D s EŒ"tCs jFt D 0;
r
sD1 f
meaning that the fundamental value st represents an unbiased estimate of the perfect
foresight price set given the information available at time t. In particular, this implies
that Cov.ut ; st jFt1 / D 0, for all t 2 N, and, hence,

Var set jFtj D Var st jFtj C Var ut jFtj Var st jFtj ; (7.5)
for all t 2 N and j 2 f1; : : : ; t 1g. Furthermore, provided that unconditional

variances are well-defined, it holds that

Var set D Var st C Var .ut / Var st ; for all t 2 N: (7.6)
The intuition behind the above inequalities is the following: the fundamental
value is an unbiased forecast of the perfect foresight price and, therefore, its
variance is smaller than the variance of the perfect foresight price (being a
conditional expectation of the latter). Inequality (7.6) is always valid regardless of
the information available to market participants and of the specific dividend process
considered, under the standing assumption that the second moments exist. These
features make inequality (7.6) empirically testable. However, in order to evaluate
the statistical significance of a violation of inequality (7.6), one needs to specify a
model for the asset dividends.
As shown in (7.6), the difference between the variance of the perfect foresight
price and the variance of the fundamental value is given by Var.ut /. Clearly, the
process .ut /t2N cannot be observed in real markets. However, if the distribution of
the unexpected component " of the variation in the stock price is stationary, so that
Var."tCs / D Var."t /, for all t; s 2 N, then it holds that
X1
1 1
Var.ut / D 2s
Var."tCs / D 2 Var."t /; for all t 2 N: (7.7)
r
sD1 f
rf 1
It has been remarked in LeRoy & Porter [1189] that the variance of asset prices
increases with the information available to market participants, while the variance
of unexpected returns goes in the opposite direction. Indeed, let us consider two
information flows F and G with the property that Ft Gt , for all t 2 N (intuitively,
the information represented by Gt is finer than the information represented by Ft )
and let us denote by sFt and sGt the conditional expectations (with respect to Ft
and Gt , respectively) of the perfect foresight price, at date t. Then, as shown in
Exercise 7.1, it holds that

Var sFt Var sG
t (7.8)
and, under the assumption that the distributions of the unexpected components "F
and "G are stationary, so that Var."FtCs / D Var."Ft / and Var."G G
tCs / D Var."t /, for all
t; s 2 N, it holds that

Var "Ft Var "G
t : (7.9)
Since the rational ex-post value set given in (7.2) cannot be observed in real
markets, one needs to estimate set on the basis of market data. In the seminal paper
Shiller [1533], the infinite sum appearing in the rational ex-post value (7.2) is
estimated as (compare with Marsh & Merton [1306])
X
Tt1
1 1 e
sOet D s dtCs C Tt s
NT ; for t D 0; 1; : : : ; T 1; (7.10)
sD1
rf rf
where T represents a large enough time horizon of the dividend time series and
sNeT an arbitrary terminal value (for instance, sNeT can be taken as the average of
the historically observed prices, see Grossman & Shiller [848]). Note that this
truncation of the infinite sum (7.2) does not automatically exclude the presence of a
non-null bubble component. The sample variance of sOet is then taken as an estimator
of the variance Var.set / of the ex-post rational price. The idea of Shiller [1533] is
that, if the time series is sufficiently long, then a reasonable estimate of the variance
of set can be obtained on the basis of (7.10).
In Shiller [1533], inequality (7.6) has been tested on the Standard and Poor index
time series over the period 1871–1979. The author verified in the first place that
the time series is characterized by a long run exponential growth rate. Removing
this trend, inequality (7.6) turns out to be empirically violated: the volatility of the
stock index is dramatically higher than the volatility of the rational ex-post price
(excess volatility phenomenon). As no specific model for dividends was assumed
(in other words, the test procedure is model-independent), the statistical significance
of this violation of inequality (7.6) cannot be statistically assessed. Nonetheless,
the remarkable difference between the two variances motivated Shiller [1533] to
claim that the empirical evidence is in favor of a strong violation of the implications
of classical asset pricing theory. This conclusion is reinforced by the comparison
between the time series of the stock price index and that of the rational ex-post
price: the latter appears to be substantially smoother and more stable than the first
one. Always in Shiller [1533], alternative tests were conducted by comparing the
variances of (de-trended) price innovations and the variance of the dividend process,
with empirical results always in contradiction with the implications of classical asset
pricing theory: stock prices appear to be excessively volatile.
In LeRoy & Porter [1189], excess volatility was also detected. The authors
assumed that dividends and stock prices (suitably taking into account a trend
component) are generated by a stationary bivariate linear process. Inequality (7.6)
was empirically tested together with the absence of correlation between prices and
return forecasting errors. In particular, inequality (7.6) is shown to hold empirically
in the opposite direction and volatility bounds are empirically violated. The null
hypothesis of the absence of correlation test is rejected but the statistical test leads
to very wide confidence intervals and does not always lead to statistically significant
results.
These results, in particular those reported in the seminal paper Shiller [1533],
stimulated a large debate on the excess volatility phenomenon. First of all, the results
were discussed concerning the statistical methodology employed. In particular,

two statistical issues deserve attention: the hypothesis of stationarity of the price-
dividend time series and the methodology used to estimate the volatility of the
time series. In Flavin [715], it is observed that the variances of st and set are both
estimated with a downward bias in small samples due to the presence of serial
correlation and, furthermore, this bias is more pronounced for the time series of
set . The bias derives from the estimation of the population mean through the sample
mean and, in the case of the time series of the rational ex-post price, from the fact
that set is computed as a moving average. Moreover, the methodology employed
by Shiller [1533] to compute the rational ex-post price set introduces a further
downward bias (see Gilles & LeRoy [782]). This remark is important since such
a downward bias can be large enough to provide a potential explanation for the
apparent violation of the volatility bounds, with the effect that variance bound tests
tend to be biased in favor of a rejection. Moreover, as we have already pointed out,
without specifying the model generating the dividends, it is difficult to quantify
the bias and, therefore, to take it properly into account. An additional problem is
represented by the presence of nuisance parameters, due to the fact that inequality
(7.6) holds independently of the agents’ information. The sample distribution of
the test statistic is affected by parameters which are unrestricted under the null
hypothesis of validity of the variance bounds. The agents’ information turns out
to be relevant in order to establish a critical value of the volatility bound test. As a
consequence, unless the agents’ information is precisely defined, it is not possible
to evaluate the statistical relevance of a violation to the volatility bounds (see, e.g.,
LeRoy & Parke [1188] and LeRoy & Steigerwald [1190]).
Kleidon [1104] deeply criticized the results reported in Shiller [1533]. His criti-
cism concerns two main issues: (i) the distinction between the relative smoothness
of the time series of set and st and the variance inequality (7.6); (ii) the stationarity of
the dividend process (and note that both criticisms are not concerned with the sample
size). First of all, Kleidon [1104] points out that the procedure adopted in Shiller
[1533] to estimate the rational ex-post price set (see equation (7.10)) relies on ex-post
information which is available only after the date when prices are set (uncertainty in
future dividends) and, therefore, the estimate of the conditional variance of the ex-
post rational price can be biased. If dividends follow an autoregressive linear process
(either stationary or non-stationary, see Hamilton [881] for a detailed presentation)
and conditional variances are calculated properly, then inequality (7.5) is satisfied
and the pattern of the rational ex-post price sOet calculated according to the Shiller
procedure can be smoother than that of the price st computed according to the
fundamental no-arbitrage solution. In particular, this smoothness effect is obtained
in the presence of a sufficiently strong autoregressive component. Moreover,
Kleidon [1104] shows that, if dividends follow a geometric random walk (so that
log-returns are non-stationary), then the time series of the rational ex-post price is
always smoother than that of the fundamental no-arbitrage price and inequality (7.5)
between the conditional variances is satisfied (even if the unconditional variances
cannot be defined). This observation leads to the conclusion that there is no direct
relation between the relative smoothness of the two time series and the variance
inequalities (7.5)–(7.6). Moreover, a variance inequality has to be regarded as a

cross-sectional evaluation and, therefore, one simple plot of the two time series and
a comparison of their smoothness is rather uninformative. In other words, ex-post
one does only observe one of the many ex-ante possible realizations of the economy
and, therefore, one cannot look at different values of set , each corresponding to
a different realization, to test whether inequalities (7.5)–(7.6) are satisfied. This
observation becomes particularly critical when the dividend time series is non-
stationary. Kleidon [1104] relies on this argument to explain the results in Shiller
[1533], arguing that a geometric random walk for dividends is a good model for
the time series at hand. In this case, unconditional variances are not well-defined
and the use of sample variances as estimators is invalid, so that the test procedure
adopted by Shiller [1533] is uninformative. Moreover, Kleidon [1104] shows by
means of a Monte Carlo analysis that Shiller’s procedure for the computation of
the unconditional variance on a small sample assuming a geometric random walk
process (with a normally distributed noise term) for dividends leads to an empirical
violation of inequality (7.6). Such a violation occurs in more than 70% of the cases
and in most of the cases the size of the violation is similar to that detected by
Shiller, while a conditional variance test for the Standard and Poor time series is
not violated. This different behavior should be attributed to the non-stationarity of
the de-trended time series investigated by Shiller. According to Kleidon [1104], an
unconditional variance test is uninformative, since the unconditional variance is not
well-defined, while the conditional variance inequality (7.5) is satisfied.
Shiller [1537] replied to the above arguments by observing that the assumptions
adopted by Kleidon [1104] produce unrealistic price-dividend ratios and by showing
(through Monte Carlo simulations, as in Kleidon [1104]) that violations with a size
equal to that detected in Shiller [1533] happen with small probability (smaller than
1%) in the case of a log-normal process of the dividends with realistic values for the
price-dividend ratio and in the case of other non-stationary dividend processes.
Summing up, we can affirm that an empirical violation of inequality (7.6)
according to the test procedure proposed by Shiller is in favor of a rejection of
classical asset pricing theory (due to the excess volatility phenomenon) if the
dividend process, once the trend component is removed, is assumed to be stationary.
However, the same conclusion cannot be drawn if de-trended dividends (or log-
dividends) are non-stationary (random walk process), because in that case the
unconditional variance is not well-defined. In the latter case, the empirical variances
estimated on small samples often do not satisfy inequality (7.6).
The first generation of excess volatility tests suffered for small sample biases
and non-stationarity problems. To overcome these deficiencies, a second generation
of volatility tests has been proposed in the literature, see in particular Mankiw
et al. [1294], West [1657], Campbell & Shiller [351], Mankiw et al. [1295]. In
Mankiw et al. [1294], a “naive forecast” of the future discounted dividend stream
is considered and a test is proposed which does not suffer for small sample biases
and does not depend on the assumption of stationarity of the dividend time series.
Suppose that, at each date t 2 N, one uses a subset of the available information to
make a “naive forecast” (which may not be a rational forecast) of future dividends
and denote by Ft ./ the corresponding forecast operator, so that Ft .dtCs / represents
the naive forecast at date t of the dividend paid at the future date t C s. Define then
X1
f 1
st WD Ft .dtCs /; for all t 2 N:
rs
sD1 f
Similarly as in the case of inequality (7.5), the following inequalities hold by

definition, as shown in Exercise 7.2 (see also Mankiw et al. [1294]):
EŒ.set st /2 jFt EŒ.set st /2 jFt EŒ.set st /2 jFt EŒ.st st /2 jFt :

f f f
and
(7.11)
Moreover, by the tower property of the conditional expectation, the above inequali-
ties also imply that
EŒ.set st /2 jKt EŒ.set st /2 jKt EŒ.set st /2 jKt EŒ.st st /2 jKt ;

f f f
and
for any subset Kt Ft of the information available at date t, for every t 2 N

(in particular, one can take Kt D Ftı , for some ı 2 N). In particular, as long
as the above conditional expectations are taken with respect to the information
available a finite amount of time before date t, the non-stationary of the dividend
time series does not pose problems concerning the existence of the conditional
expectations. The above inequalities mean that the market price is a better forecast
(in the mean-square sense) of the perfect foresight price (which is constructed as
in Shiller [1536]) than the naive forecast. In their empirical application, Mankiw
et al. [1294] suppose that the naive forecast is completely myopic, in the sense that
future dividends are naively forecasted to be always equal to the current dividend,
and show that the above inequalities are violated, but do not discuss the statistical
significance of such a violation. In Mankiw et al. [1295], a Monte Carlo sampling
distribution of the test is proposed and it is shown that, assuming a constant discount
factor smaller than 6%, the violation of the inequality turns out to be statistically
significant, while assuming higher discount factors leads to violations that are not
necessarily statistically significant.
In Blanchard & Watson [248] and West [1657], a test based on the information
available to market participants is proposed. For every date t 2 N, consider two
information flows F D .Ft /t2N and G D .Gt /t2N such that the first one contains less
information than the second one (i.e., Ft Gt , for all t 2 N) and let sFt and sG t be
the rational forecasts of set based on the two information sets Ft and Gt , respectively,
for all t 2 N. Assuming the stationarity of the forecasting errors with respect to both
information flows F and G, the following inequality holds (see Exercise 7.3):
EŒ.sFt C dt EŒsFt C dt jFt1 /2 EŒ.sG G 2

t C dt EŒst C dt jGt1 / : (7.12)
A similar inequality can be established about the return variance (both conditional
and unconditional), see LeRoy & Porter [1189]. In West [1657], an empirical test
of inequality (7.12) has been proposed. In particular, such a test does not require
the knowledge of the rational ex-post (perfect foresight) price set and does not
depend on the stationarity of the dividend process. The two information sets Gt
and Ft are respectively represented by the information available in the market at
date t and by the observation of the time series of the dividends. Inequality (7.12)
is empirically violated, with the violation being highly statistically significant.
Moreover, Monte Carlo simulations show that the test does not suffer for small
sample biases. However, Ackert et al. [13] remark that earlier studies have only
considered ordinary cash dividends, neglecting the effects of share repurchases and
takeover distributions. Taking into account these phenomena, the variance bounds
are empirically satisfied and no sign of excess volatility is detected (see Ackert et al.
[13]).
Summing up, the second generation of volatility tests provided a general
evidence of excess volatility, with excess volatility being detected also when
the dividend time series is allowed to be non-stationary. However, the statistical
significance of the results is not always strong.
A Critical Assessment
On the basis of the results reported above, a large debate started in the literature
on the interpretation of the excess volatility phenomenon. In this regard, one can
distinguish three main lines of thought: (i) studies pointing out statistical problems;
(ii) studies allowing for non-constant investment opportunities and (iii) studies
supportive of approaches going beyond classical asset pricing theory. In particular,
the contributions from the second group propose an explanation of the excess
volatility phenomenon within the realm of the classical asset pricing theory, while
the contributions from the third group suggest that the excess volatility originates
from some form of market irrationality which is not taken into account by classical
asset pricing theory (behavioral finance).
Remaining in the context of classical asset pricing theory, non-constant invest-
ment opportunities (non-constant expected returns) can be introduced by assuming
a risk averse representative agent. Indeed, in view of equation (6.71), a constant
risk free rate is obtained in the case of a risk neutral representative agent or
when the aggregate endowment process is constant over time. Risk aversion has
been introduced in LeRoy & LaCivita [1187], Grossman & Shiller [848], West
[1657]. As shown in Sects. 6.4–6.5, classical asset pricing theory establishes that the
equilibrium price of an asset at date t is equal to the discounted expected value of its
future dividends (together with a possible bubble component), with the (stochastic)
discount factor at date t for consumption at the future date t C s being given by
the representative agent’s marginal rate of intertemporal substitution between dates
t C s and t. In particular, this implies that the discount factor is stochastic if the
representative agent is not risk neutral (in which case the discount factor would be
simply given by 1=rf , see equation (6.71)).
In the presence of a risk averse representative agent, equation (6.69) suggests
that the variability of the ex-post rational price (i.e., perfectly knowing the whole
future dividend stream) could happen to be larger than that of an ex-post rational
price corresponding to a constant discount factor. In other words, the risk aversion
of the representative agent can lead to more volatile asset prices. The argument can
intuitively be explained as follows. In a pure exchange representative agent econ-
omy, in equilibrium the representative agent consumes the aggregate endowment
of the economy (no-trade equilibrium). Of course, in equilibrium the price system
should be compatible with this choice. In general, a risk averse agent would like to
smooth consumption over time. However, this cannot be realized at the aggregate
level. Hence, equilibrium prices need to be such that the representative agent
optimally chooses not to smooth consumption and simply consumes the aggregate
endowment at each date. In order to induce the representative agent not to buy stocks
in good periods and sell stocks in bad periods, stock prices must be procyclical
and, therefore, highly volatile (see, e.g., LeRoy [1184]). Assuming a power utility
function, in Grossman & Shiller [848] it has been observed that the variability of
rational ex-post prices increases with the agent’s risk aversion, with asset prices
being supposed to be coherent with (6.69). Assuming a coefficient of relative risk
aversion equal to four, a time series of the rational ex-post price compatible with the
empirical time series can be reproduced. These results have been also confirmed in
Cochrane & Hansen [467], where it is shown that the time variability of expected
returns allows to explain excess volatility if agents are strongly risk averse or if
their preferences are characterized by a habit formation process. However, as the
consumption time series are relatively smooth over time, the risk aversion coefficient
should be very high in order to reconcile classical asset pricing theory with the
empirical data: this observations is consistent with the equity premium puzzle (see
below). In addition, Mehar & Sah [1321] show that fluctuations in the subjective
discount factor and in the attitude towards risk may also explain the excess volatility
phenomenon.
A methodology that allows to investigate the presence of excess volatility with
time varying expected returns was developed in Campbell & Shiller [351, 352]
(see also Cochrane [465, Chapter 20]). The methodology is based on a simple log-
linear approximation. For all dates t 2 N, let RtC1 WD log..stC1 C dtC1 /=st / be the
logarithmic return of the asset. Then the following log-linear approximation of the
logarithmic return (centered on the logarithm of the average value of the dividend-
price ratio) can be obtained (see Exercise 7.4):
RtC1 k C log stC1 C .1 / log dtC1 log st ; (7.13)
where

WD 1= 1 C exp.log d log s/ and k WD log./ .1 / log.1= 1/:
In particular, when the dividend-price ratio is constant, then D st =.st C dt /, for

all t 2 N, with log.s=d/ representing the average value of the logarithm of the
price-dividend ratio. The precision of this approximation is rather satisfactory and,
moreover, it holds as an exact relation if the dividend-price ratio does not vary over
time.
Iterating equation (7.13) and assuming that a transversality condition of the type
lim !1 .stC C dtC / D 0 holds, the following approximation for the logarithm
of the price-dividend ratio is then obtained:
X 1
k
log dt log st C s .RtC1Cs log dtC1Cs /; (7.14)
1 sD0
with log dtC1Cs WD log dtC1Cs log dtCs , for all t; s 2 N. Approximation
(7.14) holds ex-post but, by taking the Ft -conditional expectation on the right-
hand side of (7.14), we can obtain an analogous ex-ante relation between the
logarithm of the price-dividend ratio and the forecast at date t of future changes
in dividends and returns, so that the logarithm of the price-dividend ratio turns
out to be approximately equal to the conditional expectation of future discounted
(logarithmic) returns and of future dividend growth rates. As a consequence, the
dividend-price ratio will be large if dividends grow slowly or if future returns are
forecasted to be large. Note, however, that expected dividend growth and expected
returns are positively correlated, thus yielding an offsetting effect on the logarithm
of the dividendŰ-price ratio (see Lettau & Ludvigson [1196]).
Exploiting the linearity of the above approximation, a vector autoregression (see
Hamilton [881, Chapter 11]) has been estimated in Campbell & Shiller [351] for
the logarithm of the dividend-price ratio, the first difference of the logarithm of
dividends and the logarithm of earnings relative to price. The restrictions on the
regression coefficients implied by the approximate relation (7.14) are empirically
rejected, thus showing that the dividend-price ratio time series do not agree with
(several versions of) this model. The test procedure employed in Campbell & Shiller
[351] does not suffer from possible non-stationarity problems. Moreover, Monte
Carlo simulations show that the test works well in small samples. Excess volatility
is detected assuming a constant or a variable expected rate of return (a constant plus
the expected real return on commercial paper).
Some authors have pointed out that allowing for time varying expected returns
does not suffice to explain the excess volatility phenomenon. Indeed, as shown
in Shiller [1533], Mankiw et al. [1295], Campbell & Shiller [351], variance
inequalities of asset prices are empirically violated even under the assumption of
non-constant expected returns (estimated as the risk free rate plus a time varying
risk premium component), thus proving that time varying discount factors only
explain part of the excess volatility phenomenon. To reproduce the asset price
volatility observed historically, discount rates should have a much higher standard
deviation than what is observed in real time series (see, e.g., Shiller [1533], West
[1657], Poterba & Summers [1430]).
According to the fundamental solution (see equation (6.86)), changes in asset

prices are due to changes in expected dividends and in the discount rates. This
basic observation motivated several studies on the relationship between ex-post
changes in economic fundamentals and ex-post changes in stock price-returns. In
Roll [1459], the time series of the orange juice future contract has been analysed.
Due to the specific characteristics of this commodity, its fundamental is well
proxied by the weather conditions in Florida. As a consequence, the price of the
future contract should reflect the predictable part of weather patterns and, therefore,
future prices should change in response to predictable changes in weather patterns.
Instead, the author finds that prices react to unanticipated weather changes, but these
unexpected changes only explain a small portion of the daily price variation and,
actually, around 90% of the daily price variability cannot be explained by economic
fundamentals. Similar results have been obtained for stock returns in Roll [1460],
taking into account economy, industry and firm specific factors. Considering both
CAPM and multi-factor models and including an industry factor, less than 35% of
the daily/monthly variations of individual stock returns can be explained in terms of
changes of economic fundamentals. Similar results have been obtained in Campbell
& Shiller [351], Cutler et al. [510].
In Campbell [331], Campbell & Ammer [338], Campbell [334], Cochrane [463],
by relying on the log-linear approximation (7.13), it has been shown that the
variance of aggregate unexpected returns can be explained in terms of the variability
of expected returns-risk premia (expected return news) rather than of the variability
of future cash flows (dividend news). The opposite holds for individual firms (see
Vuolteenaho [1633]) and for the book to market ratio (see Cohen et al. [472]). Only
a small fraction of the variation in book-to-market ratios is explained by variations
in expected returns. Campbell & Shiller [351] show that the dividend-price ratio
predicts future expected returns rather than dividend changes. These results confirm
that expected returns are time varying and that their variability over time is important
in order to explain price/return movements.
Summing up, we observe that a large part of the excess volatility tests has dealt
with models with a constant discount factor (risk neutral agents), showing that this
assumption does not allow to explain the magnitude of the variability observed
in financial time series. This empirical evidence can be partly explained within
the context of classical asset pricing theory by introducing risk aversion and time
varying expected returns. However, in order to explain the volatility of financial
markets, the recent literature has shown that either agents exhibit a very strong risk
aversion or discount rates are much more variable than what empirically observed.
As suggested by Fama [663] and Cochrane [458, 461], returns are predictable and
we do not have a good model to explain their dynamics: this is the origin of excess
volatility results.
An alternative interpretation of excess volatility has been proposed in Shiller
[1535]. According to his interpretation, the excess volatility phenomenon is due to
the inefficiency of the market: in particular, there are some forms of irrationality
(noise traders, feedback trading, irrational expectations) in the market that make
market prices deviate from rational prices (we will return to this topic in Chap. 9).
As a matter of fact, regardless of the interpretation of the results reported above,

excess volatility test rejections are strictly related to the predictability of future
returns (see, e.g., Campbell & Shiller [352]). Indeed, both volatility tests and
absence of correlation tests are based on the fact that ut and st are uncorrelated (see
equation (7.3)). If this is not the case, then future returns can be (at least partially)
predicted through the available information. A volatility bound violation may be due
to the fact that st and future returns are negatively correlated. With some caution,
this result can be interpreted as an evidence of negative serial correlation of returns
(see LeRoy [1184]). Indeed, referring to the decomposition (7.3), the presence
of excess volatility can be interpreted in terms of negative correlation between a
weighted average of past returns (the quantity st ) and a weighted average of future
returns (the quantity ut ).
7.2 Tests on the Price-Dividend Process: Return

Predictability
In the ’80s, the literature on excess volatility animated the debate on the efficient
market hypothesis. Afterwards, the debate moved to the more general theme of asset
return predictability (we refer to LeRoy [1184], Fama [663], Kaul [1076] for good
surveys on the topic). As we have seen above, return predictability is intrinsically
related to the excess volatility phenomenon.
According to the classical asset pricing theory as presented in Sect. 6.4, returns
are unpredictable under a risk neutral probability measure or, under the historical
probability measure, in the presence of risk neutral agents (with a constant discount
factor). Indeed, in view of Proposition 6.35, under one of these two assumptions,
the conditional expected return of any asset (or portfolio of assets) is simply equal
to the risk free rate, so that the information available at date t (represented by Ft )
does not yield any useful information for predicting future returns. However, as
we have already pointed out, in the absence of one of these hypotheses, classical
asset pricing theory does not necessarily imply that returns are unpredictable. It
is important to emphasize this point since return predictability has often been
erroneously interpreted as an evidence against market efficiency/classical asset
pricing theory.
As before, we assume that a risk free asset is available in the market, yielding
the constant risk free rate rf > 0. We also assume that agents are risk neutral (a
similar analysis can be performed with respect to a risk neutral probability measure).
Denoting by rt the return of an asset over the time period Œt 1; t, we let zt WD rt rf
denote the excess return of the asset over the same time period, for t 2 N. Risk
neutrality together with Proposition 6.35 implies that
EŒztC1 jFt D 0; for all t 2 N: (7.15)

7.2 Tests on the Price-Dividend Process: Return Predictability 363
In other words, the excess return of any asset represents a fair game: at every date t,
the Ft -conditional expectation of the asset excess return is simply equal to zero.
Equation (7.15) implies an ex-ante restriction on the asset prices time series,
which translates into an ex-post relation of the following form:
rtC1 D rf C tC1 ; (7.16)
where .t /t2N is a sequence of random variables such that EŒtC1 jFt D 0, for all
t 2 N. Relation (7.16) is related to a random walk process. Following Campbell
et al. [348, Chapter 2], three different types of random walk processes can be
identified depending on the features of the sequence .t /t2N : (1) a random walk
with independent and identically distributed innovations; (2) a random walk with
independent innovations; (3) a random walk with serially uncorrelated innovations.
Clearly, a random walk of the first type is also a random walk of the second type and,
similarly, a random walk of the second type is also a random walk of the third type.
Note also that a random walk of the second type allows for heteroskedastic return
time series (i.e., time varying volatility, see Campbell et al. [348] and Tsay [1601])
and a random walk of the third type allows for conditional heteroskedasticity in
asset returns. Time varying volatility in financial time series is a widely documented
phenomenon (see Schwert [1509]). In particular, conditional heteroskedasticity is
typically evident as volatility is a highly persistent phenomenon with clustering
patterns.1
The fundamental conditions (7.15)–(7.16) only require the return process .rt /t2N
to follow a random walk of the weakest form, with serially uncorrelated innovations.
In particular, (conditional) heteroskedasticity is perfectly compatible with the
hypothesis of efficient markets in the form (7.15)–(7.16).
Condition (7.15) captures the notion of returns unpredictability and has important
implications on the properties of asset returns time series. Indeed, (7.15) shows that
the information contained in Ft does not allow to make predictions about future
returns and, as a consequence, returns cannot be serially correlated (recall that, by
definition, the random variable rt is Ft -measurable). Moreover, as shown in the next
proposition (see Exercise 7.6 for a proof), there does not exist a trading strategy
which yields a non-null excess expected return with respect to any information set
contained in the market information Ft .
Proposition 7.1 Under the assumptions of Sect. 6.4, suppose furthermore that
agents are risk neutral. Let K D .Kt /t2N be an information flow contained in the
market information flow F D .Ft /t2N (i.e., Kt Ft , for all t 2 N) and such that
snt and dtn are Kt -measurable, for all n D 1; : : : ; N and t 2 N. Then the following
hold:
1
This phenomenon is particularly relevant in high frequency data and can be described by
relying on ARCH and GARCH models (see Akgiray [36], Pagan & Schwert [1392], Pagan
[1391], Campbell et al. [348] and Tsay [1601] for a textbook presentation of ARCH and GARCH
models in finance).
(i) returns are not serially correlated conditionally on the information K D

.Kt /t2N , i.e., Cov.rtCs ; rt jKk / D 0 for all k; t; s 2 N;
(ii) for any t 2 N, the value of P any Kt -measurable portfolio strategy tC1 2 RNC1 is
given by its market value NnD0 tC1 n
snt and the Kt -conditional expected excess
return of such a strategy is null.
The above proposition provides a theoretical foundation for the efficient market
hypothesis: under the assumption of risk neutral agents (or, under a risk neutral
probability measure), it is not possible to “beat the market” (excess return greater
than zero) by trading on the basis of any information set contained in the available
market information, since the (conditional) expected return of any portfolio is
equal to the risk free rate. Of course, since the risk free rate rf is assumed to be
deterministic, asset returns are not serially correlated if and only if the asset risk
premium is not correlated with past returns.
In view of the above observations (see in particular part (i) of Proposition 7.1), a
straightforward test of classical asset pricing theory consists in verifying the absence
of serial correlation in asset returns. However, Summers [1578] has shown that this
type of test has a rather low power against the alternative hypothesis of a persistent
autoregressive component in stock prices, so that the inability of such a test to reject
market efficiency does not provide a strong evidence for its validity.
More powerful tests have been proposed exploiting the linearity of the variance
with respect to the return time horizon. For instance, as suggested in Cochrane [457],
if the log-price is described by a first-difference stationary linear process, where the
error terms are independent and identically distributed with constant variance 2 ,
then the variance of the k-period return divided by k should be equal to the constant
term 2 (which also represents the variance of single period returns). Testing this
property leads to the so-called variance ratio tests, see Lo & MacKinlay [1233,
1234] and Campbell et al. [348]. Variance ratio tests are strictly related to serial
correlation tests: indeed, the variance of the k-period return divided by the variance
of the single period return is equal to one plus a positively weighted sum of return
autocorrelations (see Cochrane [457, Appendix A]).
Depending on the information set considered (corresponding to the information
flow K D .Kt /t2N considered in Proposition 7.1), the notion of market effi-
ciency takes three possible forms: weak market efficiency (Kt only contains the
information of past and current market prices), semi-strong market efficiency (Kt
contains all public information available at date t) and strong market efficiency (Kt
contains all possible information, including private information). In this section,
we shall limit our attention to the restrictions imposed by classical asset pricing
theory on the price-dividend processes and, hence, on the weak and semi-strong
forms of market efficiency, referring to Chap. 8 for a discussion of the role of
private information in financial markets. In particular, Proposition 7.1 implies that
future excess returns cannot be predicted on the basis of current and past returns
or on the basis of any information set Kt contained in the set Ft of available
market information (including public information such as macroeconomic factors
and accounting variables but excluding private information). According to the above
classification of market efficiency, this corresponds to the weak and semi-strong

forms of market efficiency.
As reported in the seminal paper Fama [661], the first empirical results provided
positive evidence in favor of the efficient market hypothesis. In particular, no serial
correlation was detected in the return time series. Fama [660] showed that first
order serial correlation of daily/monthly returns is positive, but in general only a
weak significance of serial correlation was detected. In the same years, negative
empirical evidence was only reported for the strong form of market efficiency. The
analysis of trading rules typically adopted by practitioners also provided support
for the efficient market hypothesis. For instance, Fama & Blume [665] analysed a
filter trading strategy, consisting in buying an asset when its price rises by a fixed
percentage and selling it when its price drops by the same percentage. Analysing
this type of trading strategy, it was shown that a filter strategy with a reference
percentage comprised between 0:5 and 1:5 was able to produce excess returns and
outperformed simple buy-and-hold strategies by trading on the basis of very short
term price swings. However, the presence of small transaction costs suffices to
eliminate the excess profits generated by such a strategy.
In the ’80s, the debate on market efficiency blazed up and many recurrent anoma-
lies were reported in the literature. Let us first discuss the so-called seasonality
effects, consisting in some calendar anomalies depending on the frequency of the
considered time series. In monthly time series, a January effect was observed:
stock returns are higher in January than in other months, especially in the case
of small capitalization stocks, see Rozeff & Kinney [1481], Roll [1458], Keim
[1079, 1080], Reinganum [1444]. The January effect was shown to be robust with
respect to different time horizons and different markets. A possible explanation
of the January effect comes from the turn-of-the-year tax related trading (tax
loss selling), see Constantinides [490], Poterba & Weisbenner [1431], Grinblatt &
Moskowitz [833]. On the other hand, on daily time series, a weekend effect was
observed: returns are on average negative from the closing of the trading activities
on Friday to the opening of trading activities on Monday, see French [739], Keim
& Stambaugh [1082], Jaffe & Westerfield [1008]. Moreover, returns are typically
higher on the day preceding a holiday, on the last day of a month and before the end
of a year (see Lakonishok & Smidt [1160]). Some of these phenomena are due to
a window dressing behavior by institutional investors: often, before reporting dates,
the composition of a portfolio is changed in order to improve the appearance of the
portfolio managed by the investor.
The results obtained in the ’70s providing positive evidence in support of return
unpredictability mostly considered short term returns. Afterwards, in the ’80s,
predictability patterns were reported in many studies based on returns over longer
time horizons. One of the first papers casting doubts about the efficient market
hypothesis in the finance community was De Bondt & Thaler [532], where the
authors showed that portfolios composed by extreme “winner” stocks over the past
three to five years (i.e., portfolios composed by stocks which exhibited in the past
three to five years a return higher than the market return) have a relatively poor
performance and, moreover, are outperformed by portfolios composed by extreme
“loser” stocks. The excess return difference between these two portfolios in three
years is around 25%: the average return on “winner” portfolios is about 5% less
than the market return, whereas the average return on “loser” portfolios is about
20% more than the market return. These results agree with CAPM anomalies
associated with size, earnings-price ratio and book-to-market ratio (see Sect. 5.3),
i.e., firms with high ratios tend to be losers in the past (value stocks, see De Bondt &
Thaler [533]). The authors attributed these results to market irrationality and, more
specifically, to an overreaction phenomenon: market participants tend to overreact
to unexpected and dramatic events and the Bayes rule is typically not applied in
practice since agents tend to overweight recent observations (see also Kahneman &
Tversky [1061] and Chap. 9). These results suggest that asset prices cannot be well
represented by a random walk process: there is a mean reversion effect and stock
prices may contain autoregressive components with a small decay rate.
In Summers [1578], it was shown that serial correlation tests based on short term
returns have a very low statistical power to discriminate the hypothesis that prices
depart from their fundamental value by exhibiting a temporary component with slow
decay. The model alternative to market efficiency proposed in Summers [1578] for
asset prices is formulated as follows:
st D st C ut ;
ut D ˛ut1 C vt ;
for all t 2 N, where st denotes the asset fundamental value (see expression (6.86))
and where 0 ˛ 1 (so that deviations from the fundamental value persist but
do not grow forever) and .vt /t2N is a sequence of independent random variables
with zero mean and constant variance. In this model, ˛ D 0 is the null hypothesis,
representing market efficiency (note also that this model well represents Shiller’s
suggestion according to which an asset price is given by the sum of the fundamental
value and of an autoregressive component describing investors’ irrationality). When
˛ is close to one (representing persistent price departures), it is unlikely that serial
correlation tests of monthly returns will reject the null hypothesis, without being
able to really discriminate the existence of the inefficient component ut (using daily
returns does not alter the conclusion). Furthermore, since it is difficult to detect
mean reversion phenomena in short horizon returns, results showing the absence of
serial correlation in returns should be interpreted with caution.
The debate on the presence of serial correlation in asset returns was animated
by the results obtained in Poterba & Summers [1430], Fama & French [666], Lo
& MacKinlay [1233]. In Poterba & Summers [1430], Fama & French [666], it has
been shown that returns computed on a horizon longer than one year are negatively
serially correlated, while in Lo & MacKinlay [1233] positive serial correlation was
detected for weekly and monthly returns. The methodology employed in Poterba &
Summers [1430], Lo & MacKinlay [1233] was based on a variance ratio test, while
in Fama & French [666] the test directly concerned serial correlation in the asset
return time series.
Elaborating on the model of Summers [1578], the following model has been
proposed in Fama & French [666] for the evolution of log-prices:
log st D qt C zt ;
qt D C qt1 C t ;
zt D zt1 C t ;
for all t 2 N, where .t /t2N and . t /t2N are two white noise sequences (i.e.,
sequences composed of independent random variables with zero mean). According
to the above model, the log-price is decomposed into the sum of two components:
a random walk component .qt /t2N , with the parameter representing a drift
parameter, and an auto-regressive component .zt /t2N with parameter < 1 (to
ensure stationarity). The stationary component .zt /t2N represents the deviations of
the log-price from the fundamental value, deviations which tend to disappear as time
goes on (mean reversion effect). The mean reversion of the stationary component
.zt /t2N induces a negative autocorrelation in log-returns. Note, however, that positive
serial correlation in log-returns cannot be generated by the model proposed in Fama
& French [666]. More precisely, if the asset log-price time series is generated
according to the above model, then the serial correlation of log-returns computed
for a small horizon is almost null and instead approaches the value 0:5 for large
time horizons. The authors tested this model by estimating the serial correlation
of log-returns of some asset portfolios (based on industry and size variables) in
correspondence of different time horizons in the period 1926–1985. While serial
correlation of one year returns is found to be negligible, negative serial correlation
is observed for horizons longer than one year. More specifically, plotting the values
of the estimated serial correlation coefficient in correspondence of increasing return
horizons reveals a U-shaped pattern, with the highest degree of correlation being
obtained in correspondence of a three to five years horizon. This pattern is consistent
with the hypothesis that log-prices have a slowly decaying stationary component,
while the random walk component dominates in long horizon returns. In the case
of three to five year returns, between 30% and 45% of the variance of log-returns
is generated by the stationary mean reverting component .zt /t2N . Moreover, return
predictability turns out to be more significant for small capitalization firms than
for large capitalization firms. However, the negative autocorrelation is shown to be
significantly weaker in the post-Second World War period and does not show the U-
shaped pattern in the overall 1926–1985 period. Since the sample size is relatively
small in the case of long horizon returns, the authors establish the significance of
their results by means of Monte Carlo simulations, mimicking the properties of
the observed log-returns time series. Fama & French [666] explain these empirical
results through time varying equilibrium expected returns generated by changing
investment opportunities and not through market irrationality. Similar results have
been obtained in Poterba & Summers [1430] by relying on a variance ratio test, with
the main difference being that the highest degree of mean reversion is observed for
a six to eight years horizon. Long horizon negative serial correlation is also detected
in Cutler et al. [510]. In particular, these studies show that the presence of serial
correlation in asset returns is not observed if one does only look at short term returns
as previously considered in many tests of the efficient market hypothesis.
The presence of serial correlation for short horizon returns turned out to be a
controversial topic. Indeed, French & Roll [741], Lo & MacKinlay [1233] show
that weekly and monthly (up to one year) asset returns do not exhibit (or exhibit
to a small extent) negative serial correlation, while significant negative correlation
was reported in Conrad & Kaul [483], Lehman [1177], Jegadeesh [1023]. In
particular, Lo & MacKinlay [1233] show that the random walk hypothesis is
strongly rejected on the basis of weekly returns on the period 1962–1985, using
aggregate return indexes as well as size-related portfolios (with more pronounced
effects in the case of small capitalization firms), and report significant positive
serial correlation for weekly and monthly returns (momentum effect, see also Lo
& MacKinlay [1237], Poterba & Summers [1430], Cutler et al. [510], Jegadeesh &
Titman [1026], Chan et al. [405]). In Lo & MacKinlay [1233], the first order serial
correlation coefficient of an equally weighted index is estimated to be approximately
30%, while, somehow surprisingly, the serial correlation of individual securities is
typically found to be negative. In the subsequent paper Lo & MacKinlay [1237],
using weekly data, it has been reported that the serial correlation of a small
size companies portfolio (the smallest quintile) is equal to 42%, while portfolios
composed by large companies exhibit a weaker serial correlation. The absence of
serial correlation in asset returns can be explained by the presence of significant
noise components in the single assets, while negative correlation may be due to a
lack of liquidity and to short term price pressure. It is also interesting to remark
that, as shown in Lo & MacKinlay [1237], a strong positive cross-covariance effect
between returns of individual stocks (capturing the situation where a high return on
a given stock today implies that the return on another stock will probably be high
tomorrow) can generate positive serial correlation in the returns of an index or of a
portfolio.
Conrad et al. [487] propose a solution to the short-horizon serial correlation
puzzle by assuming that security returns are generated by three independent com-
ponents: a positively autocorrelated common component (representing time varying
expected returns), a negatively autocorrelated idiosyncratic component (due to
microstructure effects such as transaction costs) and a white noise component. Using
weekly returns on an index, the authors show that the first two components allow to
explain about 24% of the variance of returns. At the level of individual securities,
market microstructure effects dominate, thus producing negative serial correlation,
while at the portfolio level returns are mostly driven by the common component
(since bid-ask errors are diversified away, being cross-sectionally uncorrelated),
thus leading to positive serial correlation. In this sense, the model of Conrad et al.
[487] is able to reproduce the regularities observed empirically in the return time
series. An alternative explanation for the presence of serial correlation in asset
returns is provided by non-synchronous trading: assets are exchanged in different
time periods and with a different frequency and, therefore, they incorporate new
information at different times. This fact can induce positive serial correlation in
stock index returns (in particular, equally weighted indexes). Mech [1316] shows
that transaction costs and bid-ask spreads may lead to serial correlation in portfolio
returns even if the returns of individual assets are serially uncorrelated. However,
Lo & MacKinlay [1233, 1236], Cutler et al. [510], Lo & MacKinlay [1237] show
that non-synchronous trading only accounts for a small part of the observed positive
serial correlation in portfolio returns. On the other hand, Boudoukh et al. [273] and
Ahn et al. [28] suggest that the effect of non-synchronous trading on small stocks
has been underestimated. An example of an equilibrium model generating negative
serial correlation in long horizon returns is provided in Exercise 7.7.
Momentum and Contrarian Strategies
Another way to empirically test the predictability of future asset returns consists in
evaluating the performance of suitable trading strategies. Indeed, in view of part (ii)
of Proposition 7.1 (under the assumption of risk neutrality), there exists no trading
strategy based on the available information yielding a non-null excess return on
average. Quoting Jensen (see Jensen [1031]): “a market is efficient with respect to
information set , if it is impossible to make economic profits by trading on the
basis of . By economic profits we mean the risk-adjusted rate of return, net of all
costs”. Since this result relies on the absence of serial correlation in asset returns,
the existence of serial correlation in asset returns might lead to trading strategies
yielding non-null excess returns.
In this perspective, market efficiency has been evaluated by assessing the excess
returns of trading strategies obtained from technical analysis methods (i.e., trading
strategies based on the analysis of past and recent movements in the returns time
series). In Brock et al. [308], it has been shown that technical trading rules based
on two moving averages and resistance-support levels generate returns that are not
consistent with the hypothesis of a random walk, of a first-order autoregressive
model or of a GARCH model (see Tsay [1601]), supporting the claim that technical
analysis has predictive power on future returns. Along the same lines, Lo et al.
[1238] provided evidence that classical technical analysis may have a predictive
power.
In De Bondt & Thaler [532], the performance of contrarian strategies has been
evaluated. Such strategies consists in forming portfolios with long positions on
the assets that have performed badly in a previous period (loser assets) and short
positions on the assets with a good performance in the previous period (winner
assets). It has been empirically observed that contrarian strategies produce excess
returns in the long run because of the agents’ overreaction (the profitability of such
strategies has been first observed by Graham & Dodd [818]). On a short horizon,
the existence of excess returns generated by contrarian strategies has been reported
in Jegadeesh [1023] and Lehman [1177].
Somehow in an opposite direction, it has also been observed that strategies
buying (selling, resp.) winner assets (loser assets, resp.) in the recent past generate
excess returns, exploiting short horizon trends in asset returns time series. Trading
strategies exploiting this phenomenon are called momentum strategies. The capabil-
ity of momentum strategies to generate excess returns over a medium time horizon
has been documented in Jegadeesh & Titman [1026] for the U.S. market: assets with
high returns over the past three-twelve months generate significant positive returns.
However, after twelve months, the performance tends to reverse (see Jegadeesh
& Titman [1028]). Excess returns are not due to changes in the riskiness of the
assets or to a delayed reaction of the asset price to common risk factors. The
momentum effect is stronger and persistent in small firms, growth firms (rather
than value firms), low trading volume firms, high volume markets, firms with low
analysts’ coverage, as documented in Rouwenhorst [1478], Moskowitz & Grinblatt
[1355], Hong et al. [958], Lee & Swaminathan [1175], Chan et al. [405]. In Conrad
& Kaul [486] it is shown that contrarian and momentum strategies are equally likely
to be successful: contrarian strategies earn profits in the long run (although the
profits are significant only in the 1926–1947 period), while momentum strategies
are usually profitable over a medium horizon (three-twelve months), confirming the
results reported above.
A Critical Assessment
The empirical results showing the presence of serially correlated asset returns,
the existence of excess returns generated by contrarian and momentum strategies
stimulated an intense debate on the serial correlation of asset returns. As in the
excess volatility debate, one can distinguish three main lines of thought on the
profitability of contrarian/momentum strategies: the existence of statistical pitfalls
in the test procedures, the existence of risk factors not taken into account by the
models being tested (coherently with the classical asset pricing theory) and, finally,
interpretations going beyond the classical asset pricing theory (behavioral finance).
Starting from the first of the above three lines of thought, the evidence of negative
serial correlation in asset returns has been criticized from a statistical point of view.
In particular, the sample size of long horizon returns can be very small when the
return horizon is large compared to the length of the time series, thus reducing the
statistical power of the test. Moreover, when working with long horizon returns,
overlapping observations typically occur. Due to these issues, variance ratio tests as
well as serial correlation tests have a rather low statistical power and are typically
biased towards the rejection of the classical random walk hypothesis, see Richardson
& Smith [1449], Richardson & Stock [1450], Kim et al. [1093]. In these studies,
taking into account the statistical issues discussed above, the results previously
reported in Fama & French [666] and Poterba & Summers [1430] have not been
confirmed and represent spurious deviations from the null hypothesis. Moreover, in
Richardson [1448] it has been shown that the U-shaped autocorrelation structure
reported in Fama & French [666] can also be generated when the log-price process
follows a random walk.
In Kim et al. [1093] and Jegadeesh [1024] it has been shown that mean reversion
is peculiar to a limited time period: indeed, after the Second World War, no mean
reversion is observed but rather persistence in long horizon returns (mean aversion).
In Jegadeesh [1024] it has been shown that an equally weighted index of stocks
exhibits mean reversion, but there is little evidence of mean reversion in the case
of a value-weighted index. Moreover, the mean reversion phenomenon is typically
concentrated in the month of January (January effect). Motivated by Fama & French
[666], Lamoureux & Zhou [1165] adopt a Bayesian methodology and focuses on
testing return predictability per se (as opposed to a specific null hypothesis that
implies the absence of serial correlation in asset returns), providing evidence in
favor of the random walk hypothesis. However, the evidence on the profitability of
contrarian/momentum strategies cannot be disregarded from a statistical point of
view. Indeed, even when taking into account most of the statistical issues discussed
above, Balvers et al. [127] report the existence of mean reversion in post-war data
in the case of national equity indexes of well-developed countries. Strong mean
reversion effects have been also detected in Daniel [515].
On a different ground, the existence of excess returns generated by short
horizon contrarian strategies (i.e., trading strategies exploiting weekly/monthly
mean reversion) may be due to order imbalance, lack of liquidity and price pressure,
as documented in Lehman [1177], Jegadeesh [1023]. Conrad & Kaul [485] have
shown that the long run excess returns generated by a contrarian strategy (computed
as cumulative short term returns) are typically overestimated due to the presence
of bid-ask spreads, price discreteness and non-synchronous trading. Taking into
account these effects and computing portfolio returns with a holding period of up
to three years, the excess returns generated by a contrarian strategy can be even
negative. Moreover, as reported above, contrarian profits are in general associated
with the existence of a January effect, see also Zarowin [1674], De Bondt & Thaler
[533] (however, a different interpretation is proposed in Loughran & Ritter [1247]).
Remaining within the context of classical asset pricing theory, the anomalies
observed in asset returns are typically explained in terms of time varying expected
returns. For instance, serially correlated mean reverting expected returns have been
considered in Fama & French [666], Ball & Kothari [124], Chan [399], Fama
[662], Kothari & Shanken [1124], Berk et al. [199]. In Cecchetti et al. [378], the
authors propose an intertemporal general equilibrium model producing negative
serial correlation in long horizon returns. Agents are characterized by constant
relative risk aversion and the growth rate of the endowment follows a Markov
switching process (see Exercise 7.7 for more details). An intertemporal equilibrium
model producing similar results has been also proposed in Kandel & Stambaugh
[1069]. In Conrad & Kaul [483], the following model allowing for time varying
expected returns has been tested:
rt D EŒrt jFt1 C t ;
EŒrt jFt1 D C EŒrt1 jFt2 C ut1 ;
for all t 2 N, where 1 and . t /t2N and .ut /t2N are two white noise sequences.
The expected return is evaluated by means of a Kalman filtering technique and the
data lead to a rejection of the hypothesis of constant expected returns. Moreover, the
empirical analysis shows that this model fits well the time series of weekly returns.
The variation over time of expected returns explains up to 26% of the variance of
the returns of a portfolio of stocks of small firms (this percentage decreases when
considering a portfolio composed by stocks of large firms). Similar results were
obtained in Conrad & Kaul [484] in the case of monthly returns.
Leaving the realm of classical asset pricing theory, De Bondt & Thaler [532, 533]
suggest that the existence of long run mean reversion phenomena is due to the
agents’ overreaction to recent observations: loser assets are excessively underesti-
mated by non-rational agents due to a poor recent performance while winner assets
are excessively overestimated due to a very good recent performance (overreaction).
Lehman [1177] supports this behavioral interpretation. This interpretation of the
excess returns generated by contrarian strategies is similar to the “irrational” expla-
nation of the CAPM anomalies presented in Sect. 5.3. An alternative interpretation
has been proposed in Ball & Kothari [124] and Chan [399]: considering the CAPM
as the reference model, the authors observe that negative serial correlation in returns
can be induced by time varying expected returns generated in turn by time varying
expected returns on the market portfolio. Hence, expected returns can change over
time, the ˇ coefficients are correlated with the market portfolio expected return and
change over time. In the period comprised between the date of portfolio formation
and the date of performance evaluation, the expected returns and the ˇ coefficients
of winner assets typically decrease while the expected returns and the ˇ coefficients
of loser assets increase. This implies that winner assets will be less risky while
loser assets will be riskier with respect to the date of portfolio formation. One of
the main causes of changes in riskiness is asset leverage. Indeed, since leverage is
decreasing with respect to past asset returns, a negative series of abnormal returns
will increase leverage and, consequently, the ˇ coefficient (typically, the opposite
occurs for winner assets). It has been observed that the ˇ coefficient of loser assets
typically exceeds the ˇ coefficient of winner assets by 0:62 following the date of
portfolio formation. By allowing for time varying expected returns in an equilibrium
model (e.g., the CAPM model extended with time varying ˇ coefficients), it can be
shown that contrarian strategies do not produce significant excess returns. Similarly,
in Zarowin [1673, 1674] no sign of overreaction is detected and it is shown that the
profits generated by contrarian strategies are mainly due to a size effect, which is not
captured by the ˇ coefficient. Indeed, while loser assets typically outperform winner
assets, loser assets are also typically smaller than winner assets. When loser assets
are matched with winner assets of equal size, then there is little statistical evidence
of a different performance. Fama & French [673] and Brennan et al. [293] show that
the multi-factor model proposed in Fama & French [670] captures contrarian profits
but not momentum profits. Korajczyk & Sadka [1120] show that momentum profits
persist even when transaction costs are taken into account.
Chopra et al. [431] address the robustness of the profitability of contrarian strate-
gies with respect to the above criticisms. By allowing for changing ˇ coefficients
of winner and loser assets, taking into account the existence of a size effect and
considering January and non-January strategies, they still find evidence of non-null
excess returns generated by contrarian strategies (5–10% in excess return per year in
the years following the portfolio formation). This phenomenon is more pronounced
for small firms rather than for large firms and, as already mentioned above, there
is a strong January seasonality pattern. Similarly to De Bondt & Thaler [532],
they conclude that the excess profits of contrarian strategies are mainly due to the
presence of non-rational agents overreacting to news together with the fact that small
firms are more frequently owned by individuals. Lehman [1177] reached a similar
conclusion, analysing contrarian strategies over a weekly time horizon.
Summing up, we can remark that the explanations reported above for the excess
profits generated by contrarian/momentum strategies resemble the explanations
proposed for the CAPM anomalies, namely (besides the issues related to the
statistical methodologies) the incorrect specification of the asset pricing model (see,
e.g., Fama & French [673]) and the presence of irrational agents in the market (see
Lakonishok et al. [1159]). Hence, some of the arguments presented in Sect. 5.3
about the CAPM anomalies can also be applied to the present context. Assuming
a behavioral finance perspective, the presence of negative serial correlation (mean
reversion) and excess profits generated by contrarian strategies are often referred to
as overreaction of asset prices to news. On the other hand, positive serial correlation
and the momentum effect are typically referred to as underreaction (or delayed
reaction of asset prices to news). Both phenomena, at different time horizons, are
simultaneously present in financial markets.
Adopting a behavioral finance point of view (see Lakonishok et al. [1159], La
Porta et al. [1168]), the excess profits generated by contrarian strategies can be
traced back to the presence of irrational agents in the market. Such agents are
irrational in the sense that they form expectations about future prices by extrapo-
lating recent earnings growth rates, giving an excessive weight to recent news about
earnings growth rates. In this regard, La Porta [1167] and Dechow & Sloan [536]
show that the analysts’ forecasts of future earnings growth rates are significantly
and systematically biased in the direction of overreaction (in other words, analysts’
forecasts are typically extreme). Exploiting the errors in the analysts’ forecasts to
build a contrarian strategy, excess returns are obtained. Since analysts’ forecasts
can be regarded as a relatively good proxy of the agents’ forecasts, this evidence
of overreaction can explain a part of the excess returns generated by contrarian
strategies (and there is no evidence of low return stocks being riskier than high
return stocks). Long term overreaction in analysts’ recommendations has also been
detected in La Porta [1167], Dechow & Sloan [536], De Bondt & Thaler [534], while
short horizon underreaction has been reported in Abarbanell & Bernard [1], Chan
et al. [407], Michaely & Womack [1339]. The presence of heterogeneous types of
agents (feedback and noise traders) in the market has also been used to explain
the negative/positive serial correlation in asset returns (see Cutler et al. [510] and
Chap. 9).
A severe criticism of behavioral finance came from Fama [664]. The author
pointed out that underreaction is equally frequent as overreaction and, moreover,
long term anomalies are sensitive to the factor model being tested as well as to
the statistical methodology being employed. Typically, the presence of asset pricing
anomalies is not robust to changes in the model or in the statistical methodology
and, hence, if the model and the testing procedures are well-chosen, no anomaly
should be detected. Almost all the empirical studies on market anomalies test the
hypothesis of market efficiency without clearly specifying an alternative hypothesis
(i.e., without satisfying the Khun methodological approach). In other words, market
efficiency can only be replaced by a better specific model of price formation,
while in most of the empirical studies the alternative hypothesis is just vaguely
meant to represent market inefficiency. Moreover, many of the models proposed by
the behavioral finance literature manage to explain a specific anomaly but predict
other facts that are not always empirically confirmed in real financial markets. The
evaluation of Fama [664] of the empirical literature is in favor of the efficient market
hypothesis, potentially allowing for time varying expected returns. The author
suggests that “the expected value of abnormal returns is zero, but chance generates
apparent anomalies that split randomly between overreaction and underreaction”
(see Fama [664]).
Lo & MacKinlay [1237] investigate whether the profitability of contrarian
strategies does provide evidence of overreaction. The authors give a negative answer
if there are sufficiently many traded securities in the market: the returns generated
by a contrarian strategy can be attributed to a lead-lag relation among assets
(covariance across stocks) and not to overreaction (or mean reversion) phenomena.
They attribute over one half of the expected profits generated by a contrarian
strategy to such cross correlation effects and not to the negative serial correlation
in the individual stocks. Hence, the profitability of contrarian strategies does not
necessarily imply overreaction. The authors report a positive correlation (about
28%) between weekly returns of small size companies and lagged returns of large
size companies with a lead-lag relation (so that the returns of large capitalization
stocks almost always lead those of smaller stocks, but not vice versa). This effect
also generates positive serial correlation in weekly return indexes together with
weak asset returns serial correlation, as documented in Lo & MacKinlay [1233].
The phenomenon can also be explained in terms of a different diffusion of the
information concerning common risk factors: a quick diffusion of information for
large size companies and a slow diffusion of information for small size companies
(see McQueen et al. [1315]). In particular, there is evidence of a slow response by
some small stocks to good, but not to bad, common news. In Jegadeesh & Titman
[1027] it is shown that stock prices react with a delay to common factors while
overreact to firm-specific factors: this difference generates a size related lead-lag
relation in stock returns (see also Hou [967] on the existence of lead-lag effects).
However, in contrast with the above observations, contrarian profits are mainly
due to overreaction to firm-specific information and not to delayed reaction to
common factors. Lead-lag response to common factors is also a source of profits
of momentum strategies (see Lewellen [1210]).
The existence of a lead-lag relation between the returns on stocks of large size
firms and of small size firms suggests to build a trading strategy which suitably
buys and sells portfolios of large and small companies. In Knez & Ready [1106],
it has been shown that a strategy of this type (switching between portfolios of
small and large firms on the basis of the returns observed in the previous week)
generates excess annual returns of around 15%. However, effective spreads cancel
these gains and, once spreads are appropriately taken into account, the strategy will
be outperformed by simple buy-and-hold strategies. In Jegadeesh & Titman [1027] it
is shown that a contrarian strategy applied to size-sorted portfolios does not generate
abnormal returns.
A risk-based explanation of the profitability of momentum strategies has been
proposed in Conrad & Kaul [486] and Berk et al. [199], in the sense that the
profits generated by momentum strategies are not abnormal returns but rather the
compensation for bearing systematic risks changing in predictable ways. More
specifically, Conrad & Kaul [486] observe that a momentum strategy gains from
any cross-sectional dispersion in the unconditional mean returns, while Chordia &
Shivakumar [436] and Berk et al. [199] explain the profits generated by momentum
strategies through cross-sectional variations in time varying (conditional) expected
returns and ˇ coefficients driven by macroeconomic variables related to the
business cycle (rather than stock-specific factors). Moreover, Conrad et al. [487]
report momentum effects in common factors affecting stock returns. As shown in
Moskowitz & Grinblatt [1355], the presence of a momentum effect related to a
specific industry can explain the profits of momentum strategies, in line with the
cross-sectional correlation explanation of momentum profitability. A risk-adjusted
analysis of momentum strategies based on industry factors can explain a large part
of momentum profits (around 50%, see Ahn et al. [29]). In addition, Ang et al. [67]
and Harvey & Siddique [910] have shown that returns on momentum strategies are
related to their sensitivity (high exposure) to downside risk.
In Fama & French [673], Brennan et al. [293], Grundy & Martin [850], it has
been observed that the profitability of momentum strategies cannot be explained by
mean reversion, size and book-to-market effects. Grundy & Martin [850] provide
evidence against risk-based explanations, observing that momentum profitability
reflects momentum in the stock-specific components of returns which are not
associated with risk factors. Jegadeesh & Titman [1029] show that the explanation of
momentum profits through cross-sectional differences in expected returns provided
in Conrad & Kaul [486] is affected by small sample biases: taking this issue
into account, then no evidence in support of the cross-sectional interpretation is
found. Chan et al. [407], Jegadeesh & Titman [1026, 1028], Hvidkjaer [995], Hong
et al. [958], Grundy & Martin [850], Moskowitz [1354], Hvidkjaer [995], Lee &
Swaminathan [1175] report positive evidence that momentum strategy profits are
due to initial market underreaction/overreaction followed by a delayed reaction to
firm-specific news (in particular to earnings news). Jegadeesh & Titman [1028], Lee
& Swaminathan [1175] find reversals in momentum portfolio returns after one
year, a result that contrasts with a risk-based explanation and is in favor of the
delayed reaction explanation. Underreaction to firm-specific information (slow
diffusion of information) as the origin of the momentum effect is supported by

the observed negative relation between the momentum effect and the number of
analysts following the asset (see Hong et al. [958]). In Lewellen [1210], no sign
of underreaction to news or of positive serial correlation in single asset returns is
detected and the main source of momentum profits is identified as the negative auto-
and cross-sectional correlation (lead-lag effect).
Predictability and Event Studies
The limited statistical power of tests based on univariate returns led many
researchers to study the capability of specific firm-related and macroeconomic
variables to predict future returns. Recall that, if the asset price follows a random
walk, then no subset of the available information set Ft allows to predict future
returns (however, we want to stress that return predictability is compatible
with classical asset pricing theory if expected returns are allowed to be time
varying). As far as monetary and macroeconomic variables are concerned, the
variables that turn out to have some predictive power are the (expected) inflation,
monetary growth, short term interest rates, term spread, default spread, changes
in industrial production, output, aggregate consumption-wealth ratio, aggregate
wealth, labor income-consumption ratio, market liquidity, bid-ask spread, equity
issues proportion of new securities issues and volatility (see Balvers et al.
[126], Fama & Schwert [679], Chen [421], Keim & Stambaugh [1083], Campbell
[330], Fama & French [667, 668], French et al. [742], Cutler et al. [510], Marshall
[1308], Patelis [1410], Pesaran & Timmermann [1415], Lettau & Ludvigson
[1194], Santos & Veronesi [1497], Jones [1039], Baker & Wurgler [109], Flannery
& Protopapadakis [713], Bollerslev et al. [265]).
In Rozeff [1480], Shiller [1535], Campbell & Shiller [352], Kothari & Shanken
[1125], it has been documented that the aggregate dividend-price ratio (dividend
yield), the earnings-price ratio and the book-to-market value, respectively, have a
predictive power on future market returns (in particular, Shiller [1535] shows that
stock prices appear to overreact to dividends). These results have been confirmed
in many papers (see Fama & French [667], Campbell & Shiller [351], Fama &
French [668], Cutler et al. [510], Hodrick [951], Flood et al. [720], Campbell &
Shiller [353], Pontiff & Schall [1427], Campbell [334], Lewellen [1209, 1211]). A
theoretical foundation of the predictability of future returns through the dividend
yield can be found in the dividend growth model (7.14). In general, the price is low
when expected returns are high and, therefore, high dividend yields typically predict
high returns. The predictability of future returns through these variables increases
in the return horizon: in correspondence of a four year period, the dividend-
price ratio explains about 25% of the return variance (see Fama & French [667]).
Note that it is easy to reconcile this evidence with the presence of negative serial
correlation in asset returns: indeed, the predictability (with a positive coefficient)

of future returns through the current dividend yield provides an evidence of mean
reversion in the return time series. Weak predictability in the ’90s and a poor out of
sample performance (due to parameter instability) have been detected in Cochrane
[463], Goyal & Welch [817], Schwert [1510]. In Lamont [1162] it is shown that also
the aggregate dividend-earnings ratio predicts future returns, in the sense that high
dividends forecast high returns and high earnings forecast low returns. Furthermore,
accounting variables that are useful to explain asset returns cross-sectionally are also
useful at the aggregate level to predict market returns.
As we have already pointed out in the case of other market anomalies, the
empirical evidence on return predictability through the dividend yield and other
fundamental variables can be interpreted within or outside the classical asset pricing
theory. In the first case (see, e.g., Rozeff [1480], Fama & French [667, 668], Fama
[663], Cochrane [464], Pontiff & Schall [1427]), dividends and earnings are
regarded as fundamental variables. For instance, according to Rozeff [1480],
changes in dividend yield proxy for variations in the risk premium of stocks. An
equilibrium monetary asset pricing model relating inflation to returns has been
provided in Marshall [1308], while an intertemporal equilibrium model relating
asset returns to macroeconomic fluctuations (output) is provided in Balvers et al.
[126]. On the other side, return predictability can be interpreted as an evidence
of the existence of noise traders and market irrationality (see, e.g., Shiller [1535],
Campbell & Shiller [351], Cutler et al. [510]). According to this interpretation, low
dividend yields are associated with overvalued stocks and, therefore, with lower
future returns on stocks.
The existence of underreaction/delayed reaction has also been studied in cor-
respondence of specific events (event studies), see Daniel et al. [516], Fama
[664], Hirshleifer [945], Kothari [1123], MacKinlay [1270]. In particular, in
correspondence of some public events, a post-event drift has been observed, in
the sense that the average return on the date of the event has the same sign of
the subsequent average long run abnormal performance, up to a three to five years
period. Such events typically include earning announcements/surprises (see Ball &
Brown [123], Bernard & Thomas [202], Jegadeesh & Titman [1026], Chan et al.
[407]), dividend initiations and omissions (see Michaely et al. [1336]), seasoned
issues of common stocks and initial public offerings (IPO) (see Ibbotson & Ritter
[996]) and stock market repurchases (see Ikenberry et al. [998]). In particular, a post-
earnings drift is observed: stocks with surprisingly good news tend to outperform
those with bad news. Analysts’ earnings forecasts and stock recommendations
underreact to earnings news, see Abarbanell & Bernard [1], Womack [1662],
Michaely & Womack [1339], Chan et al. [407]. As we have already pointed out,
stock underreaction can be related to analysts’ underreaction, but this only provides
a partial explanation (and the analysts’ behavior may also be unrelated to the stock
price overreaction).
7.3 Tests on Intertemporal Equilibrium Models
The intertemporal general equilibrium models presented in Chap. 6 have been exten-
sively empirically tested in many different forms. The literature on the empirical
tests of multi-period asset pricing theory is extremely rich and we refer to Singleton
[1549], Ferson [687], Ferson & Jagannathan [696], Kocherlakota [1112], Campbell
[334, 335, 336] for good surveys. In broad terms, we can distinguish three main
classes of empirical tests:
1. tests verifying the relationship between asset returns and the intertemporal rate
of substitution of consumption of the representative agent of the economy
(stochastic discount factor);
2. tests on intertemporal risk premium factor models (CCAPM, ICAPM);
3. calibration studies: the ingredients of an equilibrium model (notably the param-
eters describing the preference structure and the technology) are chosen in order
to match some empirical properties of asset returns.
Before starting our survey of the empirical literature we want to point out that a
large part of the empirical literature on multi-period asset pricing models is based
on the assumption of a time additive utility function of the power form u.x/ D
x1˛ =.1 ˛/. However, this type of utility function is affected by a significant
constraint: the parameter ˛ represents both the coefficient of relative risk aversion
as well as the reciprocal of the elasticity of the intertemporal rate of substitution (see
Hall [880] and Exercise 7.8). The identification through a single parameter of the
willingness of an agent to diversify wealth over different states of the world and to
substitute wealth intertemporally is not justified on an empirical basis (for instance,
in Barsky et al. [169] it has been shown that these two features are essentially
unrelated across individuals). In view of empirically testing the implications of
classical asset pricing theory, this represents a severe limitation. Considering a non-
power time additive utility function, the identity between the coefficient of relative
risk aversion and the reciprocal of the elasticity of intertemporal substitution holds
in an approximate form (see Exercise 7.8). The time additivity of the utility function
implies that agents tend to dislike variations in their consumption stream and
transfer wealth from periods of high consumption to periods of low consumption
in order to smooth consumption over time, similarly as risk averse agents try to
diversify consumption across different states of the world. In the remaining part of
this chapter, we shall restrict our attention to time additive utility functions, referring
to Chap. 9 for a discussion of alternative preference functionals.
Tests on the Euler Conditions
Considering a representative agent economy, a possible way to test the validity of

classical asset pricing theory consists in verifying empirically the validity of the
7.3 Tests on Intertemporal Equilibrium Models 379
optimality conditions (6.68)–(6.71) (Euler conditions). These optimality conditions

can be seen as (conditional) orthogonality conditions between the representative
agent’s one-period stochastic discount factor and the asset excess returns. However,
such relations depend non-linearly on the state variables (consumption/endowment)
and on the unknown preference structure. To deal with this issue, a generalized
instrumental variable estimation methodology has been developed in Hansen &
Singleton [895]. Assuming a log-normal stationary joint distribution of consumption
and monthly returns and a power utility function, the Euler condition (6.78) has been
empirically rejected. Furthermore, excluding the case of value-weighted portfolios,
the empirical evidence reported in Hansen & Singleton [896] is against the validity
of the Euler conditions. The estimated coefficient of relative risk aversion belongs
to the interval .0; 2/.
A semi-nonparametric test performed in Gallant & Tauchen [753] rejects the joint
hypothesis that consumption and returns follow a multivariate normal distribution
and that the utility function is time additive. The empirical evidence is in favor
of the presence of durable goods among the arguments of the utility function
(see Sect. 9.2), in the sense that there exist assets yielding an utility flow which
extends beyond the date of acquisition of the asset, thus contradicting the time
separability of the utility function. Evidence against a time separable utility function
has been also provided in Grossman et al. [845]: the authors suggest to consider a
utility function representing risk aversion and elasticity of intertemporal substitution
through different parameters (compare with Exercise 7.8). Note also that errors
in the measurement of consumption do not provide a sufficient explanation for
the negative results of these tests on classical asset pricing theory (see Singleton
[1549]).
A non-parametric methodology to test asset pricing models has been developed
in Hansen & Jagannathan [891] on the basis of the Hansen-Jagannathan bound
established in Proposition (6.36) (see also Cochrane & Hansen [467], Cecchetti
et al. [380], Gallant et al. [750], Balduzzi & Kallal [117], Hansen & Jagannathan
[892], Ferson & Siegel [698], Lettau & Uhlig [1199], Bekaert & Liu [188] and
Hansen et al. [890] for an extension to an economy with transaction costs and
market imperfections). In particular, note that a test of the Hansen-Jagannathan
bound does not require precise assumptions on the utility functions and, hence,
on the intertemporal rate of substitution of the representative agent. According
to Proposition 6.36, the excess return of every asset (normalized by its standard
deviation) should be bounded from above by the standard deviation-mean ratio
of the one-period stochastic discount factor. However, it has been pointed out in
Ferson & Siegel [699] that this type of test (in particular in an unconditional
form) is typically biased towards the rejection of the validity of the bound. The
results reported in Balduzzi & Kallal [117], Hansen & Jagannathan [891], Cecchetti
et al. [380], Cochrane & Hansen [467] show that the stochastic discount factor is
insufficiently volatile compared to stock returns for the bound to be satisfied, thus
providing empirical evidence against the validity of an intertemporal equilibrium
model with a time additive utility function. In particular, it has been shown that
an intertemporal equilibrium model with a time separable power utility function
only agrees with the data for a very high coefficient of relative risk aversion.
Intertemporal complementarity in preferences (habit formation, see Chap. 9) may
help to satisfy the bound.
Tests on the CCAPM and on the ICAPM
A second approach (which can however be related to the previous one) to test
the implications of classical asset pricing theory consists in empirically testing
the validity of the CCAPM. An empirical test of the CCAPM with positive
results has been implemented in Breeden et al. [286]. After adjusting reported
consumption data for aggregation and measurement errors, the authors have tested
cross-sectionally that expected returns are linear in the ˇ coefficients with respect to
the return of a portfolio which is maximally correlated with the consumption process
(compare with Proposition 6.37). The risk premium of such a portfolio is found to
be positive and the relation between the expected returns of the assets and of the
portfolio is approximately linear, especially if one does not include the 1929–1939
period. Consumption aggregation and measurement errors decrease the variance of
measured consumption growth and lead to an underestimation of the covariance
with asset returns, thus inducing a bias towards the rejection of the CCAPM. The
results of Breeden et al. [286] have been confirmed in Wheatley [1659] as well as
in Cecchetti & Mark [382], assuming a very high risk aversion coefficient. These
results are mainly based on seasonally adjusted data, which could in principle lead
to biases and erroneous inferences. However, even using unadjusted data, Ferson
& Harvey [692] still obtain a rejection of the model. Again, an implausibly high
coefficient of risk aversion is required to fit the data, as already observed in the case
of tests based on the Euler conditions. Bansal et al. [134] propose an intertemporal
equilibrium model explaining cross-sectional asset returns through the exposure of
asset dividends to aggregate consumption. The model explains more than 50% of
the cross-sectional variation in risk premia across different assets (momentum, size
and book-to-market sorted portfolios). The logarithm of the aggregate consumption
is modeled through an ARIMA(1,1,1) process (see Hamilton [881]) with a small
predictable component in the growth rates.
In this family of tests, the measurement of the aggregate consumption has a
crucial importance. In Jagannathan & Wang [1013], consumption betas of stocks
have been computed using yearly consumption growth rates (based on the fourth
quarter of the year). With this methodology, the CCAPM is shown to explain
the cross-section of stock returns similarly as the three-factor Fama & French
[670] model. Positive evidence on the CCAPM in explaining the cross-section of
expected returns has been also obtained in Parker & Julliard [1401], considering
the covariance with respect to consumption growth cumulated over many quarters
following the return period.
A conditional version of the CCAPM explains the cross-section of average
returns on size and book-to-market sorted portfolios when the consumption-wealth
ratio, idiosyncratic consumption risk or the consumption/labor income ratio are
used as conditioning variables, see Lettau & Ludvigson [1195], Jacobs & Wang
[1006] and Santos & Veronesi [1497], respectively. These variables summarize the
investors’ expectations about asset returns. Note that the use of the conditioning
technique improves the fit of the model because some stocks are more highly
correlated with consumption growth in bad times than in good times. Clearly, this
effect cannot be captured by unconditional models, since the latter assume constant
risk premia.
The CCAPM has been tested against the CAPM, i.e., the hypothesis that asset
expected returns are proportional to the consumption ˇ has been tested against the
hypothesis that they are proportional to the ˇ coefficient with respect to a market
portfolio. The empirical results are mostly in favor of the CAPM. In particular,
Mankiw & Shapiro [1297], Attanasio [80], Campbell & Cochrane [344] have shown
that the CAPM has a better explanatory power than the CCAPM when considering
cross-sectional average returns. In a similar direction, Chen et al. [424] have shown
that the multi-factor APT performs better than the CCAPM in explaining returns
cross-sectionally. The main reasons for the empirical failure of the CCAPM are
represented by the lack of variability in the intertemporal rate of substitution
of consumption (which corresponds to the one-period stochastic discount factor)
and the lack of covariance between consumption growth and asset returns. This
limitation has been partly addressed by considering conditional versions of the
CCAPM. Note also that these tests of the CCAPM are often affected by a joint
hypothesis problem, since the rejection of the CCAPM can also be attributed to a
failure of one of the many auxiliary assumptions needed for its specification.
In the previous section, we have reported many empirical results showing that
asset returns have a predictable component. As we have already mentioned, in
an efficient market the predictability of returns can be generated by time varying
investment opportunities and, hence, this feature should be captured by an intertem-
poral general equilibrium asset pricing model. Balvers et al. [126] have proposed
a general equilibrium model where the returns are related to macroeconomic
fluctuations. Asset returns depend on consumption which is in turn linked to
aggregate output. As a consequence, asset returns can be predicted if the output itself
is predictable. Consistently with conventional macroeconomic models, the output is
typically serially correlated and, hence, predictable. An intertemporal asset pricing
model with fluctuating mean and variance of consumption growth generating mean
reversion in asset returns has also been considered in Kandel & Stambaugh [1068]
and Cecchetti et al. [378]. Both consumption growth moments and risk premia
exhibit business cycle effects. Ferson & Harvey [691] employ a macroeconomic
multi-factor model (ICAPM) of the type described in Chen et al. [424] to capture
the time varying nature of returns and show that the stock market is the most
important factor in order to capture predictable variations of stock portfolio returns.

Moreover, the variation over time in the premium for the ˇ-risk is more important
than changes in the ˇ coefficient itself. Ferson & Korajczyk [695] have shown that
a macroeconomic or a statistical multi-factor model captures the predictability of
long horizon returns, allowing for time varying beta and risk premia. It is shown
that variations in risk premia are the primary source of predictability and five factors
suffice to explain about 80% of the variability of asset returns. An analogous task is
accomplished in Evans [658] with a two-factor model (returns on the stock market
and on corporate bonds), time varying ˇ and risk premia. Allowing for time varying
first and second moments, Attanasio [80] has shown that the traditional CAPM
accounts for predictability of returns through the dividend yield (mixed results have
been instead obtained in Kirby [1103]). Huberman & Kandel [983] provide positive
evidence on the capability of an intertemporal equilibrium model to capture return
predictability through time varying returns. Furthermore, Ferson [685] shows that
conditional ˇ changes are associated with interest rates, providing an explanation of
return predictability through interest rates.
Avramov [92] reported substantial deviations from the capability of the three-
factor model of Fama & French [670] to explain asset return predictability as well
as from the model based on firm characteristics proposed by Daniel & Titman [520].
Ferson & Harvey [693] have shown that predetermined variables used to predict
the time series of stock and bond returns also provide cross-sectional explanatory
power for stock returns on top of the three factors used in Fama & French [670]. The
conditional three-factor model is not able to capture common dynamics in returns
(e.g. those captured by the variables employed in Fama & French [667]).
The Equity Premium and Other Asset Pricing Puzzles
As we have seen in the preceding sections, the empirical literature has reported
several anomalies which apparently conflict with classical asset pricing theory. In
particular, let us emphasize the following striking empirical facts (see Campbell
[334]):
• the average asset return is rather high (in the postwar period, the average real
annual stock return has been 7.6%);
• the average risk free rate is rather small (in the postwar period, the three-month
Treasury bills average rate has been 0.8% per year);
• the time series of aggregate consumption exhibits a rather low variability (the
standard deviation of the consumption growth rate in the United States is around
1%);
• the volatility of the interest rate is rather small (the standard deviation of the real
return on U.S. Treasury bills is 1.8%);
• the correlation between consumption growth and real asset returns is rather weak
(0:22 on the basis of U.S. quarterly data).
In particular, it is not easy to explain within the context of classical asset

pricing theory a large equity premium (equity premium puzzle, see Mehra &
Prescott [1319]), a low risk free rate (risk free rate puzzle, see Weil [1648]) and
a low variability of consumption growth (consumption smoothing puzzle, see Hall
[880], Campbell & Deaton [345]). We refer to Kocherlakota [1112], Mehra &
Prescott [1320], Campbell [336], De Long & Magin [546] for an evaluation of these
puzzles. As shown also in cross-sectional tests, at the origin of these puzzles lies the
fundamental observation that the time series of consumption growth is too smooth
if compared with income growth and asset returns. Note that, in view of relation
(6.71), it holds that
1 0
EŒu0 .et /jFt1 D u .et1 /; for all t 2 N;
ırf
so that the quantity u0 .et / u0 .et1 /=.ırf / is orthogonal to all the information
available at date t 1, for every t 2 N. In particular, assuming a quadratic utility
function of the form u.x/ D ax bx2 =2 and letting ı D 1=rf , we obtain the linear
model
et D et1 C "t ; for all t 2 N;
where ."t /t2N is a sequence of uncorrelated random variables with zero mean.
As we have already remarked, many specifications of classical asset pricing the-
ory have considered utility functions of the power form, for which the coefficient of
relative risk aversion coincides with the reciprocal of the elasticity of intertemporal
substitution of consumption (see Exercise 7.8). Assuming a log-normal distribution,
the results presented after Proposition 6.38 imply that the equity premium is
increasing in the coefficient of relative risk aversion (assuming that the covariance
term is positive), while the relation between the coefficient of relative risk aversion
and the equilibrium risk free interest rate depends on both the average consumption
growth rate and the variance of the consumption growth rate. This shows that, in the
case of a power utility function, both the equity premium puzzle and the risk free
rate puzzle are related through the coefficient of relative risk aversion. A large risk
aversion coefficient, as required to match the observed values of the equity premium,
would imply a strong preference for consumption smoothing, meaning that the agent
dislikes fluctuations in consumption and is unwilling to substitute consumption
over time. On the other hand, since the variance of the consumption growth rate is
typically small, a large value of the coefficient of relative risk aversion would imply
unrealistically large values for the risk free rate. This shows the difficulty of solving
at the same time the equity premium and the risk free rate puzzles in this setting.
Note also that, as shown in Exercise 7.8, the inverse relation between the coefficient
of relative risk aversion and the elasticity of intertemporal substitution holds in
an approximate form for any time additive utility function. As a consequence, the
difficulty of solving the two puzzles together does not only depend on the choice of
a power utility function.
The estimate of the equity premium reported above, of course depending on
the estimation methodology as well as on the underlying time series, has been
criticized from a methodological point of view. In Mehra & Prescott [1319], the risk
premium has been estimated as the average historical return, so that the precision
of the estimate clearly depends on the length of the time series. However, it has
to be noted that long time series are also more sensitive to structural breaks (for
instance, the 1929 crisis). Indeed, Cogley & Sargent [470] have shown that the large
equity premium estimate can be attributed to the effects of the great depression:
even assuming that agents update their beliefs according to the Bayes rule, the great
depression induced pessimism that persisted over time inducing agents to ask for
a significant risk premium. In this direction, incorporating structural breaks in the
time series, Pástor & Stambaugh [1406] have estimated a risk premium comprised
between 4% and 6% percent in the last two centuries, with a sharp decline in
the last half century. In Fornari [726], using a conditional variance model, the
equity premium has been estimated around 5–6%, with a countercyclical behavior.
Moreover, computing the equity premium by an arithmetic average leads to an
overestimation when asset returns are mean reverting and noisy, as shown in Siegel
[1542]. Evidence of an overestimation of stock returns was also reported in Siegel
[1543] when computing the equity premium as the historical average over a long
time series, together with an underestimation of fixed income returns. In more recent
years, the equity premium seems to have declined (see for instance Mehra & Prescott
[1320] for more recent estimates).
To take into account these effects, model-based estimates of the risk premium
have been proposed. In general, these methods still report a positive equity premium,
but typically smaller than the estimate produced on the basis of average historical
returns (see, e.g., Blanchard [247]). A decline of the equity premium in recent
years has been established in Fama & French [675] and Jagannathan et al. [1011]
by relying on a dividend-earnings growth model (Gordon-type model) to estimate
the equity premium through fundamentals (dividends and earnings). The results
reported show that in the period 1951–2000 the equity premium estimates on the
basis of dividend and earnings growth models are 2.55% and 4.32%, respectively,
far below the equity premium estimate computed on the basis of average returns
(note also that in the late 20th century unusually high expected returns have
occurred). According to Lettau et al. [1197], this reduction of the equity premium
can be rationalized through a reduction in macroeconomic risk. Campbell &
Shiller [353], using the price-dividend ratio, predicted a conditional equity premium
smaller than its sample average. On the other hand, even considering model-based
estimates of the equity premium, Constantinides [494] reported large values for the
(unconditional) equity premium. Note, however, that the equity premium puzzle can
be partly explained by recognizing that agents face uninsurable and idiosyncratic
income shocks, together with borrowing constraints. Moreover, a survivorship bias
similar to that detected in Brown et al. [315] for asset returns can be at the origin of
the high equity premium in the U.S. economy. Indeed, the fact that we only observe
returns of surviving firms induces an upward bias in the estimation of the equity
premium. In Goetzmann & Jorion [797], the authors show that the U.S. market had
the highest uninterrupted real rate of appreciation of all countries (around 5% per
year, while for other countries the median real appreciation rate has been around
1.5% per a year), concluding that the large equity premium observed in the U.S.
market could be the exception rather than the rule. However, some analyses have
shown that the equity premium puzzle is pervasive in most financial markets (see
Campbell [334], Mehra & Prescott [1320], Campbell [336]).
Mehra & Prescott [1319] reported that the average annual real rate of return
on short term bills was 0.8% and the average annual real rate of return on stocks
was 6.98% in the U.S. market during the period 1889–1978. In order to produce
these estimates, the authors assumed a representative
P agent economy with a time
additive power utility function of the form 1 t 1˛
tD0 ı ct =.1 ˛/, supposing that
the aggregate consumption growth rate follows a two-state Markov chain (see the
last part of Sect. 6.4 for a detailed presentation of the model). In Mehra & Prescott
[1319], the Markov chain was calibrated to market data by matching the average
consumption growth rate, together with the sample standard deviation and first-order
serial correlation. In correspondence of an equilibrium, the aggregate consumption
coincides with the aggregate dividends and the parameters ˛ and ı describing the
preference structure of the representative agent were calibrated in order to match the
equity premium and the risk free rate observed empirically. With a discount factor
smaller than one, the authors verified that it is not possible to reproduce the observed
equity premium with a parameter ˛ 10 (the largest equity premium which
could be obtained by the model was computed to be 0:35). In order to match the
empirically observed equity premium, the coefficient of relative risk aversion has to
take unrealistically large values. The covariance of asset returns with consumption
is driven by the variance of the consumption growth rate, which is typically rather
low and, therefore, a large coefficient of relative risk aversion is needed to reproduce
the average equity return. The smoothness of the consumption growth rate plays a
fundamental role in understanding the equity premium puzzle (the lack of variability
of consumption is partly due to the presence of adjustment costs in consumption
that artificially reduce the variability of consumption, see Gabaix & Laibson [748]).
These results are partly related to those in Hall [880], where it is shown that the
consumption growth rate in the postwar years had a small variance while the risk
free rate was low and equity returns and risk free rate variations over time were
large (in general, large real interest rates and large consumption growth rates are
typically not observed simultaneously). These results support the conclusion that
the elasticity of intertemporal substitution is unlikely to be larger than 0:1, a value
confirmed by experimental results in Barsky et al. [169], which is larger than the
reciprocal of the coefficient of relative risk aversion needed to match the historical
equity premium. Summing up, these observations suggest that the relation existing
between the coefficient of relative risk aversion and the elasticity of intertemporal
substitution (see Exercise 7.8) represents a severe limitation of classical asset pricing
theory.
A significant drawback of a large coefficient of relative risk aversion, needed

to match the observed equity premium, is represented by the very large value
that it implies for the equilibrium risk free rate (risk free rate puzzle), in view of
relation (6.77). Therefore, we can summarize as follows the risk free rate puzzle:
given a large coefficient of relative risk aversion (and, hence, a small elasticity
of intertemporal substitution), an agent prefers to smooth consumption over time,
but this contrasts with the saving behavior of typical investors, which generates
an average pro capita consumption growth rate of around 2% per year. This
phenomenon can be explained by a negative time preference (i.e., ı > 1) or by a
very large risk free interest rate. However, this is in contradiction with the historical
observations. A low risk free rate and a low elasticity of intertemporal substitution
can only be explained by a consumption growth rate much lower than the 2%
observed historically (see, e.g., Deaton [530], Kocherlakota [1112], Weil [1648]).
The large values of the coefficient of relative risk aversion obtained above have
been confirmed by many calibration experiments: for instance, Cecchetti & Mark
[382], Kocherlakota [1109] have estimated ˛ D 13:7, Weil [1648] have estimated
˛ 20 and Kandel & Stambaugh [1069] have estimated ˛ D 29 (see also
Kandel & Stambaugh [1068]). The seasonality adjustments of consumption do
not suffice to explain the puzzle: indeed, as shown in Ferson & Harvey [692],
seasonally unadjusted consumption data still require large values of risk aversion
in order to solve the puzzle. These results agree with the test proposed in Hansen &
Jagannathan [891], where a large risk aversion is needed to satisfy the volatility
bound established in Proposition 6.36 on the intertemporal rate of substitution.
Hence, the large equity premium can only be explained by assuming enough
volatility in the stochastic discount factor. Note that Blume & Friend [258] have
estimated through a panel data analysis a small risk aversion coefficient (around 3),
while the experimental analysis in Barsky et al. [169] has produced larger estimates
but still not large enough to match the historical risk premium. However, these
estimates have been questioned by Wheatley [1659], Kocherlakota [1109], Kandel
& Stambaugh [1068], suggesting that a larger risk aversion coefficient may indeed
be plausible.
One of the crucial assumptions in the analysis of Mehra & Prescott [1319] is
that aggregate consumption coincides with aggregate dividends. In Cecchetti et al.
[379], this assumption has been relaxed and non-traded labor income has been
introduced, so that the aggregate consumption is given by the sum of aggregate
dividends and labor income. A bivariate log-normal process with a two-state Markov
drift is estimated for the consumption and dividend growth processes. In this setting,
the equity premium and the risk free rate empirically observed can be reproduced
through plausible parameters. However, the model does not allow to simultaneously
match the first and the second moments of returns. Considering equity as a levered
claim on aggregate consumption (without labor income) and calibrating the degree
of leverage (treated as a free parameter) to match the volatility of stock returns,
Benninga & Protopapadakis [192], Kandel & Stambaugh [1068, 1069] match the
first and the second moments of the risk free rate and of equity returns. Bonomo
& Garcia [266] estimate a heteroskedastic joint bivariate Markov process for
consumption and dividends and try to reproduce the first and the second moments
of real and excess equity returns together with negative serial correlation of excess
returns and predictive power of the dividend-price ratio. While the model captures
the main features of real returns data, the main failure comes from excess returns:
the premium is still small and the dividend-price ratio does not appear to have
significant predictive power with respect to excess returns.
The equity premium puzzle has also been addressed by introducing a small
probability of a future crash in the market. In Rietz [1451], introducing an event
crash in correspondence of which the output falls by 50% of its value with a
probability of 0.4%, the U.S. equity premium can be matched with a risk aversion
coefficient equal to ˛ D 5. Moreover, the risk aversion needed to explain the equity
premium puzzle decreases as the probability of a crash event increases. As shown
in Salyer [1490], the agents’ intertemporal rate of substitution is consistent with
the bound determined in Hansen & Jagannathan [891] (see Proposition 6.36), as
calibrated to real data, however the volatility of excess returns as predicted by the
model is significantly smaller than what observed in the data. Barro [166] further
analyses the possibility that a disaster may explain the equity premium. The author
estimates that “disaster events” yielding an output fall between 15% and 64%
occur with a probability comprised in the interval [1.5%,2%] and allowing for a
probability of a disaster (calibrated as a drop of more than 29%) of 1.7% allows to
match the equity premium with coefficient of relative risk aversion equal to 4 (see
also Barro & Jin [167]). A rare disaster (eventually with time varying intensity) also
contributes to explain other puzzles such as the presence of excess volatility and
return predictability, see Gabaix [747] and Wachter [1637]. We refer to Exercise 7.9
for a presentation of the model considered in Gabaix [747]. However, Julliard &
Ghosh [1054] and Gourio [816] express a skeptical view on the possibility that the
probability of a disaster can explain the observed values of the equity risk premium.
In Labadie [1152], the effects of stochastic inflation on the equity premium are
investigated: assuming a cash in advance constraint, inflation risk increases the
equity premium predicted by the model, but the latter is still less than half of the
equity premium observed in historical data.
The excess volatility results detected in Shiller [1533] focused on the (non-)
stationarity of the dividend process. Marsh & Merton [1306] consider the case of a
stochastic dividend process depending on the asset price process and conclude that
the Shiller’s approach cannot be used to test the validity of classical asset pricing
theory. This type of dividend process is motivated by the empirical analysis of
Lintner [1220], showing that managers typically set dividends having a target pay-
out ratio and choose dividend policies that smooth the dividend changes required
to meet their goal. As a consequence, dividends represent weighted averages of
past earnings. Assuming a constant expected (real) rate of return and that the stock
price reflects investor beliefs (which are in turn given by the expectation of future
discounted dividends), the de-trended dividend process can then be written as a
moving average of current and past de-trended stock prices (see also Marsh &
Merton [1307] and, for a microfoundation of this feedback effect, Timmermann
[1591]). According to this policy, managers typically deviate from the long run
growth path in response to changes in permanent earnings. The authors claim that
such a dividend policy well fits the observed dividend time series. Assuming this
type of dividend policy, inequality (7.6), computed according to the procedure
proposed by Shiller, holds in the opposite direction. The violation is due to the fact
that the de-trended dividend time series is non-stationary and the Shiller’s procedure
suffers from a joint hypothesis problem concerning the stationarity of the dividend
process. However, other dividend smoothing policies in the spirit of Lintner [1220]
do not generate a violation of the volatility inequality (7.6) (see Shiller [1536] for
more details). In this context, see also Exercise 7.5.
The (non)-stationarity of the dividend process has been deeply investigated in the
literature. Shiller [1534, 1536, 1537], Campbell & Shiller [351] report evidence in
favor of a trend-stationary dividend time series, while Kleidon [1104] and Marsh
& Merton [1306] find a negative evidence. If the time series has a trend, then
a geometric random walk test (without a trend component) is biased towards
accepting the null hypothesis of a unit root (see Hamilton [881, Chapter 17] for
a study of univariate processes with a unit root). Including a trend component,
the test still has a low power against trend stationary alternatives (see DeJong &
Whiteman [541]). Using a Bayesian approach, DeJong & Whiteman [541] conclude
that the dividend and the U.S. price time series are more likely to be trend-stationary
than integrated. A possible way to avoid non-stationarity problems is to perform
excess volatility tests on the price-dividend ratio time series, which is more likely
to be stationary than the dividend time series. Both absence of correlation and
volatility bound tests can be performed on the price-dividend ratio time series. In
this perspective, see Cochrane [461] for a result not indicating striking rejections of
the volatility bounds and, on the contrary, Campbell & Shiller [351, 352], LeRoy &
Parke [1188] for results in favor of excess volatility.
The presence of excess volatility can also be tested by running a linear regression
of set on st and it can be shown that this type of test is analogous to an orthogonality
test on price volatility. If asset prices are consistent with classical asset pricing
theory, then the intercept of the regression must be equal to zero and the regression
coefficient must be equal to one. However, this hypothesis is not verified empirically
(see Scott [1513]). The test is well defined in the case of stationary dividend-price
time series, while, if this is not the case, then a similar test can be implemented for
the price-dividend ratio. Even in this case, it has been shown that the intercept is
positive and the regression coefficient is not significantly different from zero.
Positive evidence on the capability of fundamentals to explain asset return
variations has been provided in Fama [662]. The author identifies three sources of
variations in returns: shocks to expected cash flows, shocks to expected returns and
time varying expected returns. Taking into account these three components, around
58% of the variation in ex-post annual stock returns can be explained. Moreover, as
shown in Kothari & Shanken [1124], variables that proxy for market expectations of
future dividends explain over 70% of variations in returns over time. Evans [659],
allowing for non-stationary aggregate dividends and discount rates, establishes that
changing forecasts of future dividend growth account for 90% of the predictable
variations in the dividend-price ratio.
Positive evidence on time varying discount rates as a source of excess volatility is
reported in Cochrane [461]. Bansal & Lundblad [137] show that cash flow growth
rates can be modeled as a stationary ARIMA(1,0,1) process (i.e., they contain a
small predictable long run component). By assuming this process together with a
time varying systematic risk factor, about 70% of the volatility of asset prices can
be explained.
The relevance of microstructure and mispricing effects (in particular liquidity
and bid-ask spreads) on the measurement of the long run performance of contrarian
strategies has been also pointed out in Ball et al. [125]. Moreover, Ball et al.
[125] observe that the profitability of a trading strategy depends on the month
of portfolio formation: contrarian strategies formed in December have positive
excess returns, while those formed in June have negative excess returns. Again,
there is evidence of a January effect. On the other hand, momentum strategies earn
negative profits in January, while positive profits are associated with the remaining
months. There are seasonality patterns in the profits generated by momentum
and contrarian strategies as well as in the predictability of future returns through
past returns, with the seasonality effect being more pronounced in the months of
December and January (therefore, it can be at least partly attributed to tax loss
selling), when effective capital gain tax rates are expected to decrease, see Grinblatt
& Keloharju [830], Grinblatt & Moskowitz [833], Hvidkjaer [995]. Momentum
strategies produce their greatest profits towards the end of the year, due to the
increasing selling pressure on loser assets and their greatest losses towards the
beginning of the year. The opposite occurs for contrarian strategies. On monthly
seasonality effects see also Heston & Sadka [941].
Return seasonality is robust across many different markets, see Hawawini &
Keim [916], Fama [663], Schwert [1510], Malkiel [1290] and also Chap. 10
for intraday patterns in stock returns. Winner-loser reversals have been observed
in national stock market indices (see Richards [1447]). Buckley & Tonks [322]
evaluate a trading strategy based on excess volatility consisting in buying-selling
the market portfolio when it is far away from its mean, showing that such a
strategy can yield excess returns. Moreover, the existence of profitable trading
strategies based on technical analysis has been confirmed by more recent studies
(see Park & Irwin [1400], Po-Hsuan & Chung-Ming [1423], Han et al. [884]). The
profitability of momentum strategies has been confirmed in Chan et al. [407] and an
analogous phenomenon has been observed in European and emerging markets (see
Rouwenhorst [1478], Chan et al. [405], Bhojraj & Swaminathan [229]). Moreover,
the profitability of momentum strategies is confirmed by the investment strategies
frequently adopted by mutual fund managers, whose performance is evaluated over
a short horizon. Note also that tax loss selling practices may induce a momentum
effect around the month of January, when agents sell badly performing assets to
offset capital gains (see Grinblatt & Moskowitz [833]).
In Chan et al. [407] and Barber et al. [141] it has been documented that
investment strategies based on analysts’ recommendations typically earn abnormal
gross returns. This empirical result can be interpreted as an evidence against market
efficiency, either due to the existence of mispricings or due to analysts’ private
information. Moreover, the performance of mutual funds performance is highly
persistent and this can also be interpreted as an evidence against market efficiency
(see Brown & Goetzmann [314]).
The predictability of future returns through variables such as dividend yields,
earnings and interest rates has been criticized from a statistical point of view (see,
e.g., Kaul [1076]). Two main issues have been pointed out, especially in the case
of long horizon returns, see Kirby [1102], Hodrick [951], Richardson & Stock
[1450], Richardson [1448], Goetzmann & Jorion [795], Nelson & Kim [1373].
First, overlapping observations induce serial correlation in errors and, therefore,
the validity of the statistical procedure is negatively affected. Second, there is a
small sample size problem. The regression on endogenous lagged variables is not
unbiased in finite samples and problems mainly arise when the predictor variable
is persistent and its innovations are highly correlated with returns. As a result,
these biases may induce spurious predictability. Taking into account these potential
biases, weak predictability through the dividend yield and earnings yield has been
detected in Goetzmann & Jorion [795], Ang et al. [65], Campbell et al. [348],
while Nelson & Kim [1373], Hodrick [951], Lewellen [1211], Campbell & Yogo
[359], Campbell & Thompson [357], Cochrane [466], Lettau & Van Nieuwerburgh
[1200] still find some evidence of return predictability through the dividend yield,
the book-to-market and the earnings-price ratio. Moreover, as in the case of serial
correlation, predictability seems to be mostly restricted to the pre-Second World
War period (see Nelson & Kim [1373]).
In addition, all the empirical results reporting predictability of future returns
through variables such as past returns, dividends and earnings suffer from a strong
survivorship bias. Indeed, structural relations between macroeconomic/monetary or
firm-specific variables and asset returns are only observed ex-post, conditionally on
the survival of the asset on which the returns are observed. In other words, especially
in the long run, we only observe the returns of the assets which did not default during
the period, thus inducing a bias in the empirical analysis. This bias can introduce
spurious mean reversion and predictability in asset returns. Indeed, among the firms
with a poor performance in the recent past, we can only observe the firms with a
relatively good recent performance which have survived so far. In Goetzmann &
Jorion [796] it is shown that the survivorship bias generates predictability of future
returns through the dividend-price ratio. Taking this fact into account, only marginal
predictability of returns through the dividend yield is observed. This problem is
similar to that identified by Kothari et al. [1126] in the case of the CAPM (see also
Brown et al. [315] on the survivorship bias).
Several studies have investigated the origins of the lead-lag effect: according
to Boudoukh et al. [273], the lead-lag effect is to be attributed to the presence of
non-synchronous trading and transaction costs and not to the existence of delayed
reaction of small sized stocks to information or to time varying expected returns.
However, McQueen et al. [1315] and Lo & MacKinlay [1237] provide evidence
against transaction costs, non-synchronous trading, time varying expected returns
as a source of lead-lag effects and in favor of delayed reaction of small firms to
information. Moreover, Badrinath et al. [104] show that past returns on stocks held
by institutional (informed) traders are positively correlated with contemporaneous
returns on stocks held by non-institutional (uninformed) traders. Somehow analo-
gously, Brennan et al. [295] show that stocks followed by many analysts tend to
lead stocks followed by fewer analysts, while Chordia & Swaminathan [440] show
that there is a lead-lag relationship between returns of high trading volume stocks
and returns of low trading volume stocks. In general terms, this empirical evidence
supports the interpretation of delayed diffusion of information as a source of lead-
lag effects.
Balduzzi & Kallal [117], Balduzzi & Robotti [119] show that multi-beta factor
models can be tested by means of the Hansen-Jagannathan bound (see Proposi-
tion 6.36). Balduzzi & Kallal [117] find evidence against the three-factor model
proposed by Fama & French [670], while Balduzzi & Robotti [119] find that several
macroeconomic factors are indeed priced by the market with a significant risk
premium (in particular: the return on a stock market proxy, the consumption growth,
the slope of the term structure, the interest rate and the default premium) and the
sign of the risk premium is generally consistent with the intuition of the ICAPM. In
general, the ICAPM has a better performance than the CAPM, the CCAPM and the
three-factor model of Fama & French [670].
In Gibbons & Ferson [776], Ferson et al. [690], Ferson [686, 687], Bekaert &
Hodrick [186], the variation over time of asset returns has been described by relying
on latent variables models, where expected returns and ˇ coefficients are conditional
on a set of latent variables. Expected returns vary over time but conditional ˇ are
fixed parameters. In Gibbons & Ferson [776] it is shown that a model based on a
single latent variable is not rejected empirically, while subsequent papers suggest
the presence of more than one latent variable.
Besides the equity premium puzzle, it has also been observed that classical asset
pricing theory encounters difficulties in explaining the risk premia observed in the
term structure of interest rates (see Sect. 6.4). Indeed, as shown in Backus et al.
[103], an intertemporal general equilibrium asset pricing model with time additive
utility is not capable of reproducing the sign, the magnitude and the variability of
the risk premia observed in forward interest rates over expected future spot interest
rates. A very large risk aversion coefficient (greater than 8) would be needed to
generate average risk premia as large as those observed in the Treasury bill markets.
In particular, classical asset pricing theory does not easily explain the positive
relation existing between maturity and risk premium. This phenomenon is typically
referred to as the term premium puzzle.
Finally, we refer to Fama [664], Barber & Lyon [142], Kothari & Warner
[1127], Kothari [1123] for a discussion of the statistical problems associated to
the measurement of abnormal long horizon returns associated with event studies.
In particular, there is often a bias towards the rejection of the market efficiency
hypothesis.
7.5 Exercises
Exercise 7.1 Prove inequalities (7.8)–(7.9).

Exercise 7.2 Prove the two inequalities in (7.11).
Exercise 7.3 As in equation (7.4), define "Ft WD sFt C dt rf sFt1 , for all t 2 N, with
an analogous definition for the information flow G. Suppose that rf > 1 and that
Var."FtCs / D Var."Ft / and Var."G G

tCs / D Var."t /; for all t; s 2 N:
Under this assumption, show that inequality (7.12) holds.

Exercise 7.4 Derive the log-linear approximation (7.13).
Exercise 7.5 Consider an infinitely lived economy (i.e., T D 1) and suppose that
the dividend process of a risky security satisfies the following dynamics:
X
N

dtC1 D .1 C g/dt C k EtkC1 .1 C g/Etk ; for all t 2 N; (7.17)
kD0
for some N 2 N, with d0 D 0 and where .Et /t2N denotes the earnings process and
fk gkD0;1:::;N is a family of non-negative weight factors. This model represents the
situation where the dividend process is set according to a growth rate equal to g, but
managers deviate from this long run growth path in response to changes in earnings
that deviate from their long run growth path. Define by dN t WD dt =.1 C g/t the de-
trended dividend and, similarly, EN t WD Et =.1 C g/t , for all t 2 N. Suppose that the
earnings process .Et /t2N is related to the firm value process .Vt /t2N by Et D rf Vt ,
for all t 2 N, where rf is the constant risk free rate and suppose also that stocks
are priced rationally, i.e., st D Vt , for all t 2 N. Show that, under the present
assumption, the rational ex-post
PT1 price as defined in (7.10) (with the terminal value
sNeT being defined as sNeT WD tD0 sNt =T, with .Nst /t2N denoting the de-trended observed
price process) admits the representation
X
T1
sOet D wtk sNk ; for all 0 t T; (7.18)
kDN
for a suitable family fwtk g of weight factors.

7.5 Exercises 393

Exercise 7.7 In this exercise, we introduce and solve the general equilibrium asset
pricing model proposed by Cecchetti et al. [378] with the purpose of showing
that negative serial correlation in long horizon stock returns is consistent with an
equilibrium model of asset pricing.
Consider an infinitely lived economy (i.e., T D 1) admitting a representative
agent with power utility function of the form u.x/ D x1C =.1 C / and discount
factor ı and where a single risky security is traded, with price process .st /t2N
and paying the aggregate endowment as dividend. Assume that the aggregate
endowment process .et /t2N is modeled as follows:
et D et1 e˛0 C˛1 Xt1 C"t ; for all t 2 N;
where ."t /t2N is a sequence of independent random variables identically distributed

according to a normal law with mean zero and variance 2 and where .Xt /t2N is a
Markov process with state space f0; 1g characterized by the transition probabilities
P.Xt D 1jXt1 D 0/ D 1 q and P.Xt D 0jXt1 D 1/ D 1 p:
This means that the logarithm of the aggregate endowment process follows a random
walk with a stochastic drift, where the process .Xt /t2N is meant to represent the
high/low growth state of the economy. The Markov chain .Xt /t2N is assumed to be
independent of the sequence ."t /t2N .
Show that the equilibrium price st is given by
st D %.Xt /et ; for all t 2 N;
where the function % W f0; 1g ! R is defined by

%.0/ WD ıQ 1 ıQ˛Q 1 . p C q 1/ = ;

%.1/ WD ıQ˛Q 1 1 ı.
Q p C q 1/ = ;
with
ıQ WD ıe˛0 .1C /C.1C / =2 ;

2 2
˛Q 1 WD e˛1 .1C /
and
Q p˛Q 1 C q/ C ıQ2 ˛Q 1 . p C q 1/:

WD 1 ı.
Exercise 7.8 Consider a multi-period economy with dates f0; 1; : : : ; Tg, for some
T 2 N, and a time additive utility function of the form:
X
T
U.c0 ; c1 ; : : : ; cN / D ı t ut .ct /;
tD0
with ı 2 .0; 1/ and ut W RC ! R, for all t D 0; 1; : : : ; T. Let us define by rur t the

coefficient of relative risk aversion of the utility function ut (see Sect. 2.2), for all
t D 0; 1; : : : ; T, and define the elasticity of intertemporal substitution of U as
d log .ct =cs /

U .s; t/ WD ; for all s; t 2 f0; 1; : : : ; Tg:
@U @U
d log @c s
= @c t
Show that the following relation holds for every s; t 2 f0; 1; : : : ; Tg:
ct ı ts u0t .ct /

1 cs u0s .cs /
U .s; t/ D : (7.19)
ct ı ts u0t .ct / r
rur t .ct / cs u0s .cs /
rus .cs /
Moreover, if the utility functions .ut /tD0;1;:::;T are CRRA with the same coefficient
of relative risk aversion for all dates t D 0; 1; : : : ; T, then it holds that
1
U .s; t/ D ; for all s; t D 0; 1; : : : ; T:
rur t .ct /
Exercise 7.9 Consider a representative

P agent economy, where the representative
ıt 1
agent’s utility function is given by 1tD0 e .ct 1/=.1 /. Suppose that, at each
period t C 1, a disaster event can occur with probability pt , with pt denoting the Ft -
conditional probability of occurrence, for all t 2 N. Depending on the occurrence of
a disaster, the aggregate endowment/consumption growth satisfies
(
etC1 eg ; if there is no disaster;
D
et eg BtC1 ; otherwise;
where BtC1 is a random variable taking values in .0; 1/, for all t 2 N, and g
represents the logarithmic rate of growth of the aggregate endowment. Define
R WD ı C g. Suppose also that the occurrence of the disaster affects the dividend
process .dt /t2N of a risky security (whose cum-dividend price process is denoted by
. pt /t2N ) in the following way:
(
dtC1 egd .1 C "tC1 /; if there is no disaster;
D
dt egd .1 C "tC1 /FtC1 ; otherwise;
7.5 Exercises 395
where gd is the logarithmic rate of growth of the dividend process, ."t /t2N is a
sequence of random variables (independent from the disaster event) with zero mean
and Ft is a random variable taking values in Œ0; 1, for all t 2 N. We also assume
that the random variable "tC1 and the random variable representing the occurrence
of a disaster are Ft -conditionally independent, for all t 2 N. Define also the asset
resilience Ht by

Ht WD pt EŒBtC1 FtC1 1jfthere is a disaster at t C 1g _ Ft
and suppose that the resilience process .Ht /t2N satisfies the following dynamics:
1 C H
HtC1 D H C e .Ht H / C "tC1 ; for all t 2 N;
1 C Ht
where H > 1 represents a long term equilibrium resilience level and ."t /t2N is
a sequence of random variables (independent from all the other random variables
introduced above) with zero mean.
Show that the equilibrium cum-dividend price pt of the risky security is given by

dt eRCgd .Ht H /
pt D 1C ; for all t 2 N; (7.20)
1 er 1 er
where r WD R gd log.1 C H /.
Chapter 8
Information and Financial Markets
We must look at the price system as a [: : :] mechanism for

communicating information if we want to understand its real
function.
Hayek (1945)
The preceding chapters have been devoted to the analysis of economies populated
by agents with homogeneous beliefs. We have always assumed that the structure
of the probability space is common knowledge among all the agents, in the sense
that every agent knows the laws of the random variables describing asset prices,
dividends, returns and endowments. Moreover, agents are assumed to be price takers
and rational, but do not recognize any informational value of asset prices concerning
the uncertainty of the economy. This assumption is certainly plausible when agents
have homogeneous beliefs and information, but it becomes quite restrictive in an
economy where agents are endowed with private information.
In this chapter, we consider economies populated by agents with different
information, in the sense that the random variables describing asset prices, dividends
and returns have different probability laws from the point of view of different agents.
In this sense, we say that agents have private information. More specifically, we
distinguish between heterogeneous information, when all the agents of the economy
observe a private signal, specific to each agent, and asymmetric information, when
the economy is populated by informed and uninformed agents, with only the
informed agents having access to some additional information about the economy.
In both cases, we assume that agents are rational and, hence, are able to identify the
informational content of asset prices. When the information is heterogeneous, asset
prices will aggregate the agents’ private information, possibly making information
fully homogeneous among the agents, while, when the information is asymmetric,
asset prices will transmit information from the informed to the uninformed agents. It
has to be noted that, when the information is heterogeneous or asymmetric, agents
will trade for two main motivations: for risk sharing, similarly as in the analysis
developed in the preceding chapters, and for informational reasons (speculation).
Due to this second motivation for trading, especially when the information is
asymmetric, there will be an adverse selection problem: less informed agents will
be afraid to trade with more informed agents and this may preclude them for trading.
The central question of the present chapter can be phrased in the following terms:

398 8 Information and Financial Markets
in a perfectly competitive market, when agents have private (heterogeneous or

asymmetric) information, can asset prices aggregate/transmit the agents’ private
information?
The above question can be easily addressed in the context of a simple two-
period economy (i.e., t 2 f0; 1g), where agents take their investment-consumption
decisions at the initial date t D 0 and consume at t D 1. In order to take decisions
at t D 0, agents need to evaluate the probability distribution of future consumption
at t D 1 and, therefore, they need to evaluate the probability distribution of the
random variables describing the dividends delivered at t D 1. Suppose that the
information is heterogeneous because each agent can observe at the initial date
t D 0 a private signal with a non-trivial informational content. By observing
the private signal, each agent will revise his beliefs concerning the probability
distribution of the random variables associated with the investment opportunities.
Since each agent is assumed to be an expected utility maximizer, his optimal
choice at t D 0 (in particular his asset demand) will depend on the probability
distribution updated on the basis of the information conveyed by the observation
of the private signal. This implies that, by trading in the market, an agent will
somehow reveal his beliefs. Investment decisions and, therefore, the resulting
market prices will carry a non-trivial informational content, resulting from the
aggregation of the agents’ private information. By observing asset prices, agents
can infer the information of other agents and, as a consequence, update their beliefs.
In this chapter, we aim at understanding to what extent asset prices make public,
aggregate and transmit private information in a perfectly competitive economy. In
particular, we are interested in understanding when this process of price observation
and revision of beliefs will lead to homogeneous beliefs among the agents, with
market prices fully revealing the agents’ private information. The relevance of this
aggregation/transmission of information phenomenon depends on the organization
of the market. In this chapter, we consider perfectly competitive markets, while in
Chap. 10 we will extend the analysis to alternative market settings.
If the answer to the above question is positive, then markets are information-
ally efficient. It goes without saying that a positive answer would be a strong
argument in favor of a free market: indeed, not only does the market allocate
resources in an efficient way (allocative efficiency, first welfare theorem), but it
also aggregates/transmits private information, thereby eliminating any advantage
associated with private information. In this chapter, we want to analyse the aggre-
gation/transmission of information function of prices and evaluate under which
conditions markets are informationally efficient. Of course, it suffices to consider
the insider trading legislation adopted in most countries and the amount of resources
spent in market research by investors to understand that the message from the
efficient markets theory is not fully satisfactory in reality (see also Sect. 10.5).
The fact that asset prices aggregate/transmit private information implies that the
classical concept of Arrow-Debreu equilibrium does no longer suffice to describe a
stable condition of the economy. To this end, the appropriate notion of equilibrium is
represented by the Green-Lucas equilibrium. The limitations of the Arrow-Debreu
equilibrium in an economy where agents have private information can be understood
8 Information and Financial Markets 399
by considering the following example, taken from Huang & Litzenberger [971,
Section 9.2].
Consider a two-period economy (i.e., t 2 f0; 1g) with five possible states of the
world ˝ D f!1 ; : : : ; !5 g. There exists a single consumption good which is traded
at t D 0 and consumed at t D 1. We assume that markets are complete, in the sense
that at t D 0 there are five markets open for the five Arrow securities paying one
unit of the single consumption good in correspondence of each state of the world !s ,
s D 1; : : : ; 5. Suppose that the economy is populated by two agents a and b. Both
agents are endowed (ex-ante) with the same beliefs, in the sense that they agree on
the probabilities of occurrence of each of the five states of the world. However, at
the initial date t D 0, both agents observe a private signal, represented by a binary
random variable yQ i , for i D a; b, taking two possible values fL; Hg. We assume that
( (
L; if ! 2 f!1 ; !2 ; !3 g; L; if ! 2 f!1 ; !2 g;
yQ .!/ D
a
and yQ .!/ D
b
H; if ! 2 f!4 ; !5 gI H; if ! 2 f!3 ; !4 ; !5 g:
The signal induces heterogeneous information: if agent a observes fQya D Lg then he

will know that the state of the world belongs to the subset f!1 ; !2 ; !3 g. On the other
hand, if agent b observes the realization fQyb D Lg, he will know that the state of the
world belongs to f!1 ; !2 g. After the observation of the private signals, the beliefs of
agent a will be represented by the partition ff!1 ; !2 ; !3 g; f!4 ; !5 gg, whereas those
of agent b will be described by ff!1 ; !2 g; f!3 ; !4 ; !5 gg. Furthermore, we assume
that the structure of these two partitions is common knowledge among the two
agents. In the context of the present example, the two agents exhibit heterogeneous
beliefs. Indeed, after the observation of the signal, the two agents believe in different
probability distributions, here represented by different partitions of ˝. In particular,
after the observation of their private signals, they can disagree about the possibility
of occurrence of the state of the world !3 .
Let us assume that the two agents a and b have strictly increasing preferences. In
this case, it can be easily shown that, if one adopts the classical notion of equilibrium
introduced in Chap. 4, then there does not exist an equilibrium of the economy.
Indeed, suppose that the realized state of the world is !1 . In this case, the signal yQ a
received by agent a takes the value L and the signal yQ b received by agent b takes
the value L as well. After the observation of the private signal, agent a learns that
the true state of the world belongs to f!1 ; !2 ; !3 g, while agent b learns that the
true state of the world belongs to the smaller subset f!1 ; !2 g. This difference in
the agents’ beliefs implies that there cannot exist an equilibrium price p3 of the
Arrow security corresponding to the state of the world !3 . Indeed, suppose that
such a security has a strictly positive price. In that case, agent b would like to short
sell an unlimited amount of the security, since he knows that the state of the world
cannot be !3 , and use the proceeds of this short sale to finance consumption in the
two states of the world f!1 ; !2 g. On the other hand, agent a, who assigns a strictly
positive probability of occurrence to the state of the world !3 , would be interested in
investing in the Arrow security corresponding to !3 , but certainly his demand would
not balance the supply of agent b. Hence, markets cannot clear and an Arrow-Debreu
equilibrium cannot exist if markets are complete. A similar argument can be applied
if the Arrow security corresponding to !3 has a non-positive price.
By examining the above example, we can easily see why the Arrow-Debreu
equilibrium does not represent the appropriate notion of equilibrium in the presence
of heterogeneous beliefs. In the classical definition of equilibrium, agents consider
prices as given and only use prices to define their budget constraints. In other words,
agents do not assign any informational value to prices. However, in the context of
the above example, since the structure of the partitions is assumed to be of common
knowledge among the two agents, it is rather unrealistic to assume that agent a will
not revise his beliefs after observing a huge short sale from agent b, for any strictly
positive price p3 of the Arrow security paying one unit of the consumption good in
correspondence of !3 . Somehow, the trading activity of agent b reveals his private
information to the market. In this case, a price equal to zero for the Arrow security
would be of equilibrium if, after observing the private signal fQya D Lg and the price
p3 D 0, agent a would infer that the realized state of the world cannot be !3 . Note
also that, after the revision of beliefs, the ex-ante difference in the agents’ private
information would be completely eliminated ex-post and both agents would believe
that the state of the world belongs to f!1 ; !2 g.
The above discussion leads us to the introduction of the Green-Lucas equilib-
rium, which is a rational expectations equilibrium. In a Green-Lucas equilibrium,
all agents determine their optimal demands by fully exploiting the information
contained in asset prices. In the context of the above example, a Green-Lucas
equilibrium with p3 D 0 can be constructed. If fQya D yQ b D Lg and p3 D 0, agent
b will find p3 D 0 coherent with his private information and agent a, knowing
the partition of agent b, will learn from observing p3 D 0 that agent b knows for
sure that the realized state of the world cannot correspond to !3 . Therefore, by
updating his beliefs via the Bayes rule, agent a will infer that the state of the world
cannot be !3 . As explained above, the (ex-ante) difference in the agents’ private
information will be fully equalized (ex-post) by equilibrium prices.
Radner [1439, p. 941] points out that “an agent who has a good understanding
of the market is in a position to use market prices to make inferences about the
(non-price) information received by other agents: : : These inferences are derived,
explicitly or implicitly, from an individual’s “model” of the relationship between the
non-price information received by market participants and the market prices: : : An
equilibrium of this system, in which the individual models are identical with the true
model, is called a rational expectations equilibrium: : : The concept of equilibrium
is more subtle, of course, than the ordinary concept of the equilibrium of supply and
demand. In a rational expectation equilibrium, not only are prices determined so as
to equate supply and demand, but individual economic agents correctly perceive
the true relationship between the non-price information received by the market
participants and the resulting equilibrium market prices. This contrasts with the
ordinary concept of equilibrium in which the agents respond to prices but do not
attempt to infer other agents’ non-price information from actual market prices”.
8 Information and Financial Markets 401
In this chapter, we shall distinguish between heterogeneous beliefs, i.e., when

agents exhibit different (posterior) probability distributions conditioned on the
observation of private signals related to the economic environment, and het-
erogeneous opinions, i.e., when agents exhibit different probability distributions
not related to the observation of private signals informative about the economic
environment. Typically, information about the economic environment is meant to
be information about the distribution of asset payoffs/dividends. The difference
between heterogeneous beliefs and heterogeneous opinions may seem subtle but
has quite distinct implications. In particular, an agent will not change his behavior or
revise his probability distribution by knowing that another agent exhibits a different
probability distribution if the latter is not motivated by private information related
to the economic conditions (heterogeneous opinions). On the contrary, an agent
will revise his behavior if he knows that the other agent’s probability distribution
is generated by private information about the economy (heterogeneous beliefs). In
general, heterogeneous opinions can be treated within the paradigm described in the
preceding chapters by introducing heterogeneous agents. In this chapter we shall be
mostly concerned with heterogeneous beliefs. We will briefly discuss the case of
heterogeneous opinions in Sect. 8.5.
The question of whether market prices fully reflect the agents’ information
represents the central question of efficient markets theory and goes back to seminal
contributions in the financial markets literature. A microeconomic approach to this
issue is somehow more recent, going back to the contributions of the rational
expectations theory (see Grossman [842]). Moreover, this question has important
implications from an empirical perspective, as discussed in Chap. 7. From this point
of view, the analysis of market efficiency can be reduced to the following problem:
given an equilibrium market model, do asset prices reflect the information available
in the market?
Any answer to this question requires on the one hand a definition of an
equilibrium model and, on the other hand, a definition of the notion of available
information. Hence, testing market efficiency corresponds to jointly testing the
validity of an equilibrium model and the fact that a given information set is embed-
ded in market prices. This is intimately related to the fundamental no-arbitrage
equation presented in Sect. 6.5 and discussed from an empirical perspective in
Chap. 7. In particular, the fundamental solution (6.86) to the no-arbitrage equa-
tion (6.85) shows that equilibrium prices are determined by three key ingredients:
the dividend process .dt /t2N , the information flow F D .Ft /t2N and the stochastic
discount factor. In particular, under the assumption of risk neutrality, future excess
returns cannot be predicted on the basis of the current information Ft and expected
excess returns are null (see Proposition 7.1). If an information set Ft does not allow
to predict future excess returns, then asset prices are informationally efficient in
the sense that they correctly reflect the available information. If this is the case,
then expected returns are equal to the risk free rate rf and any trading strategy
will not perform better on average than the market portfolio or simple buy-and-
hold strategies in any of the traded security. Somehow more generally, following
Jensen [1031], we can define (informational) market efficiency with respect to some
information flow F as the impossibility of making profits by trading on the basis of

the information contained in F.
Following a classification widely accepted in the literature, we can identify
three different forms of market efficiency, depending on the specification of the
information flow (see Sect. 7.2 and Fama [661]):
• Efficiency in weak form, when the information set includes all past realizations
of asset prices and returns;
• Efficiency in semi-strong form: when the information set includes all publicly
available information, including all past realizations of asset prices and returns;
• Efficiency in strong form: when the information set includes all publicly available
information and all private information.
According to this classification, three different information flows are associated to
three different levels of market efficiency. Clearly, these three notions of market
efficiency are increasingly stronger, in the sense that strong efficiency implies semi-
strong efficiency and, in turn, semi-strong efficiency implies weak efficiency. This
simply follows by the inclusions among the different information sets considered in
the above classification.
The typical way of empirically testing market efficiency consists in building
trading strategies on the basis of the information contained in some information
set. If such a strategy can be shown to perform better than the market, then this
is an indication that the considered information is not fully incorporated in market
prices and, hence, some form of market efficiency is rejected. In Chap. 7, discussing
the empirical results on the implications of the fundamental no-arbitrage equation,
we have considered market efficiency in the weak and in the semi-strong form.
As we have seen, there is empirical evidence showing that asset returns exhibit
predictable components. However, as we have already mentioned, these empirical
findings do not necessarily have to be interpreted as an evidence against market
efficiency (or against agents’ rationality). Indeed, as a matter of fact, returns can
be predictable because of time-varying investment opportunities or of risk averse
agents. The empirical evidence reported in the literature on market efficiency in the
strong form is mostly negative: trading strategies constructed on the basis of private
information are shown to yield positive excess returns with respect to the market.
This phenomenon is intimately related to insider trading (see Sect. 10.5).
This chapter is structured as follows. In Sect. 8.1, we study the value of
information first from the point of view of an individual decision maker and then for
the economy as a whole, introducing the notion of Green-Lucas equilibrium and the
role of information transmission/aggregation of market prices. Sections 8.2 and 8.3
respectively deal with the possibility and the impossibility of informationally effi-
cient markets, also discussing the acquisition of information by market participants.
In Sect. 8.4, we provide a brief overview on the role of information in dynamic
multi-period market models. Section 8.5 is concerned with the equilibrium and asset
pricing implications of economies where the agents exhibit different opinions. In
Sect. 8.6, we survey the empirical evidence on information-based models. At the
8.1 The Role of Information in Financial Markets 403
end of the chapter, we provide a guide to further readings as well as a series of

exercises.
8.1 The Role of Information in Financial Markets
In the most simple context, the role of information in financial markets can be
explained by analogy to the example of the farmer given at the beginning of
Chap. 4, by introducing an intermediate date between the sowing and the harvesting
dates. At this intermediate date, the farmer observes a private signal carrying
some information about the future weather conditions. The present simple example
consists of three dates (i.e., t 2 f1; 0; 1g): at t D 1 the farmers sow, at the
intermediate date t D 0 they receive private signals and at t D 1 they harvest. Note
that, even though all the farmers start at t D 1 with the same prior beliefs about
the future weather conditions, if they receive different private signals at t D 0 then
they will form heterogeneous beliefs about the future weather conditions.
Generalizing this example, let us consider an economy populated by I 2 N
agents (i D 1; : : : ; I). At the starting date t D 1, all agents share homogeneous
beliefs about the future state of the world, described by a common probability space
.˝; F ; P/. For simplicity, we assume that the probability space is finite, i.e., the set
˝ consists of S 2 N elementary states of the world f!1 ; : : : ; !S g and we denote by
s WD P.!s / > 0 the probability of occurrence of state !s , for all s D 1; : : : ; S. The
structure of the probability space .˝; F ; P/ is assumed to be common knowledge
among the I agents. We assume that each agent is characterized by the couple
.ui ; ei /, where ui W RC ! R is a strictly increasing and strictly concave utility
function and the vector ei D .ei1 ; : : : ; eiS / 2 RSC denotes the endowment of agent i
(expressed in terms of wealth or some reference consumption good) in the S possible
states of the world, for all i D 1; : : : ; I. Similarly as in Chap. 4, a consumption plan
is represented by a vector x D .x1 ; : : : ; xS / 2 RSC , with xs representing consumption
in the state of the world !s . We assume that consumption is non-negative and only
takes place at the final date t D 1. All the elements of the economy introduced so far
are assumed to be common knowledge among the agents. We want to point out that
the present setting can be generalized to the case of state dependent utility functions.
At the intermediate date t D 0, each agent i observes a private signal represented
by the random variable yQ i . The signals are private in the sense that each agent i only
observes his own signal yQ i , without having access to the signals of the other agents.
For simplicity, we assume that each random variable yQ i can take J 2 N possible
values fyi1 ; : : : ; yiJ g, for all i D 1; : : : ; I, and we define . yij / WD P.Qyi D yij /. We
denote by yQ the I-dimensional random vector .Qy1 ; : : : ; yQ I / and let . y/ WD P.Qy D y/,
for all y 2 Y , with Y denoting the set of all possible realizations of the random
vector yQ . We assume that the probability distribution f. y/I y 2 Y g of the random
vector yQ is common knowledge among the I agents and is known at the initial date
0
t D 1. Note that we allow for yij ¤ yij , for i; i0 2 f1; : : : ; Ig, and it may happen
0
that agent i observes the realization yij while agent i0 observes the realization yij0 , for
some j; j0 2 f1; : : : ; Jg with j ¤ j0 .
As soon as an agent observes his private signal at the intermediate date t D 0, he
will update his beliefs according to the Bayes rule. Let y D . y1 ; : : : ; yI / 2 Y denote
a realization of the random vector yQ of the private signals. For each i D 1; : : : ; I, the
updated beliefs of agent i are represented by the conditional distribution
i i
v1 . y /; : : : ; vSi . yi / ; (8.1)
with vsi .yi / WD P.!s jQyi D yi / representing the probability of the state of the world
!s conditionally on the observation of signal yi 2 fyi1 ; : : : ; yiJ g by agent i, for all
i D 1; : : : ; I. At date t D 0, the agents have heterogeneous beliefs since they observe
different signals, leading to different conditional probability distributions for the
unknown state of the world. It is important to remark that the private signals are
assumed not to directly affect the agents’ endowments nor their utility functions.
However, the signals convey useful information on the unknown state of the world.
Summing up, in the present setting we can distinguish three different dates:
• t D 1 (ex-ante): nature selects a state of the world;
• t D 0 (interim): agents observe private signals and update their beliefs;
• t D 1 (ex-post): the state of the world is revealed and agents consume.
In the remaining part of this section, we shall first analyse the value of receiving
a private signal from the individual point of view and then the effect of information
from a social point of view, introducing the notion of Green-Lucas equilibrium. The
presentation of this section is inspired from Laffont [1154, Chapters 4 and 9].
The Value of Information
In order to analyse the impact of additional information on the optimal choice of an

individual, let us consider an optimal choice problem of the following form, for a
strictly increasing and strictly concave utility function u W RC ! R (since we now
consider an individual optimization problem, we drop the superscript i):
X
S
max s u.xs /; (8.2)
.x1 ;:::;xS /2B
sD1
where B RSC represents the set of feasible consumption plans, with respect to a
budget constraint defined in terms of a given endowment .e1 ; : : : ; eS /. We assume
that an optimal solution .x1 ./; : : : ; xS .// 2 B exists, where we emphasize the
dependence on the probability distribution D .1 ; : : : ; S / representing the prior
beliefs. We denote by U ./ the optimal value of the expected utility associated
to problemP(8.2) in correspondence of the prior beliefs D .1 ; : : : ; S /, i.e.,

U ./ WD SsD1 s u.xs .//.
Consider the case of an agent who can observe a signal yQ carrying some
information about the unknown state of the world. We assume that yQ takes values
in the set fy1 ; : : : ; yJ g, for some J 2 N, with corresponding probabilities yQ D
.. y1 /; : : : ; . yJ //. As explained above, after the observation of the signal, the
agent will update his beliefs via the Bayes formula, leading to the conditional
distribution .v1 .yj /; : : : ; vS . yj //, where yj denotes the realization of the random
variable yQ , for some j D 1 : : : ; J. The agent takes a decision after the observation
of the signal but before the revelation of the state of the world. Hence, given the
observation of the signal fQy D yj g, for some j D 1; : : : ; J, the optimal choice
problem (8.2) can be transformed into
X
S
max vs . yj /u.xs /: (8.3)
.x1 ;:::;xS /2B
sD1
As in the case of problem (8.2), the optimal consumption plan solving problem
(8.3) will depend on the conditional probability distribution .v1 . yj /; : : : ; vS . yj //.
We denote the optimal solution to (8.3) by .x1 .v. yj //; : : : ; xS .v. yj ///, assuming
that an optimal solution to (8.3) exists. We denote by U .v. yj // the optimal value
of the expected utility conditionally on the realization fQy D yj g, i.e.,
X
S

U v. yj / WD vs . yj /u xs .v. yj // ;
sD1
for yj 2 fy1 ; : : : ; yJ g. We are now in a position to establish the following

proposition, which compares the maximal expected utilities associated to problems
(8.2) and (8.3).
Proposition 8.1 Let U ./ denote the maximal expected utility associated to
problem (8.2) and, for each possible realization yj 2 fy1 ; : : : ; yJ g of the random
variable yQ , let U .v. yj // denote the maximal expected utility associated to problem
(8.3) conditionally on the event fQy D yj g, for j D 1; : : : ; J. Then it holds that
X
J

U ./ . yj /U v. yj / : (8.4)
jD1
Proof Let j 2 f1; : : : ; Jg. By definition, since the set B does not depend on the
probability distribution, it holds that
X
S
XS

vs . yj /u xs ./ vs . yj /u xs .v. yj // ;
sD1 sD1
for every j D 1; : : : ; J. In turn, multiplying both sides by . yj /, summing over all

j D 1; : : : ; J and noting that vs . yj /. yj / D P.!s ; yQ D yj /, for all s D 1; : : : ; S and
j D 1; : : : ; J, we have that
X
S
XJ X
S

U ./ D s u xs ./ D . yj / vs . yj /u xs ./
sD1 jD1 sD1
X
J X
S
XJ

. yj / vs . yj /u xs .v. yj // D . yj /U v. yj / ;
jD1 sD1 jD1

u
The result of Proposition 8.1 is in line with economic intuition. Indeed, it shows
that the optimal expected utility of an uninformed agent (corresponding to problem
(8.2)) is always smaller than the expected optimal utility achieved by an informed
agent (corresponding to problem (8.3)). In other words, at least from an individual
point of view, additional information is always beneficial and has a non-negative
value. In the worst possible case, the additional information conveyed by the signal
yQ has no value and the informed agent can at least do as well as the uninformed
agent. Note that this result does not require any assumption on the utility function
(in particular, it does not rely on risk aversion) nor on the feasibility set B (apart
from the fact that it is not state dependent) and is a consequence of the fact that the
optimal expected utility U ./ is convex with respect to the vector of probabilities
D .1 ; : : : ; S /, see Exercise 8.1. The introduction of additional information also
affects optimal saving decisions (see Sect. 3.4). Indeed, as shown in Exercise 8.2,
if the additional information leads to an earlier resolution of uncertainty, then it can
induce a reduction in the level of saving by a prudent agent.
The observation of the signal yQ induces an information structure represented by
the couple . yQ ; v/, where we denote by v the .S J/-dimensional matrix with
elements vs . yj /, for all s D 1; : : : ; S and j D 1; : : : ; J. Each column of the matrix v
represents the conditional probability distribution of the S states of the world given
one of the J possible realizations of the signal. Note that, by the Bayes rule, it
holds that D v yQ , where we recall that D .1 ; : : : ; S / denotes the prior
probability distribution of the S states of the world. Considering the two extreme
cases, the information structure . yQ ; v/ can be fully informative if the observation
of the signal fully reveals the state of the world or completely uninformative if the
conditional probabilities coincide with the prior probabilities. In the latter case, the
result of Proposition 8.1 holds with an equality and the signal does not convey any
useful information (see also Gollier [800]). In view of Proposition 8.1, it is evident
that a fully informative information structure is always preferred to a completely
uninformative one.
Proposition 8.1 shows that the value of information is always non-negative (we
call this effect the Blackwell effect). Let us now assume that an agent can choose
between the observation of two possible signals yQ a and yQ b . Both signals convey some
information on the unknown state of the world, but they do not generate the same
information structure. In this context, an agent is interested in determining which
one of the two signals is more informative. For simplicity, we assume that yQ a and yQ b
take values in the same set fy1 ; : : : ; yJ g but have different probability distributions
a D .1a ; : : : ; Ja / and b D .1b ; : : : ; Jb / and lead to different conditional
distributions v a . y/ D .v1a . y/; : : : ; vSa . y// and v b . y/ D .v1b . y/; : : : ; vSb . y//, for each
y 2 fy1 ; : : : ; yJ g, where, for ` D a; b, we denote by v ` the .S J/-dimensional
matrix where the j-th column v ` . yj / is the probability distribution of the S states of
the world conditionally on the observation of fQy` D yj g, for some j D 1; : : : ; J.
In this context, we say that the information structure represented by . a ; v a /
is finer (or more precise) than the information structure . b ; v b / if, for any utility
function u W RC ! R and for any feasible set of choices B RSC , it holds that
X
J
XJ

a . yj /U v a . yj / b . yj /U v b . yj / ; (8.5)
jD1 jD1
where U .v ` . yj // is defined as in (8.3) with respect to the vector of conditional

probabilities .v1` . yj /; : : : ; vS` . yj //, for ` D a; b and j D 1; : : : ; J. In other words,
. a ; v a / is finer than . b ; v b / if the expected maximal utility associated to . a ; v a /
is greater than the expected maximal utility associated to . b ; v b /.
The following proposition, due to Blackwell [246], provides a necessary and
sufficient condition for an information structure to be finer than another (see
Exercise 8.3 for a proof of the sufficiency part).
Proposition 8.2 Let . a ; v a / and . b ; v b / be two information structures as
described above. Then . a ; v a / is finer than . b ; v b / if and only if there exists
a .J J/-dimensional matrix K, satisfying Kij 0, for all i; j D 1; : : : ; J, and
PJ
iD1 Kij D 1, for all j D 1; : : : ; J, such that
vb D va K and a D K b : (8.6)
Condition (8.6) can be equivalently formulated as the property that the condi-
tional probability distribution associated to the information structure . b ; v b / is a
mean-preserving spread 1 of the conditional distribution associated to . a ; v a /.
1
Given two random variables xQ1 and Qx2 , we say that the law of Qx2 is a mean-preserving spread of
the law of xQ1 if it holds that xQ2 is equal in law to xQ1 C Qz, where Qz is a random variable satisfying
EŒ Qz jQx1 D 0 (compare with part (iii) of Proposition 2.10).
Pareto Optimality and Green-Lucas Equilibrium
In the previous subsection, we have analysed the value of information from the
point of view of an individual decision maker. Let us now consider the role of
information in an economy populated by I agents, as introduced at the beginning
of this section. In comparison with the two-period economy considered in Chap. 4,
the present setting features an intermediate date in correspondence of which agents
observe the private signals represented by the random vector yQ . Hence, besides the
notions of ex-ante and ex-post Pareto optimality (see Definitions 4.1 and 4.2), we
can formulate a third notion of Pareto optimality.
Definition 8.3 A feasible allocation fxi 2 RSC I i D 1; : : : ; Ig is interim Pareto
0
optimal if there is no feasible allocation fxi 2 RSC I i D 1; : : : ; Ig such that
X
S
0
X
S
vsi . yij /ui .xis / vsi . yij /ui .xis /; for all i D 1; : : : ; I and j D 1; : : : ; J;
sD1 sD1
with at least one strict inequality, for some i 2 f1; : : : ; Ig and j 2 f1; : : : ; Jg.
The notions of ex-ante, interim and ex-post Pareto optimality differ with respect
to the date at which the optimality of an allocation is evaluated, respectively at date
t D 1, before any information is released (ex-ante), at date t D 0, when agents
observe the signals but do not known the state of the world (interim), and at date
t D 1, when the state of the world is revealed (ex-post). If all prior and conditional
probabilities are strictly positive, it is easy to see that ex-ante optimality implies
interim optimality and, in turn, interim optimality implies ex-post optimality (see
Exercise 8.4 and compare also with Holmstrom & Myerson [956]).
Having introduced the notions of ex-ante, interim and ex-post Pareto optimality,
it is interesting to analyse the interplay between information and allocative effi-
ciency. In particular, an important result due to Milgrom & Stokey [1341] shows
that, if an allocation is ex-ante Pareto optimal, then the observation of private
signals cannot create any incentive to trade. This result is sometimes called the
no-trade theorem. To analyse this property, we adopt the setting introduced at the
beginning of the present section, with an economy f.ui ; ei /I i D 1; : : : ; Ig and S
possible states of the world. Let the random vector yQ D .Qy1 ; : : : ; yQ I / represent the
private signals observed by the agents. Following Milgrom & Stokey [1341], we say
that fzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig is a set of feasible signal-
contingent trades if
eis C zi .s; yij / 0; for all s D 1; : : : ; S; i D 1; : : : ; I and j D 1; : : : ; J;

and
X
I
zi .s; yi / 0; for all s D 1; : : : ; S and for every realization . y1 ; : : : ; yI / 2 Y :
iD1
If fzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig does not depend on the signal

(i.e., each zi ./ is a function only of the state of the world and not of the signal and,
thus, reduces to a function of one variable) and is feasible, then it is simply said
to be a set of feasible trades. We are now in a position to establish the following
proposition.
Proposition 8.4 Let fzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig be a set of
feasible signal-contingent trades. Then there always exists a set of feasible trades
(i.e., not depending on the signal) which is ex-ante (weakly) preferred by every
agent.
Assume furthermore that there exists at date t D 1 a complete set of markets
for the S Arrow securities and that the allocation fei I i D 1; : : : ; Ig is an equilibrium
allocation at t D 1. Then the following hold:
(i) the allocation fei I i D 1; : : : ; Ig cannot be ex-ante Pareto improved by any set
of feasible signal-contingent trades;
(ii) suppose that the markets for the S Arrow securities are open also at the interim
date t D 0. Then every agent chooses not to trade at date t D 0.
Proof Let fzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig be a set of feasible
signal-contingent trades and define
X
J
zNi .s/ WD P. yQ i D yij j!s /zi .s; yij /; for all s D 1; : : : ; S and i D 1; : : : ; I:
jD1
The fact that fzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig is feasible implies

that fNzi W f1; : : : ; Sg ! RI i D 1; : : : ; Ig is also feasible, i.e., it is a set of
feasible trades. Moreover, since the utility function ui is assumed to be concave,
the conditional version of Jensen’s inequality implies that
X
S
XS X
J

s ui eis C zNi .s/ P. yQ i D yij ; !s /ui eis C zi .s; yij / ;
sD1 sD1 jD1
for all i D 1; : : : ; I, thus proving the first part of the proposition.

Furthermore, if markets are complete and the initial allocation fei I i D 1; : : : ; Ig
is of equilibrium, then fei I i D 1; : : : ; Ig is also ex-ante Pareto optimal (see Sect. 4.2)
and claim (i) of the proposition follows.
In order to prove part (ii), arguing by contradiction, suppose that at date t D 0 the
S markets reopen and each agent i observes a signal yiki , for some ki 2 f1; : : : ; Jg,
for i D 1; : : : ; I, such that fziki W f1; : : : ; Sg ! RI i D 1; : : : ; Ig is a set of feasible

P
trades, meaning that eis C ziki .s/ 0 and IiD1 ziki .s/ 0, for all s D 1; : : : ; S and
i D 1; : : : ; I, and such that
X
S
XS

vsi . yiki /ui eis C ziki .s/ vsi . yiki /ui eis ; for all i D 1; : : : ; I; (8.7)
sD1 sD1
with strict inequality holding for some i 2 f1; : : : ; Ig. Define then the set of feasible
signal-contingent trades fOzi W f1; : : : ; Sg fyi1 ; : : : ; yiJ g ! RI i D 1; : : : ; Ig by
(
ziki .s/ if j D ki ;
zO
i
.s; yij / WD
0 otherwise;
for all s D 1; : : : ; S and i D 1; : : : ; I. For every i D 1; : : : ; I, it then holds that
X
S X
J

P.Qyi D yij ; !s /ui eis C zOi .s; yij /
sD1 jD1
X
J X
S

D . yij / vsi . yij /ui eis C zOi .s; yij /
jD1 sD1
X
J X
S
X
S

D . yij / vsi . yij /ui eis C . yiki / vsi . yiki /ui eis C ziki .s/
jD1 sD1 sD1
j¤ki
X
J X
S
X
S
X S

. yij / vsi . yij /ui eis C . yiki / vsi . yiki /ui eis D s ui eis ;
jD1 sD1 sD1 sD1
j¤ki
with strict inequality holding for some i 2 f1; : : : ; Ig (under the standing assumption
that . yij / > 0, for all i D 1; : : : ; I and j D 1; : : : ; J). However, this contradicts part
(i) of the proposition and, hence, inequality (8.7) must hold as an equality, for all
i D 1; : : : ; I. Moreover, since the utility functions ui are assumed to be strictly
concave, for all i D 1; : : : ; I, this implies that ziki .s/ D 0, for all s D 1; : : : ; S and
i D 1; : : : ; I. Indeed, if this was not the case, then the strict concavity of the utility
functions would imply that the set of trades fziki =2 W f1; : : : ; Sg ! RI i D 1; : : : ; Ig
is feasible and yields a strict ex-ante Pareto improvement of the initial allocation,
contradicting part (i) of the proposition. t
u
Even though the assumptions of Proposition 8.4 are rather strong, the result is
important. Indeed, it shows that, if a Pareto optimal allocation has been reached
ex-ante, then the arrival of new information does not provide by itself a sufficient
incentive for trading, even if markets reopen. In other words, letting the markets be
open after the arrival of the additional information represented by the signal yQ does
not change the equilibrium allocation (provided that markets are complete). This
result is in line with the fact that ex-ante Pareto optimality implies interim Pareto
optimality and, in turn, interim Pareto optimality implies ex-post Pareto optimality
(see Exercise 8.4).
It is important to remark that the result of Proposition 8.4 strongly relies on
the implicit assumptions that agents form rational expectations, have homogeneous
prior beliefs and that the feasibility of a set of trades is common knowledge in the
economy. In particular, the fact that the feasibility of a set of trades is common
knowledge is crucial for obtaining the no-trade result of part (ii) of Proposition 8.4.
Indeed, since the initial allocation is assumed to be Pareto optimal, any further
trading activity at t D 0 cannot be motivated by insurance or risk-sharing purposes.
As a consequence, any trading activity at t D 0 must be driven purely by speculation
(i.e., agents trade hoping to find advantageous bets with other agents), after the
revision of beliefs on the basis of the private signals received. However, if the
feasibility of a trade is common knowledge, then, as stated by Milgrom & Stokey
[1341], “the willingness of the other traders to accept their parts of the bet is
evidence to at least one trader that his own part is unfavorable”. In other words,
at least one agent will suffer a loss from trading and this suffices to prevent agents
from participating into the trading. This precludes trading motivated by differences
in private information (to this regard, compare also with the result of Exercise 8.5).
Arrow-Debreu and Green-Lucas Equilibria
In view of the no-trade result of Proposition 8.4, we now focus on the case of an
economy where markets are only open at date t D 0, after the observation of
the private signals by the agents. Recall that, after the arrival of this additional
information, agents exhibit heterogeneous beliefs. We consider an economy where
N securities are traded at t D 0. Each security n pays a random dividend dQ n at t D 1,
taking the possible values fd1n ; : : : ; dSn g. We assume that the total supply of each
traded security is normalized to zero. In this setting, for all i D 1; : : : ; I, the optimal
choice problem of agent i can be represented as follows:
!
XS XN
max vs . y /u es C
i i i i
zn dsn ;
i
(8.8)
.zi1 ;:::;ziN /2RN
sD1 nD1
where, as above, vsi . yi / denotes the conditional probability P.!s jQyi D yi /, for every
s D 1; : : : ; S and yi 2 fyi1 ; : : : ; yiJ g. In problem (8.8), the budget constraint is
given by
X
N
zin pn 0; (8.9)
nD1
where .p1 ; : : : ; pN / is the price vector of the N traded securities. Note that problem
(8.8) can be also generalized to the case of state dependent utility functions ui . Given
a realization y D . y1 ; : : : ; yI / 2 Y of the signal random vector yQ , we denote by
zi . yi / 2 RN the optimal solution to (8.8)–(8.9) given the observation of yi by agent
i, for each i D 1; : : : ; I.
In the present setting, the notion of Arrow-Debreu equilibrium can be specialized
as follows (compare with Definition 4.6).
Definition 8.5 Let y D . y1 ; : : : ; yI / 2 Y be a realization of the signal random
vector yQ D .Qy1 ; : : : ; yQ I /. A price-trade vector couple .p . y/; z1 . y1 /; : : : ; zI . yI //,
with p . y/ 2 RN and zi . yi / 2 RN , for all i D 1; : : : ; I, constitutes an Arrow-
Debreu equilibrium contingent on fQy D yg if
(i) for every i D 1; : : : ; I, given the price vector p . y/, the trade vector zi . yi / is a
solution of the optimum problem (8.8)–(8.9) of agent i;
P I
iD1 zn . y / 0, for all n D 1; : : : ; N.
i i
(ii)
According to the above definition, the Arrow-Debreu equilibrium at date t D 0 of
the economy depends on the realization y of the random signal yQ . Indeed, while the
utility functions and the endowments of the agents do not depend on the realization
of the signals, the beliefs of the agents do depend on the signals received and, hence,
also the equilibrium price vector will depend on the realization of the signals.
We always work under the implicit but crucial assumption that all the relevant
ingredients of the economy (i.e., the initial allocation, the utility functions of all
the agents, the prior distributions of the endowment and of the private signals, the
feasibility of a trade vector) are common knowledge among the agents. Unlike in
the analysis developed in Chap. 4, in the present setting the notion of equilibrium is
contingent on the realization of the random signal and the equilibrium price vector
will depend on the signals received by the agents. We represent this property by
introducing a function W Y ! RN such that p . y/ D . y/, for every possible
realization y 2 Y of the signal random vector yQ . We call the equilibrium price
functional.
Since signals are private, each agent cannot observe the realizations of the signals
received by the other agents. However, the fact that equilibrium prices depend on
the information received by every agent is common knowledge. In particular, since
agents have a perfect knowledge of the structure of the economy, they also know
the structure of the equilibrium price functional . In turn, the knowledge of the
equilibrium price functional implies that each agent can try to infer the signals
received by the other agents on the basis of the observation of the equilibrium prices,
which depend on the signals received by all the agents, as explained above.
In other words, Arrow-Debreu equilibrium prices transmit information. In the
classical setting considered in Chap. 4, prices were only used to define the agents’
budget constraints. In the present setting, besides defining the budget constraints,
prices also carry additional information: every agent, when confronted with an
equilibrium price vector p will try to infer the signals received by the other agents
by determining the set of realizations y 2 Y that are compatible with the price
vector p . In this way, each agent will learn that the realization of the signal random
vector yQ belongs to the set fy 2 Y W . y/ D p g. In particular, if the equilibrium
price functional is injective, then every agent can perfectly infer the signals received
by every other agent by simply looking at equilibrium prices. If is not injective,
then an agent will determine the set 1 .p / WD fy 2 Y W . y/ D p g Y which
is compatible with the prices observed. Of course, this represents an additional
information and, being assumed to be rational, agents will revise their beliefs taking
into account the information conveyed by market prices. This revision of the agents’
beliefs leads to a new conditional distribution

v1i . yi ; 1 .p //; : : : ; vSi . yi ; 1 .p // ; for all i D 1; : : : ; I and yi 2 fyi1 ; : : : ; yiJ g;
(8.10)
where, for every realization yi 2 fyi1 ; : : : ; yiJ g of the signal yQ i and for every price
vector p, we denote by vsi . yi ; 1 .p// the probability of !s , from the point of
view of agent i, conditionally on the information contained in the signal yi and
in the price vector p. In general, the conditional distribution (8.10) is different
from the conditional distribution (8.1), since it incorporates an additional source
of information. In turn, this implies that the optimal solution to problem (8.8)–(8.9)
will no longer be optimal, since the optimal behavior of the agents will now be
determined as the solution to
!
X S
i 1 i i X N
max vs y ; . p / u es C
i
zn dsn ;
i
(8.11)
.zi1 ;:::;ziN /2RN
sD1 nD1
for all i D 1; : : : ; I. Therefore, the price vector p D . y/ obtained from the Arrow-
Debreu equilibrium will no longer be of equilibrium. This observation shows that
the notion of Arrow-Debreu equilibrium contingent on a realization of the signal
random vector yQ (see Definition 8.5) is not the appropriate notion of equilibrium,
since the associated price vector leads agents to revise their beliefs and, therefore,
their optimal choices. In the present context, an appropriate notion of equilibrium
should represent a stable point in this process of price observation - revision of
beliefs and agents should not want to revise their behavior once they observe the
market clearing prices.
As outlined in the introduction to this chapter, an appropriate notion of equi-
librium is represented by the Green-Lucas equilibrium (see Lucas [1251], Green
[820]), which captures the idea that at equilibrium there should be no incentive to
revise beliefs and trades. For a price vector p 2 RN , we denote by zi n . yj ; p/ the
i
optimal solution to problem (8.11) for agent i given the observation of the private
signal yij and of the price vector p.
Definition 8.6 Let W Y ! RN and z D fzi W fyi1 ; : : : ; yiJ g RN ! RN I

i D 1; : : : ; Ig. The couple . ; z / constitutes a Green-Lucas equilibrium if the
following two conditions hold, for every realization y D . y1 ; : : : ; yI / 2 Y of the
signal random vector yQ :
the equilibrium price vector p is given by p D . y/;
(i) P
I i i
(ii) iD1 zn . y ; p / 0, for all n D 1; : : : ; N.
Definition 8.6 captures the fact that equilibrium prices transmit information.
Agents are rational and are assumed to know how private signals affect equilibrium
prices and, therefore, optimally take this information into account.
A Green-Lucas equilibrium is said to be fully revealing if the equilibrium
price functional is injective, so that the observation of the equilibrium prices
completely determines the realization of the signals received by all the agents.
Otherwise, we say that the equilibrium is only partially revealing, since the
observation of the equilibrium prices only allows agents to determine a set of
possible signals which are compatible with the observed prices. Of course, in view
of studying market efficiency from an informational point of view, as discussed in
the introduction to this chapter, it is of particular interest to determine under which
conditions a Green-Lucas equilibrium is fully revealing. Indeed, if the equilibrium is
fully revealing, then all available (private) information will be conveyed by market
prices.
Given an economy as described in the present section, in order to determine
a Green-Lucas equilibrium, one can proceed along the following successive steps
(see also Brunnermeier [319, Section 1.2.1]):
• define the agents’ prior beliefs, the prior distributions of all the random variables
appearing in the model and specify a conjectured price functional ;
• taking the conjectured price functional as given, derive the posterior beliefs of
each agent, conditionally on the observation of the price vector and of the private
signal, and compute the optimal trade vector for each agent, as a function of the
price functional, of the signals received and of the endowment;
• impose the market clearing condition to derive the actual relation between the
conjectured price functional and the signals;
• impose the rational expectations condition (i.e., the conjectured price functional
corresponds to the true price functional; in other words, the price conjecture must
be self-fulfilling) in order to fully determine the equilibrium price functional.
Under suitable conditions, a closed form Green-Lucas equilibrium exists. Classical
conditions to find a closed form equilibrium with a linear price functional are
multivariate normality together with exponential utility functions (compare also
with the models presented in Chap. 10). In the following sections, we shall
consider several instances where the above steps can be solved and a Green-Lucas
equilibrium can be explicitly constructed.
8.2 On the Possibility of Informationally Efficient Markets 415
8.2 On the Possibility of Informationally Efficient Markets
As we have seen at the end of the previous section, equilibrium prices convey
information about the private signals received by the agents. Under suitable
conditions, a Green-Lucas equilibrium can be fully revealing. If this is the case, then
equilibrium prices aggregate and transmit all private information and markets are
informationally efficient (in a strong form). Sufficient conditions for Green-Lucas
equilibrium prices to fully reveal the private information of the agents have been
established in a series of papers, see Grossman [839, 840, 841].
To analyse the aggregation-transmission of information by Green-Lucas equi-
librium prices, it is useful to introduce an artificial economy, where every agent
is assumed to observe all private signals received by all agents. In this artificial
economy, each agent solves the optimal choice problem (8.8), with the probability
distribution of the state of the world being conditioned on the realization of the
whole signal random vector yQ D .Qy1 ; : : : ; yQ I /. As in Sect. 8.1, we assume that the
signal random vector takes values in a set Y and denote by y D . y1 ; : : : ; yI / 2 Y a
generic realization of yQ . Similarly as in (8.1), for y 2 Y , we define the conditional
probability distribution v. y/ D .v1 . y/; : : : ; vS . y//, with
vs . y/ WD P.!s jQy D y/ D P.!s jQy1 D y1 ; : : : ; yQ I D yI /; for all s D 1; : : : ; S;

(8.12)
representing the probability of the state of the world !s conditionally on the

realization fQy D yg. Observe that in the artificial economy there are no heterogenous
beliefs, since every agent observes the full set of private signals.
Similarly as in Grossman [841], we assume that at t D 0 there exist S open
markets for the S Arrow securities corresponding to the S states of the world (hence,
markets are complete). As in Sect. 8.1, the economy is populated by I agents with
strictly increasing and strictly concave utility functions ui W RSC ! R satisfying
0 0
the Inada conditions (i.e., limx&0 ui .x/ D C1 and limx!C1 ui .x/ D 0, for all
i D 1; : : : ; I). We also assume that the initial endowment of every agent is strictly
positive. For every i D 1; : : : ; I, we denote by zi . y; p/ the optimal solution to
problem (8.8) in correspondence of the vector of prices p 2 RS and of the probability
distribution (8.12) conditional on the realization fQy D yg. We adopt the notation
zi . y; p/ in order to emphasize the dependence on the vector of prices p. Note
that the Inada conditions ensure that the optimal consumption is strictly positive
in correspondence of each state of the world.
For this artificial economy, we can adopt the notion of Arrow-Debreu equilibrium
contingent on the realization of the signal random vector (see Definition 8.5). Under
the present assumptions, it can be shown that, for every realization y 2 Y of the
signal random vector yQ , the artificial economy admits an Arrow-Debreu equilibrium
contingent on fQy D yg. In particular, the vector pa . y/ of Arrow-Debreu equilibrium
prices of the artificial economy (where the superscript a stands for artificial) depends
on the realization y of yQ and, similarly as in the discussion following Definition 8.5,

we can regard pa ./ W Y ! RS as an equilibrium price functional.
The following lemma provides a sufficient condition for the equilibrium price
functional pa to be injective (compare also with Laffont [1154, Chapter 9]).
Lemma 8.7 Let pa W Y ! RS denote the price functional associated to an Arrow-
Debreu equilibrium of the artificial economy introduced above and suppose that
vs . y/ > 0, for all s D 1; : : : ; S and y 2 Y . Let y; y0 2 Y be two realizations of yQ
such that v. y/ ¤ v. y0 /. If

zj y; pa . y/ ¤ zj y0 ; pa . y0 / ; for some j 2 f1; : : : ; Ig;
then it holds that pa . y/ ¤ pa . y0 /.

Proof Arguing by contradiction, suppose that pa . y/ D pa . y0 / DW pN 2 RS ,
meaning that the two realizations y and y0 generate the same equilibrium prices
and, consequently, the same budget constraint. Then, it holds that
X
S
XS

0
vs . y/ui eis C zi
s . y; N
p / vs . y/ui eis C zi
s .y ;p
N/ ;
sD1 sD1
for all i D 1; : : : ; I, with strict inequality holding for i D j, due to the strict concavity
of the utility function uj . For brevity of notation, let us denote zi s WD zs . y; p N / and
i
i0 i 0
zs WD zs . y ; pN /, for all s D 1; : : : ; S and i D 1; : : : ; I. The concavity of ui , for all
i D 1; : : : ; I, implies that
X
S
0
i0 i0 i
vs . y/ zi
s zs u es C zi
s 0; for all i D 1; : : : ; I; (8.13)
sD1
0
with ui denoting the first derivative of the utility function ui and where a strict
inequality holds for i D j. The optimality conditions for problem (8.8) in
correspondence of y0 , recalling that the optimal consumption is strictly positive (as
a consequence of the Inada conditions), imply that
0 0
vs . y0 /ui .eis C zi
s /Dp
i
N s; for all s D 1; : : : ; S and i D 1; : : : ; I; (8.14)
where i denotes the Lagrange multiplier associated to the optimal choice problem
of agent i, for i D 1; : : : ; I. By combining conditions (8.13) and (8.14), it follows
that
X
S
i Ns
i0 p
vs . y/ zi
s zs 0; for all i D 1; : : : ; I;
sD1
vs . y0 /
with strict inequality holding for i D j. In turn, dividing by i and summing over all
agents, this implies that
X
S
pN s vs . y/ X
I
0
.zi
s zs / > 0;
i
sD1
vs . y0 / iD1
PI PI i0
which is in contradiction with the market clearing condition i
iD1 zs D iD1 zs ,
for s D 1; : : : ; S. t
u
Lemma 8.7 shows that, in the artificial economy where each agent observes the
full set of private signals, if different posterior distributions lead to different optimal
demands then the equilibrium prices must be different. In turn, this result allows us
to prove the following proposition, which shows that the Arrow-Debreu equilibrium
of the artificial economy corresponds to a Green-Lucas equilibrium of the original
economy.
Proposition 8.8 Let pa W Y ! RS denote the price functional associated to an
Arrow-Debreu equilibrium of the artificial economy introduced above and suppose
that vs . y/ > 0, for all s D 1; : : : ; S and y 2 Y . Then pa is a Green-Lucas
equilibrium price functional of the original economy in which each agent i observes
his private signal and the equilibrium prices.
Proof Define the candidate Green-Lucas equilibrium price functional W Y !
RS by . y/ WD pa . y/, for every y 2 Y . We want to prove that the definition

zi yi ; . y/ WD zi y; pa . y/ ; for all i D 1; : : : ; I; (8.15)
for every realization y D . y1 ; : : : ; yI / 2 Y , is well posed and, together with the

functional , defines a Green-Lucas equilibrium of the original economy, in the
sense of Definition 8.6. In (8.15), the right hand side represents the optimal demand
of agent i in the artificial economy, while the left hand side represents the candidate
optimal demand of agent i in the original economy, associated to the candidate
equilibrium price functional .
Suppose first that the functional pa W Y ! RS is injective. In this case, the
observation of pa . y/ corresponds exactly to the observation of the realization
fQy D yg of the whole set of signals. Therefore, the conditional distribution v. y/
considered in the artificial economy (see (8.12)) coincides with the conditional
distribution introduced in (8.10). Indeed, for all i D 1; : : : ; I, it holds that

vsi yi ; ;1 . pa . y// D P.!s jQyi D yi ; yQ D y/ D P.!s jQy D y/ D vs . y/; (8.16)
for all i D 1; : : : ; I, s D 1; : : : ; S and y 2 Y . In particular, this shows that the

information available to each agent in the original economy coincides with the
information available in the artificial economy, where every agent observes the full
set of private signals. In view of the formulation of problem (8.11), the claim then
follows.
Suppose now that pa is not injective. Clearly, there is nothing to prove if, for
every i D 1; : : : ; I, s D 1; : : : ; S and y 2 Y , relation (8.16) holds. Therefore, let us
assume that for some i 2 f1; : : : ; Ig and j 2 f1; : : : ; Jg and pN 2 RS , the set
Yj i WD fy 2 Y W yi D yij and pa . y/ D pN g
contains at least two elements y and y0 for which v. y/ ¤ v. y0 /. In other words, the
set Yj i contains all the realizations of the signal random vector yQ which generate
different conditional distributions but are indistinguishable from the point of view
of agent i, given the observation of his private signal yij and the vector of prices
pN . In view of Lemma 8.7, since pa . y/ D pa . y0 / D pN , it necessarily holds that
zi . y; pN / D zi . y0 ; pN /. In other words, the optimal demand of agent i in the artificial
economy is constant across all the elements of the set Yj i and, hence, the definition
zi . yi ; . y// WD zi . y; pa . y// is well posed. Since market clearing holds in the
artificial economy and, hence, in the original economy as well, we have constructed
a Green-Lucas equilibrium of the original economy in the sense of Definition 8.6.
t
u
Together with Lemma 8.7, the proof of the above proposition shows that, if two
realizations y and y0 of the signal random vector yQ generate different conditional
distributions v. y/ and v. y0 / but generate the same Arrow-Debreu equilibrium prices
in the artificial economy, then the optimal demand of every agent in the artificial
economy will be the same in correspondence of y and y0 . In turn, this implies
that the optimal demand at the equilibrium of the original economy, where each
agent observes his private signal and the market prices, coincides with the optimal
demand at the equilibrium in the artificial economy. In particular, a Green-Lucas
equilibrium in the original economy can be constructed by considering the Arrow-
Debreu equilibrium in the artificial economy.
We have reached an important result. Indeed, under the present assumptions,
in correspondence of equilibrium prices all information is embedded in market
prices and, hence, the market is efficient in a strong form from an informational
point of view. Indeed, while the equilibrium price functional is not necessarily
injective (and, hence, agents are not always able to infer the realization of the
whole random vector yQ given the observation of equilibrium prices), the optimal
demands of the agents are constant across all realizations of yQ that lead to the
same equilibrium prices. Hence, under the assumption of complete markets, in
a rational expectation equilibrium the price system fully symmetrizes all the
differences in information among the agents. Moreover, if prices fully symmetrize
the informational differences, then agents will only trade for hedging purposes
and there is no speculation, in the sense that there is no trading activity purely
motivated by differences in information/beliefs. An explicit example of a fully
revealing Green-Lucas equilibrium will be provided at the end of this section.
The relation between Green-Lucas equilibria of the original economy with

private information and Arrow-Debreu equilibria of the artificial economy can
be analysed in a more general setting where the private signals are represented
by random variables admitting probability density functions. To this end, let us
introduce the notion of sufficient statistic. Let zQ, xQ and yQ be three random variables
and denote respectively by z, x and y their generic realizations. We say that zQ is
a sufficient statistic for the conditional density f . y; zjx/ if and only if there exist
functions g1 ./ and g2 ./ such that, for every .z; x; y/, it holds that
f . y; zjx/ D g1 . y; z/g2 .z; x/:
As shown in Exercise 8.6, zQ is a sufficient statistic for the conditional density

f . y; zjx/ if and only if the conditional distribution of xQ given fQy D y; zQ D zg does
not depend on y (see also Huang & Litzenberger [971, Section 9.8]).
The notion of sufficient statistic can be applied in the context of a Green-
Lucas equilibrium. More specifically, assuming that the market comprises N traded
securities paying dividends .dQ 1 ; : : : ; dQ N / and given a vector of equilibrium prices p ,
we say that p is a sufficient statistic if the probability distribution of the dividends
.dQ 1 ; : : : ; dQ N / conditioned on the observation of the private signal yQ i and of the prices
p does not depend on the realization of the private signal yQ i , for each i D 1; : : : ; I.
In this sense, if p is a sufficient statistic, then the information conveyed by the
private signals becomes redundant once agents observe market prices. This property
implies that equilibrium prices embed all useful information and, hence, the market
is informationally efficient.
In Grossman [840], an economy with N C 1 traded assets and I agents is
considered. The N C 1 assets are available for trading at t D 0 and include N
risky assets (with dividends represented by the random variables .dQ 1 ; : : : ; dQ N /) and
one riskless asset yielding the riskless return rf . As above, each agent i D 1; : : : ; I
observes at t D 0 (before the opening of the market) a private signal represented
by the random variable yQ i . In Grossman [840], it is assumed that the random vector
.dQ 1 ; : : : ; dQ N ; yQ 1 ; : : : ; yQ I / is distributed according to a multivariate normal law and,
under this distributional assumption, the following results are obtained:
• there exists a Green-Lucas equilibrium price functional which is a sufficient
statistic if and only if there exists an equilibrium price functional of the artificial
economy (where every agent observes all private signals) which is a sufficient
statistic (see Exercise 8.8 for a proof of this property). Note that this result does
not require the assumption of a multivariate normal distribution;
• the vector of expected dividends conditioned on the observation of yQ (i.e., the
vector composed by mn . y/ WD EŒdQ n jQy D y, for n D 1; : : : ; N) is a sufficient
statistic;
• the equilibrium price functional of the artificial economy, whenever it
exists, depends on the realization of the signals only through the vector
.m1 . y/; : : : ; mN . y//.
• as a consequence of the multivariate normality assumption, agents act as if there

were only two assets: the riskless asset and a mutual fund composed by the N
risky assets (two mutual funds separation; see Chap. 3). Moreover, if there exists
an equilibrium price functional for the artificial economy, then it is an invertible
function of .m1 . y/; : : : ; mN . y// and a sufficient statistic. Under these conditions
and if the demand of the risky mutual fund is a decreasing function of its price
(i.e., the mutual fund is not a Giffen good), then there exists a Green-Lucas
equilibrium which is a sufficient statistic (see Grossman [840, Theorem 1]).
A sufficient condition for the risky mutual fund not to be a Giffen good is
represented by a non-increasing coefficient of absolute risk aversion of the agents’
utility functions. Together with the joint normality assumption, this condition
suffices to ensure that the rational expectations equilibrium price is a sufficient
statistic. In particular, note that the results of Grossman [840] do not require
markets to be complete. Moreover, generalizing a previous result in Grossman
[839], Grossman [840] shows that, if the rational expectations equilibrium price
is a sufficient statistic, then the equilibrium allocation cannot be Pareto improved by
a central planner having access to all available information. In this sense, Grossman
[840] argues that the appropriate notion of informationally efficient market is the
one in which the price vector is a sufficient statistic.
Of course, not every Green-Lucas equilibrium price functional, in the sense
of Definition 8.6, is fully revealing nor corresponds to informationally efficient
markets. Sometimes, only partially revealing equilibria exist. For an example of
an economy which does not admit fully revealing equilibria, admits fully revealing
equilibria or partially revealing equilibria depending on the preference parameters,
we refer to Exercise 8.7 and to Radner [1438]. A simple model where a fully
revealing Green-Lucas equilibrium does not exist is reported in Exercise 8.12.
The Blackwell and Hirshleifer Effects
In the first part of Sect. 8.1, we have analysed the value of information from the
point of view of an individual decision maker. From this perspective, we have shown
that information always has a non-negative value (see Proposition 8.1). Let us now
analyse the value of information from a social point of view. In a nutshell, we aim
at understanding whether the introduction of additional information can increase or
decrease the social welfare in the economy.
As argued above, a fully revealing Green-Lucas equilibrium implies market
efficiency in a strong form from the informational point of view. Moreover, from
the allocative point of view, a fully revealing Green-Lucas equilibrium under market
completeness is also ex-post Pareto efficient, as a direct consequence of the First
Welfare theorem (see also Grossman [841]). However, interim Pareto optimality
and ex-ante Pareto optimality do not necessarily hold (see Laffont [1153] for
an example of a Green-Lucas equilibrium which is interim Pareto inefficient).

Moreover, partially revealing equilibria are typically ex-post Pareto inefficient (see
Laffont [1153]).
The study of the social impact of information in financial markets requires the
understanding of two competing effects: the Blackwell effect and the Hirshleifer
effect. As we have already pointed out, better information always leads to better
individual choices and, hence, to a higher personal welfare. This is indeed the con-
tent of Propositions 8.1 and 8.2 (Blackwell effect). On the contrary, the Hirshleifer
effect is due to a lack of optimal risk sharing: Hirshleifer [948] remarked that the
revelation of information can eliminate profitable risk sharing opportunities. In order
to explain the Hirshleifer effect, let us consider the simple situation (as illustrated
in Sect. 4.1) where there is no aggregate risk and agents do not have private
information. In a complete market economy, the agents are willing to trade in order
to share the individual risks, thus improving their expected utilities. In particular,
in the absence of aggregate risk, they can fully eliminate the randomness of their
future consumption plans. However, if the true state of the world is revealed before
the opening of the markets, then optimal risk sharing becomes impossible, since
the agent whose endowment is larger in the realized state of the world would not
participate to the trade. This elementary example shows that an equilibrium where
the information is revealed cannot be Pareto efficient, since it implies a sub-optimal
risk sharing to the extent that it induces agents not to trade at all. In other words,
realized risks cannot be insured and, hence, the revelation of information in the
economy may eliminate profitable insurance opportunities. Due to the Hirshleifer
effect, information can be detrimental from the social point of view and the Pareto
efficiency of the equilibrium allocation can in some cases be improved if the
equilibrium is only partially revealing. An example of a simple economy where
the Hirshleifer effect is particularly evident is given in Exercise 8.9.
The above discussion can be rephrased in the setting introduced at the beginning
of the present section. Let us consider an economy with I agents who can all observe,
before the opening of the markets, the same signal yQ , here represented by a random
variable taking values in the set fy1 ; : : : ; yJ g, for some J 2 N, with corresponding
probabilities .. y1 /; : : : ; . yJ //. We assume that the complete set of S Arrow
securities is available for trading (complete market economy). Conditionally on a
realization fQy D yj g of the common signal, let us denote by fxi . yj / 2 RSC I i D
1; : : : ; Ig an Arrow-Debreu equilibrium allocation contingent on the realization
fQy D yj g (see Definition 8.5), where xi s . yj / denotes the optimal consumption of
agent i in the state of the world !s conditionally on the observation of fQy D yj g. In
particular, note that this equilibrium allocation is allowed to be contingent on the
realization of the random signal. Let us also denote by fxi;0 2 RSC I i D 1; : : : ; Ig
the “uninformative” equilibrium allocation reached before the agents are allowed
to observe the common signal. We can now establish the following proposition,
which can be proved by the same arguments used in the first part of the proof of
Proposition 8.4 (see also Gollier [800, Proposition 97]).
Proposition 8.9 Under the above assumptions, for every j D 1; : : : ; J, the equilib-
rium allocation fxi . yj / 2 RSC I i D 1; : : : ; Ig contingent on the signal realization
fQy D yj g observed by the agents cannot ex-ante Pareto dominate the allocation
fxi;0 2 RSC I i D 1; : : : ; Ig obtained when the agents cannot observe yQ .
The above proposition shows that when more precise (common) information
becomes available in the economy, then it cannot be beneficial to all market
participants. There exists at least one agent whose expected utility decreases after
the arrival of new information.
In general, the Blackwell and the Hirshleifer effects go in opposite directions.
While information increases the individual expected utility, it may lead to sub-
optimal risk-sharing and, therefore, it may reduce the social welfare. Indeed, as
shown in Gottardi & Rahi [813], the effect on social welfare of a change in the
information available prior to trading can be either beneficial or adverse. Schlee
[1504] shows that the introduction of information is Pareto inferior if one of
the following conditions is satisfied: there exists a risk neutral agent, there is no
aggregate risk or agents have linear absolute risk tolerance with common slope
(see also Gollier [800, Proposition 98] and Exercise 8.10). Eckwert & Zilcha [626]
show that, in an economy with production, information has a positive value if the
information refers to non-tradable risks, while information related to tradable risks
may reduce the welfare if the consumers are strongly risk averse.
Due to the Hirshleifer effect, in an economy with asymmetric information
(informed and uninformed agents), an incomplete financial structure may be
optimal, as shown in Marin & Rahi [1305]. In that paper, uninformed agents face
an adverse selection problem due to the presence of informed agents. In particular,
uninformed agents are afraid to trade with more informed agents. In this context,
the revelation of information induces two different effects: the Hirshleifer effect
and a reduction of the adverse selection phenomenon. As discussed above, the
Hirshleifer effect leads to negative consequences for the welfare of the economy,
while the second effect has positive consequences for the uninformed agents. In the
model considered in Marin & Rahi [1305], markets are initially incomplete and
the equilibrium is only partially revealing. At a second stage, a new security is
introduced in order to complete the market, leading to a fully revealing equilibrium.
When passing from the partially revealing to the fully revealing equilibrium, the
Hirshleifer effect and the adverse selection reduction come into play and the overall
impact on the social welfare depends on the relative strength of these two effects.
If the Hirshleifer effect outweighs the reduction of adverse selection, then it is
better to have incomplete markets and a partially revealing equilibrium, because
less information is revealed and this leads to better risk sharing possibilities. Note
also that the two effects also affect the trading volume. In order to add noise to
the information transmitted by prices, it may be optimal to introduce a purely
speculative security unrelated to endowments and preferences (see Marin & Rahi
[1304]). Related results have been obtained in Dow & Rahi [586], assuming that
the asset is not perfectly correlated with the initial risk exposure represented by
the agents’ endowment. In this context, it is confirmed that greater revelation of
information that agents wish to insure against reduces their hedging opportunities
(negative Hirshleifer effect). On the contrary, early revelation of information that is
uncorrelated with hedging needs allows agents to hedge risks more efficiently.
A Fully Revealing Equilibrium
The existence of a fully revealing equilibrium has been established in the context
of a tractable model in Grossman [839]. Let us consider a two-period economy
(t 2 f0; 1g), with a single consumption good which is consumed at t D 1. There
are two traded assets: a riskless asset paying the constant rate of return rf and a
risky asset delivering a random dividend dQ at t D 1. The price of the risky asset
at t D 0 is denoted by p and the total supply of the riskless asset is normalized
to zero. Similarly as above, we assume that the economy is populated by I agents
i D 1; : : : ; I. The initial endowment
Pof agent i is expressed in terms of ei units of
I
the risky asset and we suppose that iD1 ei D 1.
Letting wi be the demand of the risky asset by agent i (in terms of units of the
asset), his wealth at date t D 1 is described by the random variable
e i D .ei wi /prf C wi d;
W Q for all i D 1; : : : ; I:
We assume that each agent i is characterized by a negative exponential utility

function of the form ui .x/ D ea x , with ai > 0, for all i D 1; : : : ; I.
i
Ex-ante (at t D 1), all the agents believe that dQ is distributed as a normal random
variable with mean dN and variance 2 . At the interim date (at t D 0), each agent i
observes a private signal yQ i of the form
yQ i D dQ C "Qi ;
where "Qi is a random variable such that the random vector .d; Q "Q1 ; : : : ; "QI / is
distributed according to a normal multivariate law with uncorrelated (and, hence,
independent) components. As above, we denote by yQ the random vector . yQ 1 ; : : : ; yQ I /,
with generic realization y D . y1 ; : : : ; yI / 2 RI . We assume that EŒQ"i D 0
and we denote by "Q2i the variance of "Qi , for all i D 1; : : : ; I. The I agents have
private information and, after the observation of their signals, revise their beliefs
via the Bayes rule. More specifically, by the properties of the multivariate normal
distribution, for every i D 1; : : : ; I, the conditional distribution of dQ given yQ i is
normal and characterized by
Q i D dN C ˇ i . yi d/
EŒdjy N and Q i / D 2 .1 ˇ i /;
Var.djy
where we denote EŒdjy Q i WD EŒdjQQ yi D yi , similarly for the conditional variance

Q /, and where ˇ WD =. 2 C 2i /, for all i D 1; : : : ; I. Note that the
Var.djy i i 2
"Q
conditional variance does not depend on the realization of the signal. After agents
observe their private signals, markets open and trading occurs. As a consequence
of formula (3.12), the optimal demand of agent i given the observation of the
realization fQyi D yi g of his private signal is given by
Q i prf
EŒdjy
wi D : (8.17)
Q i/
ai Var.djy
Observe that, besides the properties discussed in section “Closed Form Solutions
and Mutual Fund Separation”, the optimal demand wi is an increasing function of
the precision of the information (if EŒdjyQ i prf ), here measured by the quantity
Q
1= Var.djy /.
i
PI
The market clearing condition iD1 w
i
D 1 then implies that, for every
realization y D . y ; : : : ; y / 2 R , the equilibrium price p of the risky asset is
1 I I
given by
0 PI EŒdQ jyi !1 1
X
I
1 A 1:
iD1 ai Var.dQ jyi /
p . y/ D @P (8.18)
I 1 Q i/
ai Var.djy rf
iD1 ai Var.dQ jyi / iD1
The price p computed in equation (8.18) represents the Arrow-Debreu equilib-

rium price contingent on fQy D yg (see Definition 8.5). Moreover, p . y/ has several
nice and intuitive features: it is related to a weighted average of the conditional
expectations of the dividend by the I agents, where the weight associated to each
agent is given by the reciprocal of the coefficient of absolute risk aversion multiplied
by the agent’s estimate of the variance of the dividend. Hence, the more an agent
is risk averse and the smaller is the precision of his private information, the smaller
is his weight in the determination of p . y/. The rationale for this property is that
an agent exhibiting a large coefficient of absolute risk aversion and with a large
conditional variance will also exhibit a small demand of the risky asset and, hence,
will not influence significantly equilibrium prices.
The Arrow-Debreu equilibrium price p . y/ is a function of the realization of
the signals yQ D .Qy1 ; : : : ; yQ I /. Hence, as discussed above, agents can learn about
the other agents’ private signals by observing equilibrium prices. Motivated by this
observation, we now aim at deriving the Green-Lucas equilibrium of this economy
in the presence of heterogeneous information.
Let us assume that "Q2i D "2 , for all i D 1; : : : ; I, for some "2 > 0. Under this
additional assumption, there exists a fully revealing Green-Lucas equilibrium, due to
the normality assumption P together with the fact that agents exhibit constant absolute
risk aversion. Let yQ WD . IiD1 yQ i /=I denote the average of the signals received by
the agents. By exploiting the assumption of multivariate normality, it can be easily
shown that yQ is a sufficient statistic for the distribution of .Qyi ; y/
Q conditioned on d,
Q
for all i D 1; : : : ; I (see Grossman [839, Lemma 1]). This implies that
Q yi D yi ; yQ D y DW EŒdjy
EŒdjQ Q i ; y D EŒdjy N d/;
Q D dN C ˇ.y N
(8.19)
Q yi D yi ; yQ D y/ DW Var.djy
Var.djQ Q i ; y/ D Var.djy/ N
Q D 2 .1 ˇ/;
for every realization yi and y of the random variables yQ i and yQ and where we define
ˇN WD 2 =. 2 C 2 =I/. Moreover, one can also show that yQ is a sufficient statistic for
Q conditioned on dQ and it holds that
the joint distribution of .Qy1 ; : : : ; yQ I ; y/
EŒdjy Q
Q 1 ; : : : ; yI ; y D EŒdjy and Q 1 ; : : : ; yI ; y/ D Var.djy/:
Var.djy Q
We are now in a position to establish the following result, which corresponds to

Grossman [839, Theorem 1].
Proposition 8.10 In the context of the model described above (see Grossman
[839]), there exists a fully revealing Green-Lucas equilibrium, characterized by the
equilibrium price functional W RI ! R defined by
!1
N d/
dN C ˇ.y N N X
2 .1 ˇ/
I
1
1
.y ;:::;y / D I
i
: (8.20)
rf rf iD1
a
Proof In view of Proposition 8.8, to show that (8.20) defines a Green-Lucas

equilibrium price functional of the economy, we first show that it corresponds to the
equilibrium price functional in an artificial economy where every agent observes all
the private signals yQ D .Qy1 ; : : : ; yQ I /. As a preliminary, recall that yQ is a sufficient
Q conditioned on d.
statistic for the joint distribution of .Qy1 ; : : : ; yQ I ; y/ Q Hence, the
equilibrium of the artificial economy where every agent observes the full set of
private signals coincides with the equilibrium of an artificial economy where every
agent only observes the sufficient statistic y. Q In this artificial economy, the optimal
demand of agent i at equilibrium is given similarly as in (8.17) by
Q pa .y/rf
EŒdjy
wi D ; for every i D 1; : : : ; I;
Q
ai Var.djy/
where pa .y/ denotes the equilibrium price corresponding to the artificial economy
contingent on the realization fyQ D yg. Imposing the market clearing condition, this
leads to the price functional
!1
Q
EŒdjy XI
1 1
p . y1 ; : : : ; yI / WD p .y/ D
a a Q
Var.djy/ ;
rf iD1
ai rf
which corresponds to the right hand side of (8.20), making use of the conditional
expectation and variance given in (8.19). It order to show that the Green-Lucas
equilibrium price functional of the original economy corresponds to pa , it

suffices to compute the optimal demand of every agent in the original economy
when the price of the risky asset is equal to pa .y/. Indeed, noting that the map
y 7! pa .y/ is invertible, it holds that
Q i ; pa .y/ pa .y/rf

EŒdjy Q i ; y pa .y/rf
EŒdjy
D
Q i ; pa .y//
ai Var.djy Q i ; y/
ai Var.djy
Q pa .y/rf
EŒdjy
D D wi ;
Q
ai Var.djy/
where we have used the fact that yQ is a sufficient statistic for the joint distribution
Q conditioned on d,
of .Qyi ; y/ Q for every i D 1; : : : ; I. We have thus shown that the
optimal demand in the original economy of each agent i, when faced with the price
functional pa DW , corresponds exactly to wi . Since the price functional clears
the market, this proves the proposition. t
u
The equilibrium price functional obtained in Proposition 8.10 has several
intuitive properties. For instance, we see from (8.20) that is increasing with
respect to the average signal y observed by the agents and is decreasing with respect
to the coefficients of risk aversion of the agents. Moreover, the equilibrium price
is a sufficient statistic and, therefore, the private information becomes redundant
once agents are able to observe the equilibrium price. This shows that, as we
have already remarked after Proposition 8.8, the equilibrium price functional
eliminates the heterogeneity in the agents’ information and implies that the market
is informationally efficient.
However, as remarked in Admati [16], the fully revealing equilibrium constructed
in Proposition 8.10 presents some problems. First, we can observe that the optimal
demands of the agents are independent of their initial wealth (as a consequence of
the assumption of exponential utility functions), of the private information (indeed,
as a consequence of the fact that the equilibrium is fully revealing, the optimal
demand only depends on the sufficient statistic) and, most surprisingly, of the price
of the risky asset in equilibrium (compare also with Exercise 8.11). In particular,
the independence of the optimal demand on the private information leads to a
paradox that will be discussed later in this chapter: since prices perfectly aggregate
and transmit information, there is no incentive for the agents to collect costly
private information. But then it is difficult to understand how equilibrium prices can
aggregate private information if the optimal demands of the agents are independent
from their private information. If equilibrium prices are a sufficient statistic for the
private information, agents do not have any incentive to collect private information,
thus leading to the paradox that there is no private information to aggregate. The
property that in equilibrium the optimal demand is independent of the price of the
risky asset is due to the fact that, in addition to the classical substitution and income
effects, in the present context a price change also has an information effect. The
substitution and information effects go in opposite directions: on the one hand, if
8.3 On the Impossibility of Informationally Efficient Markets 427
agents’ expectations are constant, a price increase induces the agents to invest more
in the riskless asset; on the other hand, a price increase is also interpreted as a signal
that the risky asset is more valuable and the consequent revision of beliefs offsets
the substitution effect. DeMarzo & Skiadas [555] show that the equilibrium of the
Grossman [839] model is unique, but minor changes in the normality assumption
lead to indeterminacy and partially revealing equilibria.
8.3 On the Impossibility of Informationally Efficient Markets
In the previous section, we have presented conditions under which Green-Lucas

equilibrium prices transmit and aggregate private information and make agents’
beliefs homogeneous, thereby providing a microfoundation to the efficient markets
theory. However, by means of the example discussed at the end of the previous
section, we have also seen that a fully revealing equilibrium is problematic, in the
sense that, if equilibrium prices are a sufficient statistic, then the agents’ optimal
demands no longer depend on the private information and agents have no incentive
to collect costly private information.
This paradox does not occur if equilibrium prices are only partially revealing.
Indeed, if equilibrium prices are not a sufficient statistic for the agents’ private
information, then the optimal demand will depend not only on market prices but
also on the private information and, hence, agents have an incentive to collect
private information, since it can improve the optimal expected utility. In this case,
market efficiency does not hold in the strong sense. Typical settings where a rational
expectations equilibrium is only partially revealing are represented by the presence
of random supply of the assets, costly observation of private information and
imperfect competition. In this section, we focus our attention on the first two cases,
referring to Chap. 10 for a discussion of the role of imperfect competition.
Random Supply and Heterogeneous Information
Following Hellwig [931], let us consider a generalized version of the model

introduced in Grossman [839] and discussed in Sect. 8.2. As before, we assume
that the economy is populated by I 2 N agents i D 1; : : : ; I, with preferences
characterized by the utility functions ui .x/ D ea x , with ai > 0, for all i D
i
1; : : : ; I. In the economy, there are two traded assets: a riskless asset paying the
constant rate of return rf > 0 and a risky asset delivering the random dividend dQ at
date t D 1. The total supply of the riskless asset is normalized to zero, while the total
supply of the risky asset is assumed to be random and given by the realization of
the random variable uQ . The total supply of the risky asset can be random due to the
presence of random trading by agents trading for liquidity reasons (noise or liquidity
traders, see also Chap. 10). Each agent i D 1; : : : ; I observes a private signal of the
form
yQ i D dQ C "Qi :
We assume that the random vector .Qu; d; Q "Q1 ; : : : ; "QI / is distributed according to a
multivariate normal law, with mean vector .Nu; d; N 0; : : : ; 0/ and a diagonal covariance
2 2 2 2
matrix with the vector .u ; d ; "1 ; : : : ; "I / along the diagonal. In particular, this
implies that the random variables .Qu; d; Q "Q1 ; : : : ; "QI / are mutually independent.
Ex-ante (i.e., at date t D 1), all agents share the same beliefs and the structure of
the economy is assumed to be common knowledge. In particular, every agent knows
that the total market supply of the risky asset is a realization of the random variable
uQ but, of course, cannot observe the exact realization of uQ . In this context, we can
establish the following proposition, which derives a partially revealing equilibrium
price functional.
Proposition 8.11 In the context of the model described above (see Hellwig [931]),
there exists a partially revealing Green-Lucas equilibrium, characterized by the
equilibrium price functional W RIC1 ! R defined by
X
I
.u; y1 ; : : : ; yI / D 0 C i yi u; (8.21)
iD1
where the coefficients 0 ; i 2 R, i D 1; : : : ; I, and ¤ 0 can be explicitly

determined as the solutions of a system of non-linear equations (see Hellwig [931]).
Proof In order to construct the Green-Lucas equilibrium price functional, we follow
the steps outlined at the end of Sect. 8.1. First, we conjecture a linear price functional
of the form (8.21). Given this conjectured price functional, we then compute the
agents’ optimal demands, given the observation of the price and of the private
signals. Finally, we aggregate the individual optimal demands and impose market
clearing in order to characterize the coefficients appearing in (8.21).
As a preliminary, note that the random variable Q WD .Qu; yQ 1 ; : : : ; yQ I / given
as in the right hand side of (8.21) is normally distributed, being a linear com-
bination of normally distributed random variables. Moreover, the random vector
Q is multivariate normal. By the properties of the multivariate normal
Q yQ 1 ; : : : ; yQ I ; /
.d;
law, the conditional distribution of dQ given the observation of fQ D g and fQyi D yi g,
for some 2 R and yi 2 R, i 2 f1; : : : ; Ig, is again normal, with conditional mean
and variance given by
Q yi D ˛0i C ˛1i yi C ˛2i

EŒdj; and Q yi / D ˇi ;
Var.dj;
where the coefficients ˛0i , ˛1i , ˛2i and ˇi can be explicitly computed, for i D
1; : : : ; I, and depend on the parameters of the model and on the coefficients
appearing in the right hand side of (8.21). Under the conjecture that the equilibrium
price is given by , the optimal demand of each agent i D 1; : : : ; I is then given by,
similarly as in (8.17),
Q yi rf
EŒdj; ˛0i C ˛1i yi C .˛2i rf /
wi D D :
Q yi /
ai Var.dj; ai ˇi
Market clearing implies that, for every realization fQu D ug of the random aggregate
supply of the risky asset, it holds that
X
I X
I
˛0i C ˛1i yi C .˛2i rf /
uD w D
i
:
iD1 iD1
ai ˇi
Solving for then gives

!1 !
XI
rf ˛2i X
I
˛0i C ˛1i yi
D u :
iD1
ai ˇi iD1
ai ˇi
In turn, the last relation allows us to identify the coefficients appearing in (8.21) as
the solutions to
!1
XI
rf ˛2i
D ;
iD1
ai ˇi
˛1i
i D ; (8.22)
ai ˇi
XI
˛0i
0 D iˇ
;
iD1
a i
for all i D 1; : : : ; I. Recalling that the coefficients ˛0i , ˛1i , ˛2i and ˇi , for every
i D 1; : : : ; I, in turn depend on 0 , i and , system (8.22) represents a system of
non-linear equations for 0 , i and , which can be explicitly solved as shown in
Hellwig [931, Section 3]. It remains to show that ¤ 0. Arguing by contradiction,
suppose that D 0. In that case, (8.22) would imply that 0 D i D 0, for all
i D 1; : : : ; I. In turn, this implies that the equilibrium price functional given as in
the right hand side of (8.21) becomes identically equal to zero and, hence, ˛2i D 0,
for all i D 1; : : : ; I. But then the first relation in (8.22) would imply that ¤ 0, thus
yielding a contradiction. t
u
In particular, the fact that ¤ 0 in (8.21) implies that the equilibrium price
functional is only partially revealing:
P equilibrium prices are determined not only
by the private signals, via the term IiD1 i yi , but also by the random aggregate
supply. Relation (8.22) also shows that the coefficient i is inversely proportional to
agent i’s risk aversion and to the variance of the private signal observed by agent
i. Hence, in equilibrium, very risk averse agents and agents with very imprecise
information will affect the equilibrium price in a limited way.
In the context of Proposition 8.11, it is interesting to study the limit behavior of
the Green-Lucas equilibrium when u2 ! C1 (i.e., the noise due to the random
supply dominates): in this case (see Hellwig [931, Proposition 4.2]), variations in
prices reflect mostly variations in the supply of the risky asset rather than variations
in the agents’ private signals and, therefore, agents cannot extract any useful
information from prices. Hence, if u2 ! C1, then the equilibrium price functional
converges to the equilibrium price functional obtained in an economy where agents
condition their expected utilities only on their private signals. On the other hand,
if u2 ! 0, then the equilibrium price functional obtained in Proposition 8.11
converges to that obtained in Proposition 8.10, where the aggregate supply is known
ex-ante and the equilibrium price is a sufficient statistic for the agents’ private
signals (see Hellwig [931, Proposition 4.3]). In general, the presence of the noise
component represented by the random asset supply prevents the equilibrium price
from being a sufficient statistic. In Hellwig [931, Proposition 5.2] it is also shown
that, as the number of agents increases to infinity, the private information effect
vanishes and the equilibrium price only depends on the realization of dQ (positively)
and uQ (negatively). In particular, dQ represents the common element in the agents’
private signals (in the limit, the noise terms "Qi vanish by the law of large numbers).
Note that, even in this limit case, the equilibrium price functional is only partially
revealing because agents cannot distinguish whether a high price is due to a high
realization of dQ or to a low realization of uQ .
A model similar to that considered in Hellwig [931] has been analysed in
Diamond & Verrecchia [569], under the assumption that the (random) aggregate
supply is given by the sum of the randomP endowments of the I agents (in terms
I
of units of the risky asset), so that uQ D Q i , where eQ i denotes the random
iD1 e
endowment of agent i and is assumed to be normally distributed with zero mean
and variance e2 , for all i D 1; : : : ; I. The random endowments are assumed to be
mutually independent and independent of all the other random variables appearing
in the model. In addition to observing the private signal and the market price, as
considered above, every agent also observes his own endowment eQ i . Under the
additional assumptions that ai D a > 0 and "2i D 1, for all i D 1; : : : ; I, the same
arguments given in the proof of Proposition 8.11 allow to show that the Green-Lucas
equilibrium price functional P is linear with respect to the aggregate supply uQ and
the average private signal yQ WD IiD1 yQ i =I observed by the agents. More precisely,
as shown in Exercise 8.13, it holds that:
X i
I
.u; y/ D ˛ dN C y u D ˛ dN C y u; (8.23)
I iD1
with
e2 C 1 2 .e2 C I/
˛D ; D and D ;
1 C I 2 C e2 .1 C 2 / 1 C I 2 C e2 .1 C 2 / I
where we recall that dQ is assumed to be normally distributed with mean dN and

variance 2 .
The equilibrium price functional (8.23) is increasing with respect to the average
signal received by the agents and decreasing with respect to the aggregate supply.
In the presence of a random endowment and differently from the result of Proposi-
tion 8.10, the optimal demand of each agent depends on the private signal because
the presence of the random supply does not make the private signal irrelevant.
Hence, there exists an incentive to collect private information, thus solving the
paradox generated by fully revealing equilibria discussed at the end of Sect. 8.2.
Random Supply and Asymmetric Information
A general model with asymmetric information has been proposed in Vives [1630,
Chapter 4], encompassing several variations of the models originally proposed in
Diamond & Verrecchia [569] and Hellwig [931]. As above, we consider a two-
period economy (t 2 f0; 1g) with two traded securities: a riskless asset paying a
constant rate of return rf D 1 and a risky asset delivering at t D 1 the random
dividend d.Q We assume that the economy is populated by a continuum of risk
averse agents i indexed in the interval Œ0; 1 endowed with the Lebesgue measure.
Similarly to the models considered above, the preferences of each agent i 2 Œ0; 1 are
characterized by a negative exponential utility function of the form ui .x/ D ea x ,
i
with ai > 0. The initial wealth of the agents is supposed to be non-random and
normalized to zero (this is without loss of generality, due to the assumption of
CARA utility functions). Under this hypothesis, denoting by wi 2 R the demand
of the risky asset by agent i, the corresponding wealth at t D 1 is given by the
random variable W e i D wi .dQ p/, for i D 1; : : : ; I, where p denotes the price at date
t D 0 of the asset.
In the economy, there are informed and uninformed agents. A priori (i.e., at date
t D 1), all agents believe that the random dividend dQ is normally distributed
with mean dN and variance 2 . At the interim date t D 0, each informed trader i
observes a private signal yQ i of the form yQ i D dQ C "Qi . On the contrary, uninformed
traders do not have access to any private information and can only infer information
from market prices. The proportion of informed traders is denoted by 2 Œ0; 1,
with the proportion of uninformed traders being consequently 1 . We assume
that the two classes of agents have homogeneous risk aversion, in the sense that
ai D ainf > 0 for all i 2 Œ0; and ai D aun > 0 for all i 2 .; 1, where the
superscripts inf and un stand for informed and uninformed, respectively. The random
variables dQ and "Qi , are assumed to be normally distributed and mutually independent,
with EŒQ"i D 0 and Var.Q"i / D "2 for all i 2 Œ0; (i.e., all informed traders
observe signals of the same precision and private signals are uncorrelated across
agents). Besides informed and uninformed traders, there are also noise traders and
we assume that their aggregate demand is represented by the random variable uQ ,
which is normally distributed, uncorrelated with all the random variables introduced
above, with zero mean and variance u2 . Noise traders represent investors trading for
exogenous liquidity reasons. As above, we assume that the structure of the economy
as well as all the distributional assumptions are common knowledge among all the
agents.
In deriving a rational expectations equilibrium for this economy, we focus our
attention on symmetric equilibria, meaning that traders of the same class exhibit
optimal demands of the same form. Let us denote by winf . yi ; p/ the demand of
each informed trader i 2 Œ0; (which depends on both the private signal yi and
the price information p) and by wun . p/ the demand of each uninformed trader
i 2 .; 1 (depending on the price information only). In this context, a symmetric
rational expectations equilibrium is characterized by two conditions: both informed
and uninformed agents maximize their expected utilities conditionally on their
information sets and markets clear. In particular, the market clearing condition is
given by
Z Z 1
w . y ; p/di C
inf i
wun . p/di C u D 0; (8.24)
0
where u 2 R is a realization of the random variable uQ representing the aggregate

demand of noise traders. In this context, we can establish the following proposition
(compare with Vives [1630, Proposition 4.1] and see Exercise 8.14 for a proof),
where we say that a rational expectations equilibrium is linear if the associated
demand functions are given by linear functions.
Proposition 8.12
R In the context of the model described R above (see Vives [1630]),
suppose that 0 "Qi di D 0 almost surely, so that 1 0 yQ i di D dQ almost surely.
Then there exists a unique partially revealing rational expectations symmetric linear
equilibrium characterized by the price functional W R2 ! R given by

.d N Cu
d/
ainf "2
.d; u/ D C d; (8.25)
1
ainf "2
C binf C .1 /bun
where
1
2 ainf inf
binf D and bun D b :
1
inf C aun aun
ainf C a
"2 u2
Moreover, for p WD .d; u/, the optimal demands winf; and wun; of the informed
agents and of the uninformed agents are respectively given by
1 N
winf; . yi ; p/ D . yi p/ binf . p d/; for all i 2 Œ0; ; (8.26)
ainf "2
and
N
wun; . p/ D bun . p d/; for all i 2 .; 1; (8.27)
The rational expectations equilibrium characterized in Proposition 8.12 is not

fully revealing: similarly as in the models considered in the first part of the present
section, traders observing a high price cannot determine whether such a price is
motivated by a high dividend or by a large demand by the noise traders. In other
words, the presence of noise traders does not allow the equilibrium price to fully
reveal the private information of the informed agents. As shown in Exercise 8.15,
the equilibrium price obtained in Proposition 8.12 can be represented as a weighted
average of the conditional expectations of the future dividend of the two classes of
agents plus a noise term due to the presence of noise traders. If aun ! 0, then the
equilibrium price functional converges to the conditional expectation EŒdjp. Q
The result of Proposition 8.12 leads to an interesting interpretation of the
behavior of the two classes of agents. Since bun > 0, formula (8.27) shows
that uninformed agents trade against the market (i.e., they sell the asset when its
price is above the a priori expected dividend). Uninformed agents face an adverse
selection problem, because the equilibrium is not fully revealing and they do not
know whether trading is motivated by informed traders or by noise traders. The
behavior of informed traders is determined by two distinct components, as shown
by (8.26): informed agents tend to buy (sell, resp.) the risky asset if its price is
below (above, resp.) their private signal and tend to sell (buy, resp.) the asset if
its price is above (below, resp.) the a priori expected dividend. In particular, the
first component reflects a speculative behavior motivated by private information.
For the informed traders, the sensitivity to the private signal is measured by the
quantity 1=.ainf "2 /, which is inversely proportional to the coefficient of absolute risk
aversion and directly proportional to the precision of the private signal. Moreover,
the quantity .1=.ainf "2 / C binf / C .1 /bun provides a measure of market depth,
i.e., how much the market can absorb a change in the noise traders’ demand without
a significant price change. This measure of market depth is a weighted sum of the
parameters describing the traders’ responsiveness to the market price, as can be
deduced from Proposition 8.12.
Costly Private Information and the Grossman-Stiglitz Paradox
The impossibility of market efficiency when the acquisition of private information

is costly has been shown in Grossman & Stiglitz [849]. This paper is concerned
with a phenomenon already discussed in Sect. 8.2: if acquiring private information
is costly and market prices transmit private information, then it can be convenient
for some agents to adopt a free riding behavior and not buy private information,
inferring instead information from market prices, since market prices reflect (at least
partially) the information acquired by the other agents.
The model considered in Grossman & Stiglitz [849] is a two-period model
(t 2 f0; 1g), with an economy populated by a continuum of agents indexed in the
interval Œ0; 1 and two traded assets: a riskless security with return rf > 1 and a risky
asset delivering at date t D 1 the random dividend d. Q More specifically, the random
variable dQ is assumed to be of the form
dQ D dQ o C "Q;
where the pair of random variables .dQ o ; "/

Q is assumed to have a bivariate normal
distribution, with
EŒQ" D 0; Cov.dQ o ; "Q/ D 0 and Q dQ o D do / D Var.Q"/ DW "2 ;

Var.dj
for every do 2 R. Furthermore, the per capita supply of the risky asset is random and
given by the realization of the random variable uQ , which is independent of .dQ o ; "Q/
and normally distributed with mean uN and variance u2 . Let also 2 WD 2 .d/Q D
2 .dQ o / C "2 .
Letting wi0 and wi denote the initial wealth2 and the demand of the risky asset,
respectively, of agent i, the random wealth at date t D 1 is given by
e i1 D .wi0 wi p/rf C wi d;
W Q for all i 2 Œ0; 1;
where p denotes the price at date t D 0 of the risky asset. Similarly as in the models
presented above, we assume that the preferences of each agent i are characterized
by a negative exponential utility function of the form ui .x/ D eax , with a > 0,
for all i 2 Œ0; 1. In particular, the coefficient of absolute risk aversion is assumed to
be the same for all the agents.
At date t D 0, agents can choose to observe the realization of the random
variable dQ o by paying a fixed cost c > 0, while the residual component dQ dQ o D "Q
remains unobservable. As usual, we suppose that each agent maximizes his expected
2
For the sake of simplicity and without loss of generality, we assume that the initial wealth of each
agent is non-random. In the original paper Grossman & Stiglitz [849], the initial wealth is specified
in terms of the two traded securities and, hence, depends on the equilibrium price, which is in turn
random.
utility, exploiting all the available information and deciding whether to buy or not
the information related to the observation of dQ o . In other words, each agent has to
take two decisions: whether to buy or not the additional information and how much
to invest in the risky asset. We say that an agent is informed if he has chosen to
observe the component dQ o and uninformed otherwise (in the following, we shall
use the superscript inf when referring to informed agents and the superscript un
when referring to uninformed agents). Agents have to decide ex-ante whether to buy
the information or not. Apart from the information set used to take the investment
decisions, the two types of agents are identical and the structure of the economy as
well as all the distributional assumptions are common knowledge among the agents.
We denote by 2 Œ0; 1 the proportion of informed agents in the economy. In a
first step, we shall consider as an exogenous variable and then, in a second step,
determine the endogenous value of corresponding to an overall equilibrium of the
economy.
Considering as a fixed exogenous variable, a Green-Lucas equilibrium of the
economy, similarly as in Definition 8.6 (and compare also with Proposition 8.11),
is characterized by an equilibrium price functional .do ; u/ W R2 ! R in
correspondence of which the market clears. The equilibrium price functional
depends on the realization of the random variables dQ o and uQ (representing the
information available to the informed agents and the random per capita supply,
respectively). The subscript makes explicit the dependence of the equilibrium
price functional on the proportion of informed agents. In the following proposition,
which corresponds to Grossman & Stiglitz [849, Theorem 1], we derive the
optimal demands of the informed and uninformed agents and the equilibrium price
functional, for fixed.
Proposition 8.13 In the context of the model described above (see Grossman &
Stiglitz [849]), let the proportion 2 Œ0; 1 of informed agents be fixed. Then the
equilibrium price functional W R2 ! R is given by
.do ; u/ D ˛1 ./ C ˛2 ./ .do ; u/; for all d o 2 R and u 2 R; (8.28)
where
(
a"2
.do ; u/ WD do
.u uN /; if > 0;
0 .d ; u/ WD u;
o
if D 0;
and ˛1 ./ 2 R and ˛2 ./ > 0. Moreover, for every realization do and u, the optimal
demand of each informed agent in correspondence of the price p D .do ; u/ of the
risky asset is given by
do prf
winf; .do ; p/ D ; (8.29)
a"2
and the optimal demand of each uninformed agent is given by
Q .dQ o ; uQ / D p prf
EŒdj
wun; .p/ D : (8.30)
Q .dQ o ; uQ / D p/
a Var.dj
Proof Similarly as in the case of the models considered earlier in this section, we
conjecture that the equilibrium price functional is a linear function of the random
variables dQ o and uQ as in the right-hand side of (8.28).
Let us start by computing the optimal demand of an informed agent. By
assumption, each informed agent observes the realization of the random variable
dQ o . Since uQ and "Q are independent, an informed agent observing the realization of dQ o
does not receive any useful further information from market prices. Hence, it holds
that
o o
Q dQ D do ; .dQ o ; uQ / D p D E dj
E dj Q dQ D do D do ;
where the last equality follows since Cov.dQ o ; "Q/ D 0. Similarly, it holds that
o o
Q dQ D do ; .dQ o ; uQ / D p D Var dj
Var dj Q dQ D do D "2 :
As a consequence, by the assumption of an exponential utility function, the

optimal demand of each informed agent is given by (8.29), for every realization
do 2 R. Considering now the case of an uninformed agent, (8.30) simply follows
by computing the optimal demand of an agent maximizing an exponential utility
function (together with the assumption of a normal distribution, compare with
formula (3.11)), conditionally on the information conveyed by the observation of
the market price.
In order to explicitly characterize the equilibrium price functional, consider first
the case D 0 (i.e., there are no informed agents). In this case, the market clearing
condition reduces to the condition wun; .p/ D u, for every realization u 2 R.
Let us conjecture that the equilibrium price functional is more specifically given
by
EŒdQ o a 2
0 .do ; u/ D u; (8.31)
rf rf
for every .do ; u/ 2 R2 . Note that 0 .dQ o ; uQ / is independent from d,

Q since it does not
depend explicitly on dQ o (because there are no informed agents) and uQ is assumed to
be independent from d.Q Hence:
o
Q 0 .dQ ; uQ / D p D EŒdQ D EŒdQ o
E dj
Q .dQ o ; uQ / D p/ D 2 . This implies that the optimal demand of

and Var.dj 0
every uninformed agent in correspondence of the price p D 0 .do ; u/ is given
by
EŒdQ o prf
wun; . p/ D :
a 2
By construction, the equilibrium price functional defined in (8.31) satisfies the
market clearing condition wun; .0 .do ; u// D u and the linear representation (8.28)
of the equilibrium price functional is also satisfied. Moreover, the optimal demand
is clearly of the form (8.30).
Let us now consider the more interesting case where 2 .0; 1. Define
.1/EŒdQ j .d o ;u/

a Var.dQ j .d o ;u//
EŒQu a"2
˛1 ./ WD and ˛2 ./ WD ;
1 1
rf a"2
C a Var.dQ j .d o ;u//
rf a"2
C a Var.dQ j .d o ;u//
where we denote by EŒdj Q .do ; u/ the conditional expectation of dQ given

Q
the observation of f .d ; uQ / D .do ; u/g (and similarly for the conditional
o
variance), and let us conjecture that the equilibrium price functional is given
by
.do ; u/ D ˛1 ./ C ˛2 ./ .do ; u/; (8.32)
for every realization .do ; u/ 2 R2 of the random variables dQ o and uQ , where .do ; u/
is defined as in the statement of the proposition. By the properties of the normal
multivariate distribution, it holds that

Q .do ; u/ D EŒdQ o C 2 .dQ o /
E dj .do ; u/ EŒdQ o ;
Var. .dQ o ; uQ //
!

Q
2 Qo 2 .dQ o /
Var dj .d ; u/ D .d / 1
o
C "2 ; (8.33)
Var. .dQ o ; uQ //
2
a"2
Var .dQ o ; uQ / D 2 .dQ o / C u2 :

Moreover, since the map .do ; u/ 7! .do ; u/ is linear and ˛2 ./ ¤ 0, it holds
that
o o
E dj Q .do ; u/ and
Q .d ; u/ D E dj Q .d ; u/ D Var dj
Var dj Q .do ; u/ :
By using (8.29), it can be checked that the market clearing condition

u D winf; do ; .do ; u/ C .1 /wun; .do ; u/
do .do ; u/rf Q .do ; u/ .do ; u/rf
EŒdj
D C .1 /
a" 2 Q .do ; u//
a Var.dj
holds for every .do ; u/ 2 R2 , thus showing that definition (8.32) gives indeed the
equilibrium price functional, satisfying the linear representation (8.28).
t
u
The equilibrium price functional obtained in Proposition 8.13 is only partially
revealing. Indeed, uninformed agents cannot learn the realization of dQ o by observing
the equilibrium price because the equilibrium price is also affected by the random
aggregate supply of the risky asset. Hence, changes in the equilibrium price can be
due to changes in the informed agents’ information as well as to changes in the
supply of the risky asset.
The equilibrium price .dQ o ; uQ / is informationally equivalent to the random
variable .dQ o ; uQ /, which is in turn a mean-preserving spread of dQ (this simply
follows by the fact that EŒ .dQ o ; uQ /jdQ o D do D do , since the random variables
dQ o and uQ are assumed to be independent). In particular, the noise produced by
the random supply uQ prevents uninformed traders from perfectly learning the
realization of dQ o from the observation of .dQ o ; uQ /. More precisely, the quality of
the information conveyed by .dQ o ; uQ / to uninformed agents is measured by
a2 4
Var .dQ o ; uQ /jdQ o D do D 2" u2 :

We observe that, if the variance u2 of the supply is large, then the equilibrium price
is scarcely informative about the component dQ o of the random dividend of the risky
asset. Furthermore, the informativeness of the equilibrium price is decreasing with
respect to the coefficient of absolute risk aversion and, in line with intuition, is
increasing with respect to the proportion of informed traders in the economy. Note
also that, if D 0 (i.e., there are no informed agents), then the equilibrium price
does not convey any information on dQ o .
So far, the proportion 2 Œ0; 1 of informed agents in the economy has been
considered as an exogenous fixed variable. Indeed, the equilibrium price functional
obtained in Proposition 8.13 is explicitly characterized for a given 2 Œ0; 1.
However, the proportion of informed agents should be considered as an endogenous
variable in the model: indeed, agents can a priori choose whether to buy the
information (paying the cost c > 0) or not. Hence, we can define an overall
equilibrium as a couple . ; .dQ o ; uQ //, where is such that
• if D 0, then, in correspondence of the price 0 .dQ o ; uQ /, the optimal expected
utility of an informed agent is less or equal than the optimal expected utility of
an uninformed agent;
• if 2 .0; 1/, then, in correspondence of the price .dQ o ; uQ /, the optimal
expected utility of an informed agent is equal to the optimal expected utility of
an uninformed agent;
• if D 1, then, in correspondence of the price .dQ o ; uQ /, the optimal expected
utility of an informed agent is greater or equal than the optimal expected utility
of an uninformed agent.
If an agent i 2 Œ0; 1 chooses to buy the information, then his optimal wealth at t D 1
in correspondence of the equilibrium price .dQ o ; uQ / will be given by

e inf;i; WD wi0 c winf; .dQ o ; .dQ o ; uQ // .dQ o ; uQ / rf C winf; .dQ o ; .dQ o ; uQ //d;
W Q
1;
while, if agent i 2 f1; : : : ; Ig chooses not to buy the information,

e un;i; WD wi0 wun; . .dQ o ; uQ // .dQ o ; uQ / rf C wun; . .dQ o ; uQ //d;
W Q
1;
where winf; .dQ o ; .dQ o ; uQ // and wun; . .dQ o ; uQ // denote the optimal demands of
informed and uninformed agents, respectively, as considered in Proposition 8.13.
Observe that, due to the assumption of an exponential utility function, the optimal
demand of the risky asset does not depend on the initial wealth and, hence,
does not depend on the cost c. Since wi0 is supposed to be non-random and
since we consider exponential utility functions, it is easy to see that the ratio
EŒu.We inf;i; /=EŒu.W e un;i; / does not depend on i and, hence, we can define the
1; 1;
function W Œ0; 1 ! RC by

e inf;i; /
E u.W 1;
./ WD ;
e un;i; /
E u.W 1;
for 2 Œ0; 1. We can now establish the following proposition, which characterizes
the overall equilibrium . ; .dQ o ; uQ // where the proportion of informed traders is
endogenous (compare with Grossman & Stiglitz [849, Theorem 3]).
Stiglitz [849]), the function 7! ./ is strictly increasing. For D 0 it is given by
r
1
.0/ D eacrf
1Cn
and, for > 0, it admits the representation

s r
Var.djQ dQ o D do / 1Cm
./ D e acrf
D e acrf ; (8.34)
Q .dQ o ; uQ / D .do ; u//
Var.dj 1 C m C nm
where
2
a"2 u2 2 .dQ o /
m WD and n WD :
2 .dQ o / "2
Moreover, the following hold:

(i) if, for some 2 Œ0; 1, it holds that . / D 1, then the couple
. ; .dQ o ; uQ // is an overall equilibrium,
(ii) if .1/ < 1, then the couple .1; 1 .dQ o ; uQ // is an overall equilibrium,
(iii) if .0/ > 1, then the couple .0; 0 .dQ o ; uQ // is an overall equilibrium,
where .dQ o ; uQ / is defined as in Proposition 8.13, for every 2 Œ0; 1. Moreover,
for every equilibrium price functional which is a monotone function of (as
defined in Proposition 8.13), there exists a unique overall equilibrium . ; /.
Proof The fact that the function 7! ./ is strictly increasing and can be
represented as in (8.34) is proved in Exercise 8.16. The second part of the
proposition easily follows from the fact that 7! ./ is strictly increasing together
with the assumption of negative exponential utility functions and the definition of
overall equilibrium. t
u
The quantity n introduced in Proposition 8.14 can be regarded as a measure of
the quality of the information accessible by informed agents, while the quantity 1=m
represents a measure of the quality of the information conveyed by the equilibrium
price. Indeed, if 2 .0; 1/ (so that . / D 1 in correspondence of the overall
equilibrium), then the correlation between the equilibrium price and the signal
dQ o (and, therefore, the degree of informativeness of equilibrium prices) can be
expressed as
s
ˇ ˇ 1 e2acrf 1
% WD ˇCorr.dQ o ; .dQ o ; uQ //ˇ D p D 1 : (8.35)
1Cm n
The following proposition presents several comparative statics for the equi-
librium characterized in Proposition 8.14 (see Grossman & Stiglitz [849, Sec-
tion II.H]).
Stiglitz [849]), the overall equilibrium . ; .dQ o ; uQ // satisfies the following
properties:
(i) if 2 .0; 1/, then % is increasing with respect to n and decreasing with
respect to c and a;
(ii) if 2 .0; 1/, then % is constant with respect to changes in u2 or with respect
to changes in 2 if n is constant;
(iii) if 2 .0; 1/, then is increasing with respect to u2 , decreasing with respect
to c and increasing with respect to 2 if n is constant;
(iv) if 2 .0; 1/ and 2 is constant, then there exists a value nN such that is
decreasing with respect to n for n > nN and increasing with respect to n for
n < nN ;
(v) there exists cN such that D 1 for every c cN ;
(vi) if "2 ! 0 and 2 is constant (so that 2 .dQ o / increases to 2 ), then the
equilibrium price functional becomes fully revealing and converges to
zero;
(vii) if u2 ! 0, then converges to zero and the informational content of the
equilibrium price is unchanged;
p
(viii) if u2 D 0 and eacrf < 1 C n, then an overall equilibrium does not exist;
(ix) if "2 D 0 (i.e., the signal dQ o is fully informative about the random variable
Q then an overall equilibrium does not exist;
d),
(x) the mean and the variance of the trading volume are zero for c sufficiently
large or small and the mean and the variance of the trading volume converge
to zero for n ! C1.
Proof Parts (i)–(ii) are direct consequences of (8.35), while part (iii) easily follows
from the explicit expression of ./ given in (8.34) together with the fact that the
map 7! ./ is strictly increasing. Part (iv) is proved in Exercise 8.17. Part (v)
simply follows from the observation that there exists a value cN satisfying
s
Var.djQ dQ o D do /
e aNcrf
D 1:
Q 1 .dQ o ; uQ / D 1 .do ; u//
Var.dj
Concerning part (vi), the fact that the equilibrium price functional becomes fully
revealing can be deduced from (8.35), since if "2 ! 0 and 2 .dQ o / ! 2 , then
the quantity n introduced in Proposition 8.14 explodes, so that % ! 1 (fully
informative price system). Moreover, representation (8.34) implies that, as "2 ! 0,
the quantities m and nm converge to zero for any fixed > 0, thus implying that
./ ! eacrf > 1, for any fixed > 0. Therefore, for the equilibrium to be
maintained, must also converge to zero. Parts (vii), (viii) and (ix) are proved in
Exercise 8.18. Finally, we refer the reader to Grossman & Stiglitz [849, Theorem 6]
for the proof of part (x). t
u
Proposition 8.15 shows several interesting properties of the overall equilibrium
characterized in Proposition 8.14. In particular, the informativeness of the equilib-
rium price (as measured by the correlation coefficient % introduced in (8.35)) is high
if the quality of information is high (i.e., if n is large), if the coefficient of absolute
risk aversion is small and if the cost of purchasing the information is small (see part
(i) of Proposition 8.15). Moreover, (8.35) shows that any change in the parameters of
the model that does not affect a, c or n leaves unchanged the informational content
of the equilibrium price.
If the variance u2 of the random supply increases, then, for any fixed , the
informational content of the equilibrium price decreases. However, in view of
Proposition 8.14, an increase in u2 also increases the proportion of informed traders,
since it increases the benefits of acquiring information. In equilibrium, as shown

in part (ii) of Proposition 8.15, these two effects offset exactly and, therefore, the
informational content of the equilibrium price is not affected by changes in u2 .
As shown in part (iii) of Proposition 8.15, an increase in the variance u2 of the
random supply, an increase in the variance 2 of the dividend (provided that n is
constant) or a decrease in the cost of information will increase the proportion of
informed agents. Moreover, part (iv) shows that the proportion of informed agents
is increasing with respect to the quality of the information (as measured by n) if the
latter is sufficiently poor, otherwise it will be decreasing.
This last result leads to a paradox (Grossman-Stiglitz paradox): as the economy
converges to the ideal conditions of absence of noise (u2 D 0) or of perfect
information ("2 D 0), the proportion of informed agents converges to zero (see parts
(vi)–(vii) of Proposition 8.15), since the equilibrium price will perfectly transmit
all the private information to the uninformed agents. As explained in Grossman &
Stiglitz [849], “there is a fundamental conflict between the efficiency with which
markets spread information and the incentives to acquire information”. This is to
say that there exists a positive information externality in the activity of information
acquisition by informed agents. As the economy tends to ideal conditions, the
equilibrium price becomes fully revealing and provides a very accurate estimate
of the dividend. In turn, this leads agents to behave as free riders: since the
cost of acquiring the information is strictly positive, agents prefer not to buy the
information, learning it from the price observation. But if there is no incentive to
buy information, then the only possible equilibrium is the one with no informed
agents. On the other hand, if everyone is uninformed and c is small enough, then
it is clearly optimal for some agent to acquire information, hence the paradox
of the inexistence of an overall equilibrium (compare also with the discussion in
Grossman & Stiglitz [849, Section IV]). This is the content of parts (viii)–(ix)
of Proposition 8.15, showing that an overall equilibrium with costly information
acquisition does not exist if markets are informationally efficient. The behavior of
the trading volume (part (x) of Proposition 8.15) confirms this result. Note that
the key feature of the Grossman & Stiglitz [849] model is that the proportion of
informed agents is considered as an endogenous variable (see also Sect. 8.7 for a
further discussion of the features of the model of Grossman & Stiglitz [849]).
An economy with costly private information has been analysed in Verrecchia
[1618] by relying on the model proposed in Hellwig [931]. Similarly as above, the
dividend dQ of the risky asset is assumed to be distributed as a normal random variable
with mean dN and variance 2 . The preferences of each agent i are characterized
by a constant coefficient of absolute risk aversion ai > 0 (CARA preferences).
The private information is represented by the observation of a signal of the form
yQ i D dQ C "Qi , where "Qi is a normal random variable with zero mean and variance
2 .Q"i / D 1=si , with si > 0, for all i D 1; : : : ; I. The supply of the risky asset is
distributed as a normal random variable with zero mean and all the random variables
introduced in the model are assumed to be mutually independent. The quantity si
can be regarded as a measure of the precision of the information of agent i, for
each i D 1; : : : ; I. The distinguishing feature of the model of Verrecchia [1618] is
that the quality of the information is endogenously determined and agents choose
the precision of their signal before trading. More specifically, the cost of observing
a signal with precision s is determined by a continuous cost function s 7! c.s/
satisfying c.s/ > 0, c0 .s/ > 0 and c00 .s/ 0, for every s 2 .0; 1/ (i.e., c is a
strictly increasing convex function). Similarly as in the models considered earlier in
this section, the fact that the aggregate supply is random yields an equilibrium price
functional which is not fully revealing. As above, CARA preferences together with
the assumption of multivariate normality represent crucial assumptions in order to
derive closed-form solutions. It is shown that there exists a Green-Lucas equilibrium
with an endogenous level of precision of information. In equilibrium, the level of
precision of the information acquired by an agent is a non-decreasing function of
his risk tolerance and a non-increasing function of the informational content of the
equilibrium price (which is in turn decreasing with respect to the variance of the
supply and the cost of acquiring information).
Diamond [568] considers a model similar to Diamond & Verrecchia [569] and
Hellwig [931], with infinitely many traders and two traded assets and where the
traders have the possibility of acquiring information about a firm’s return. In equilib-
rium (see Diamond [568, Lemma 3]), it is shown that all the agents can decide not to
buy information (in the case of a high cost and a low precision of the signal) or all the
agents buy information (in the case of a small cost and a very precise information).
In all the intermediate cases, only a fraction of the agents will acquire information.
In a model similar to Diamond & Verrecchia [569] and Grossman & Stiglitz [849],
two equilibria with strategic substitution/complementarity effects may emerge when
the aggregate endowment and the individual endowments are positively correlated
(see also Ganguli & Yang [754]). These two equilibria have opposing properties
concerning the informational content of equilibrium prices. In the context of a model
with a convex cost of precision of information and where the valuation of each agent
has both a private and a common component, Vives [1631] shows that there exists
a fully revealing equilibrium and an incentive to acquire information, provided that
the common value component does not dominate. Furthermore, the introduction of
derivative assets provides an incentive to acquire costly information and is shown to
positively affect the price efficiency (see Cao [361]).
Markets for Information
In all the models presented so far, it has been always assumed that information
is a good with features and a cost defined ex-ante. In Admati & Pfleiderer [18],
these hypotheses have been relaxed by assuming the existence of a market for
information, with a monopolistic agency selling information about the random
dividend of the risky asset (information is represented by a noisy signal of the
dividend). The agency can decide the features of the information sold to the agents,
by determining the level of noise added to the signal, by deciding the price of the
information, by selling “personalized” signals to different agents and by deciding
how many agents can become informed. Similarly as in the models considered
above, there is a noise component (random aggregate supply) in the economy that
makes equilibrium prices only partially revealing. As above, agents are assumed
to have exponential utility functions and the random variables appearing in the
model are jointly normally distributed. Note that, as discussed in the case of the
Grossman & Stiglitz [849] model, although prices are partially revealing, agents can
behave as free riders by choosing not to buy any information, learning it from the
observation of market prices. This phenomenon is particularly severe when prices
reveal information precisely (i.e, the noise due to the random supply is small) and
when the information is accurate.
If the agency sells the same information to every agent and the information is
precise, then the agency will decide to add a noise component to the information
sold. In the case where very precise information is sold, then it is used very
aggressively by informed agents. The agency can contrast this phenomenon by
deciding to sell imprecise information. When the equilibrium price is a relatively
precise signal of the information sold (so that the externality effect in the acquisition
of information is important), the agency does sell information only to a fraction of
the agents and the information sold will be made imprecise. On the contrary, if the
noise component of the economy is large (so that equilibrium prices will transmit
less information) and the signal is by itself imprecise, then the agency will sell
information to a large portion of the market without adding further noise.
As a further step, Admati & Pfleiderer [18] analyse the case where the agency can
sell personalized information (i.e., signals with an idiosyncratic noise component)
to different agents. Selling personalized information in a large market increases the
informational content at the aggregate level and, therefore, encourages a free riding
behavior. On the other hand, personalization can make information more attractive
to each agent. If the agency chooses to sell identical information to every agent,
then informed traders do not learn any additional information from prices. On the
contrary, when agents can buy personalized signals, all agents will infer information
from the equilibrium price. The agency, which is assumed to maximize profits, aims
at maximizing the amount of information sold to the informed agents and, at the
same time, at minimizing the amount of information embedded in the price, in order
to reduce the free riding behavior. To compensate these two effects, it is optimal for
the agency to add noise to the signals in such a way that its realization does not affect
the equilibrium price. When the economy is large, selling personalized information
is generally better than selling the same photocopied signal and the agency will
choose to sell identically distributed signals to the entire market. In particular, in
order to reduce the possibility of a free riding behavior by uninformed agents, it is
better to add personalized noise components and not to restrict the number of agents
who can acquire information. Moreover, if the variance of the noise component
in the economy is sufficiently large or if the agency’s own information is very
imprecise, then it is optimal for the agency not to add any noise component and
to sell the information to the entire market.
8.4 Information in Dynamic Market Models 445
8.4 Information in Dynamic Market Models
In this chapter, we have so far discussed the role of heterogeneous/asymmetric

information in the context of economies with a single trading period. In this section,
we give an overview of the possible extensions to economies where agents are
allowed to trade over several periods and receive private information over time.
At each trading date, (informed) agents receive a private signal and can also infer
information from the observation of market prices. Note that, at each trading date,
agents can infer information not only from private signals and market prices at
the current date, but also from past signals and past prices. This is also related
to technical analysis, in which the past price information is exploited to construct
trading strategies.
The main difficulty of multi-period models with heterogeneous/asymmetric
information consists in the definition and the computation of the rational expecta-
tions equilibrium. Indeed, in a rational expectations equilibrium agents will forecast
the forecasts of the others: market participants try to infer information about the
beliefs of the other market participants, who in turn try to infer information about
the beliefs of the other market participants, with the latter doing the same thing
and so on (higher-order beliefs). When the private signals received over time are
serially correlated, this phenomenon leads to an infinite regression problem, as
shown in the seminal paper Townsend [1598], since agents try to infer information
over time from endogenous variables and, hence, the dimensionality of the state
variables increases over time. Models of this type are complex and in some cases
can only be solved numerically (see, e.g., Hussman [994], Singleton [1548], Zhou
[1681], Makarov & Rytchkov [1289]). Typical ways to avoid the infinite regression
problem are represented by the assumption that agents are hierarchically informed
(i.e., the information of a group of agents includes the information of the other
agents).
The no-trade result of Proposition 8.4 has been extended to a multi-period
economy. In particular, considering a finite number of risk neutral agents and an
asset paying a dividend at each trading date, it has been shown in Tirole [1594]
that, if agents exhibit common prior distributions and the initial allocation is ex-
ante Pareto optimal, then there are no speculative bubbles in a rational expectations
equilibrium (regardless of whether short sales are allowed or not). As a consequence,
there is no speculative behavior in correspondence of a rational expectations
equilibrium, in the sense that no agent is willing to pay for an asset more than what
he would pay if he was obliged to hold the asset forever (this notion of speculative
behavior is due to Harrison & Kreps [903]). As stated in Tirole [1594], “speculation
relies on inconsistent plans and is ruled out by rational expectations”. This result
is intimately connected to the impossibility of speculative bubbles in a dynamic
rational expectations equilibrium (compare with Sect. 6.5). Keeping the assumption
of a finite number of risk neutral agents with common priors (and considering an
asset with no fundamental value, a feature assumed to be of common knowledge
in the economy), Bhattacharya & Lipman [226] show the possibility of speculative
bubbles if each agent’s initial endowment is private information. In this context,

assuming that the initial allocation is interim Pareto optimal (i.e., conditionally on
the private information of each agent), a bubble can exist in equilibrium.
In a finite horizon economy with risk averse agents and a single risky asset, Allen
et al. [49] distinguish between expected and strong bubbles: a strong bubble occurs
if there exists a state of the world in correspondence of which every agent knows
that the asset price is greater than the expected value of the future dividends, while
an expected bubble occurs if there exists a state of the world in correspondence of
which the asset price is greater than every agent’s marginal valuation of the asset.
Allen et al. [49] show that necessary conditions for a strong bubble to occur are
that: (i) each agent has private information in the date and state in which the bubble
occurs; (ii) each agent is short sale constrained at some date in the future with
positive probability; (iii) agents’ trades are not common knowledge. In particular,
as shown by an example in Allen et al. [49], the fact that agents’ beliefs are not
common knowledge plays a key role: even if every agent realizes that the asset is
overpriced (and it is not common knowledge that everybody believes that its price
will fall) but assigns a positive probability to the event of reselling the asset at
a greater price to some other agent at some future date/state, then a bubble may
arise and the no-trade result does not hold (provided that the short sale constraint
is binding). Every agent knows that the bubble will eventually burst, but there is
uncertainty on when this will happen.
In the models surveyed so far, all informed agents are assumed to receive their
private information simultaneously. Hirshleifer et al. [947] consider an economy
where some agents can observe a private signal about the fundamental value
of a risky asset earlier than other agents, showing that the timing at which
investors receive information may be even more important than the accuracy of
the information itself. The equilibrium where some agents receive information
before other agents can be fundamentally different from the equilibrium where
all the agents receive information simultaneously. Referring to Exercise 8.19 for
a detailed presentation of the model and for the derivation of the equilibrium price
functional, Hirshleifer et al. [947] consider an economy with two successive trading
periods, populated by a continuum of risk averse agents, where early informed
agents can observe a random component of the terminal value of a risky asset at
date t D 1, while late informed agents can observe this component only at the
successive date t D 2. The economy is also populated by liquidity traders (noise
traders) and by risk neutral competitive market makers (who do not have any private
information and set prices in a competitive way ensuring market clearing). As shown
in Exercise 8.19, the equilibrium price functional is linear, the optimal demand of
the early informed traders at the first trading date (t D 1) is determined by two
components (see equation (8.48)): a first component exploiting the expected price
appreciation between t D 1 and t D 2 and a second component to lock in at the
current price the expected optimal demand at the successive date t D 2. The early
informed agents reverse a fraction of their previous position at date t D 2, when the
trades of the late informed agents cause the price to more fully reflect the available
information. For the late informed agents, it is optimal not to trade during the first
trading period. As shown in part (ii) of Exercise 8.19, the price changes at t D 1
and t D 2 are both positively correlated with the private signal. Moreover, as shown
in Hirshleifer et al. [947, Proposition 2], the late informed agents appear to “follow
the leader”, in the sense that their trades are positively correlated with those of the
early informed agents. The model also allows for an explanation of the “herding”
behavior among investors.
As discussed at the beginning of Sect. 8.3, the introduction of a random supply
(noise) in the economy is one of the most typical ways to prevent the problems
associated with a fully revealing rational expectations equilibrium. In the presence
of noise, the agents’ private information does not become redundant after observing
market prices. In Hellwig [932], considering a multi-period economy where one
riskless asset and one risky asset are traded and agents formulate their optimal
demands according to a mean-variance criterion, it is shown that, if all agents
condition their expectations on past market prices and the time span between two
successive trading dates converges to zero, then the equilibrium price converges
towards the fully revealing rational expectations equilibrium price but the returns
of being informed are still bounded away from zero. In this sense, the market will
approximate full informational efficiency arbitrarily closely. The model of Hellwig
[931] has been extended to a continuous time setting in Naik [1366], assuming
a continuum of infinitely lived agents with CARA preferences (with possibly
heterogeneous risk aversion). If the asset supply is random and agents observe noisy
private signals of different precision on the growth rate of the dividend process
(which includes a random unobservable component), then the stationary rational
expectations equilibrium price reflects the information of the common part of the
private signals of a large set of agents. However, the agents’ private information
does not become redundant and the equilibrium is only partially revealing. In Naik
[1367], the above model is analysed under the assumption of the presence of a
monopolistic seller of information endowed with perfect information. It is shown
that the optimal strategy for the information seller consists in selling information
only to a fraction of the market, without adding noise components to the information
sold.
The transmission of information in a continuous time economy with informed
and uninformed traders is analysed in Wang [1643] (see also Biais et al. [233] for
an analysis of the effects of asymmetric information in an overlapping generations
model with multiple assets, in line with Admati [15]). In the model of Wang
[1643], informed agents possess information on the future dividend growth rate.
The information flow is hierarchical (thus avoiding the infinite regression problem),
in the sense that informed agents know all the information known by uninformed
agents. Since the growth rate of the dividend process determines the rate of
appreciation of the stock price, changes in prices provide a signal on the future
dividend growth rate and uninformed agents rationally extract this information from
the observation of market prices. The aggregate supply of the risky asset is assumed
to be random and, hence, equilibrium prices do not fully reflect the agents’ private
information. Wang [1643] shows that the presence of uninformed agents increases
the asset risk premium, can increase price volatility and induces negative serial
correlation in returns (due to the mean reversion in the underlying state variables
affecting the expected excess returns). In particular, the risk premium is increasing
in the number of uninformed agents because of the adverse selection problem (i.e.,
uninformed agents demand an additional premium for the risk of trading against
better informed agents) and because the price is less informative and there is more
uncertainty in the stock’s future cash flows. The effect of the presence of uninformed
agents on price volatility depends on two effects: on the one hand, uninformed
agents are characterized by a less volatile expected cash flow, thus reducing price
volatility; on the other hand, future cash flows are more uncertain, so that investors
demand a greater risk premium for noise trading and prices become more sensitive
to supply shock, thus increasing price volatility. If noise trading is significant, then
the price volatility is increasing with respect to the number of uninformed traders.
It is interesting to remark that the presence of informed/uninformed agents can
also lead to an explanation of the equity premium puzzle (see Sect. 7.3). Indeed,
Zhou [1682] shows that a high risk premium together with a low risk free rate
and a plausible level of risk aversion can be obtained in a model with asymmetric
information under the assumption that uninformed agents cannot buy the market
portfolio. In this model, the adverse selection effect plays a crucial role.
Information, Trading Volume and Price Volatility
In the context of multi-period economies, the presence of private/asymmetric

information has also interesting implications on the trading volume and on the
volatility of the equilibrium price process. In the context of a multi-period economy
with private information and random aggregate supply, He & Wang [920] show
that, in correspondence of the (partially revealing) equilibrium, trading volume is
significantly correlated with the inflow of information (both private and public) in
the economy. In the model considered in He & Wang [920], the information flow
is not hierarchical and the economy is populated by infinitely many investors. At
each trading date, investors receive both private and public information concerning
the fundamental value of a stock. When the information is private and dispersed,
agents trade in the market even after the arrival of information and, therefore, trading
volume can be serially correlated and may lag behind the information inflow, being
also related to the private information previously received. When the information is
public, agents trade only simultaneously with the arrival of information. While the
information inflow generates trading volume and price changes, trading generated
by existing information is not accompanied by price changes. Note that movements
in asset prices may not necessarily be caused by external news, but only by the
revelation of information by the trading process itself, if the equilibrium aggregates
the private information in an imperfect way (see Romer [1463] and Grundy & Kim
[852]).
The relation between trading volume and asymmetric information has been
also explored by Watanabe [1646] in the context of an overlapping generations
model with multiple assets, heterogeneous information and random supply. The
model generates multiple equilibria that can exhibit strong or weak correlations
between asset returns. Trading volume has a hump-shaped relation with respect to
the average level of information accuracy and is positively correlated with absolute
price changes. Less informed agents trade in the same direction of price changes
and behave like trend followers, while better informed agents trade by adopting
contrarian strategies.
Campbell et al. [346] propose a model with asymmetric information to analyse
the relationship between trading volume and the serial correlation of asset returns
(see Gallant et al. [751] and Sect. 8.6 for some empirical evidence). A detailed
presentation of this model together with the derivation of the equilibrium price is
provided in Exercise 8.20. In the economy, there are two traded assets (a riskless
asset and a risky asset) and two classes of agents: risk averse “market makers”
(utility maximizers) and “liquidity” traders (with changing risk aversion), who trade
for exogenous reasons. In the model there is no private information, in the sense that
all the agents are assumed to observe at each trading date a common noisy signal
about the future dividend shock. Price changes are determined by three sources:
innovations in the current dividend, innovations in the information about future
dividends and innovations in the time-varying risk aversion of the liquidity traders.
By construction, in equilibrium the trading volume is positive only due to changes
in the risk aversion and the serial correlation of asset returns is negatively affected
by the trading volume. The rationale of this result can be explained as follows:
if there is a price change and the trading volume is large, then the price change
is due to preference shocks (uninformative trades), while, if the trading volume is
small, then the price change is due to the inflow of new public information. As a
consequence, price changes accompanied by high volume will tend to be reversed
(thereby generating negative serial correlation in asset returns).
The relationship between trading volume and asymmetric information has been
further analysed in Wang [1644], considering an economy with uninformed and
informed agents and showing that a different nature of investors’ heterogeneity
implies different behaviors for the trading volume. While all the agents are assumed
to have CARA preferences with common risk aversion parameter and discount
factor, the two classes of agents differ in their information sets as well as in their
investment opportunities. Informed agents have access to private information about
the stock’s future dividends and trade both for informational reasons (speculation)
and liquidity reasons. Moreover, informed agents have access to a private investment
opportunity, which is not accessible to uninformed agents. Uninformed agents,
besides the realized dividends and market prices, observe a noisy public signal
of the private information received by informed agents (note that this hierarchical
information flow avoids the infinite regression problem). Informed agents trade
when they receive private information (informational trading) as well as when their
private investment opportunity changes (non-informational trading). Uninformed
agents only trade for non-informational reasons and rationally extract information
from the publicly available information. The aggregate supply of the risky asset is
constant and the structure of the economy is supposed to be common knowledge. In
this setting, the no-trade result does not hold, since uninformed agents are willing to
trade with informed agents since the latter also trade for non-informational reasons.
However, similarly as in other models with asymmetric information, uninformed
agents face an adverse selection problem, because they do not know whether
informed agents are trading for informational reasons or not. Due to this adverse
selection problem, trading volume is decreasing with respect to the asymmetry of
information between informed and uninformed agents. However, the correlation
between trading volume and the inflow of public information, excess returns
and absolute price-dividend changes is increasing with respect to the information
asymmetry. Since agents are risk averse, trading is always accompanied by price
changes, the trading volume is positively correlated with absolute price changes and
public information on the stock’s future dividends generates abnormal trading. The
implications of the current return and trading volume on future returns depends on
two effects: a high return accompanied by high trading volume implies high future
returns if the trading activity of informed agents is mostly due to informational
reasons, while it implies low future returns if the trading activity of informed agents
is mostly due to non-informational reasons (similarly as in Campbell et al. [346]).
A momentum effect associated with informative trading is due to the fact that prices
do not fully reveal private information. Similar implications are obtained and tested
with positive evidence in Llorente et al. [1232].
Note that, in almost all the models presented above, the trading volume is
related to price changes as well as to the quality of information. In models with
symmetric-private information, trading volume is increasing with respect to the
investors’ heterogeneity, while in models with asymmetric information (i.e., models
where there are informed and uninformed traders) high volume results when the
information asymmetry is limited (i.e., when the adverse selection effect is limited).
In particular, as considered in Bhattacharya & Spiegel [227], if the adverse selection
problem is particularly severe (more precisely, when the variance of the signal which
is observable to informed agents and unobservable to uninformed agents exceeds a
critical value) and there is a single monopolistic informed investor, then uninformed
agents refrain from trading, thus leading to a market breakdown (see Bhattacharya
& Spiegel [227, Proposition 3] and compare also with Chap. 10). This result has
been extended in Spiegel & Subrahmanyam [1559], showing that, if uninformed
agents do not know the variance of the informational variable, then the equilibrium
price functional is non-linear and a market breakdown occurs if the upper bound
of the support of the variance exceeds the variance of the liquidity component
of the demand. In this setting, public disclosure by firms can bound the variance
uncertainty and, therefore, mitigate the market breakdown problem.
8.5 Difference of Opinions 451
8.5 Difference of Opinions
As explained in the introduction to the present chapter, agents are said to have
heterogeneous opinions if they exhibit different probability distributions which
do not result from conditioning on information related to the fundamentals of
the economy. The analysis of economies populated by agents with heterogeneous
opinions is qualitatively different from the case of heterogeneous beliefs discussed
so far. Indeed, unlike heterogeneous beliefs, heterogeneous opinions are not related
to the observation of informative private signals. The most important consequence
of this difference is that, in the case of heterogeneous opinions, agents do not assign
any informational value to market prices (i.e., there is no private information to
aggregate and to infer from the observation of market prices). In this section, we give
an overview of some of the main results in the case of economies with heterogeneous
opinions.
No-Trade Results under Heterogeneous Opinions
The seminal no-trade result of Tirole [1594] that trading motivated by speculation
cannot occur in a rational expectations equilibrium does no longer hold if agents
have heterogeneous prior distributions (opinions). Indeed, in this case agents can
agree to disagree, since the fact that they exhibit different probability distributions
is unrelated to the fundamentals of the economy and does not embed private
information related to future dividends/payoffs. In comparison with the case of
heterogeneous beliefs, differences in opinions are not due to private information and
this prevents the no-trade result, since the willingness to trade is not interpreted by
other agents as the evidence of superior private information. Therefore, trade may
occur in equilibrium and the difference in opinions simply means that the agents
interpret in different ways the available information.
The fact that differences in opinions can lead to trade has been verified in
Varian [1611], Shalen [1522], Biais & Bossaerts [232]: if agents have different
opinions (i.e., different prior distributions about the asset’s fundamental value or
about the relationship between the realization of a signal and the value of the asset),
then the no-trade result does not hold and agents are willing to trade. Moreover,
trading volume and volatility are positively related to the dispersion of opinions
and trading volume and price changes are correlated. Morris [1352] deals with the
problem of identifying the different varieties of heterogeneous prior distributions
(opinions) that can lead to trade, pointing out that differences in opinions leading
to different interpretations of new information may represent the most important
source of trading in response to new information. Considering an economy with a
finite number of risk averse agents each observing a private signal and exhibiting
heterogeneous prior distributions, Morris [1352] shows that the no-trade result may
also hold in the presence of heterogeneous opinions, under suitable conditions:
differences in opinions of observing one’s own signal will not lead to trade; if
agents are allowed to make trades contingent on some event before the arrival of
information, then differences in opinions about that event will not lead to trade;
differences in opinions where agents undervalue their own signal will not lead to
trade (assuming that trades are incentive compatible); if trade is to be common
knowledge, then differences in beliefs about events which are not publicly revealed
will not lead to trade.
In Harris & Raviv [901] it is shown that the no-trade result does not hold when
agents receive public information but interpret information according to different
models (there is no private information but only different opinions). In this case, a
public announcement induces trading volume also in an economy without private
information and difference of opinions among agents leads to a positive correlation
between absolute price changes and trading volume as well as positive serial
correlation in the trading volume. Moreover, trading volume is positively related
to the heterogeneity of opinions (see also Kondor [1118], Banerjee & Kremer
[130], Cao & Ou-Yang [362]). In Kandel & Pearson [1066], it is shown that trading
volume can be positive in the presence of a public announcement even in the absence
of a price change, as observed empirically.
Asset Pricing Under Heterogeneous Opinions
In Miller [1342], Harrison & Kreps [903], Morris [1353] it is shown that, in the pres-
ence of heterogeneous opinions, short sale constraints can generate overpricing. A
simple example from Miller [1342] provides a clear illustration of this phenomenon
(we follow the presentation of Scheinkman & Xiong [1503, Section 2]). Let us
consider a two-period economy (i.e., t 2 f0; 1g), with a riskless asset paying the
constant rate of return rf D 1 and one risky asset. The random dividend dQ of the
risky asset is given by
dQ D dN C "Q; (8.36)
where "Q is a normal random variable with zero mean and variance "2 . Consumption
occurs only at date t D 1. The economy is populated by a continuum of agents and
Q More specifically,
each agent exhibits a personal opinion about the distribution of d.
instead of the true distribution (8.36), each agent believes that dQ is distributed
according to a normal law with mean dN i and variance "2 . In other words, all the
agents agree on the variance of the random dividend, but have different opinions on
the expected dividend. Since each agent is assumed to be characterized by his view
on the expected dividend, we index agents by dN i and assume that dN i is uniformly
distributed on the interval ŒdN k; dN C k, where k can be regarded as a measure of
the heterogeneity of opinions in the economy. Note that the average opinion in the
N corresponding to the true expected dividend. The preference structure

market is d,
of each agent dN i is characterized by a negative exponential utility function, with a
common risk aversion parameter a, and agent dN i is endowed with initial wealth wi0 .
The aggregate supply of the risky asset is supposed to be constant and given by u.
In this context, we can establish the following result.
Proposition 8.16 In the context of the model described above, if agents are not
allowed to short sell the risky asset, then its equilibrium price p at date t D 0 is
given by
(
dN a"2 u; if k < a"2 uI
p D p (8.37)
dN C k 2 ka"2 u; otherwise:
Proof Let us denote by wi the demand of the risky asset by agent dN i in terms of
units of the asset. The corresponding random wealth at t D 1, in correspondence of
e i1 D wi0 C wi .dQ p/. Given the short
a price p for the risky asset, is then given by W
sale constraint and since each agent has a negative exponential utility function, the
optimal demand wi is given by
N
di p
wi D max I 0 ; for all dN i 2 ŒdN k; dN C k:
a"2
Given the above optimal demand, the market clearing condition requires that
Z dN Ck
1 dN i p N
ddi D u:
2k max.p;dN k/ a"2
The latter condition directly implies that the equilibrium price p is given by (8.37).
t
u
As can be seen by examining the proof of Proposition 8.16, the agents believing
that the expected dividend is less than the current market price do not take part
to the trade, due to the prohibition of short sales. In the absence of short sale
constraints, the equilibrium price ppwould simply be given by dN a"2 u. It is easy
to see that dN a"2 u dN C k 2 ka"2 u. Formula (8.37) shows that, when the
difference in opinions is limited (i.e., when k < a"2 u), then the short sales constraint
is not binding for any investor and the equilibrium price coincides with the one
obtained under homogeneous expectations given by the average opinion d. N On the
contrary, when heterogeneity becomes more relevant (i.e., when k a"2 u), then the
equilibrium price only reflects the opinions of the more optimistic investors with the
equilibrium price being higher than the one obtained under homogeneous opinions.
In this sense, difference of opinions together with short sale constraints may lead to
an overvaluation of the asset. In particular, observe that an increase in the dispersion
of opinions (as measured by k) will lead to an increase in the equilibrium price p .
This model has been extended in Chen et al. [418] by allowing for the presence
of a group of fully rational arbitrageurs who can take either long or short positions
in the risky asset. Similarly as in the above setting, Chen et al. [418] consider a
two-period economy where a riskless asset with constant rate of return rf D 1 is
traded together with a single risky asset with random dividend dQ D dN C ", Q where
"Q is a normally distributed random variable with zero mean and unitary variance.
The total supply of the risky asset is supposed to be constant and given by u.
There are two groups of agents. The first group is composed of risk averse agents
(whose preferences are represented by negative exponential utility functions with
a common risk aversion parameter a) with different opinions about the expected
dividend of the risky asset: similarly as above, each agent believes that the expected
dividend is given by dN i and we assume that dN i is uniformly distributed on the interval
ŒdN k; dN C k. For all the agents belonging to this first group, short sales are
prohibited. The second group of agents is composed of fully rational arbitrageurs
(correctly believing that the expected dividend is equal to d)N who can take either long
or short positions in the risky asset. The agents of this second group have negative
exponential utility functions and their aggregate risk aversion is represented by the
parameter aarb . Under these hypotheses, we can establish the following version of
Proposition 8.16 (the proof follows the same steps of the proof of Proposition 8.16
and is given in Exercise 8.21).
Proposition 8.17 In the context of the model described above (see Chen et al.
[418]), if the agents of the first group are not allowed to short sell the risky asset
and the agents of the second group are not constrained, then the equilibrium price
p at date t D 0 of the risky asset is given by
8 1 1
ˆ
<dN 1a C 1
u; if k < 1
C 1
uI
aarb a aarb
p D q
:̂dN C k C 2ak 1
1
C 1
C u
; otherwise.
aarb a2arb aaarb ak
(8.38)
Chen et al. [418] also analyse the breadth of ownership, defined as the fraction
of agents of the first group who hold long positions in the stock, and show that
as the divergence of opinions increases (i.e., k increases) then both the breadth
of ownership and the expected return dN p decrease. Observe that the result of
Proposition 8.16 can be recovered from Proposition 8.17 (in the special case "2 D 1)
by letting aarb ! C1, representing the fact that arbitrageurs are not trading in the
market.
The interplay of heterogeneous opinions and short sale constraints has been also
analysed in the context of a multi-period model in Harrison & Kreps [903]. They
show that the combined presence of these two features may generate speculative
behavior and may lead to an equilibrium price which is greater than the fundamental
value of the asset (in other words, there exists a bubble, see Sect. 6.5). According
to Harrison & Kreps [903], agents are said to exhibit speculative behavior if “the
right to resell a stock makes them willing to pay more for it than what they would
pay if obliged to hold it forever”. In the model proposed in Harrison & Kreps [903]
(compare also with the presentation in Scheinkman & Xiong [1503, Section 3]), the
economy is characterized by an infinite trading horizon, with trading dates t 2 N.
Similarly as in Chap. 6, the information flow is represented by F D .Ft /t2N . In
the economy, there is one unit of a single risky asset, delivering a non-negative
random dividend dt at each date t 2 N. The dividend process .dt /t2N is assumed to
be adapted to F. In this context, a price process for the risky asset is represented by
a non-negative stochastic process .pt /t2N adapted to F. The economy is populated
by A 2 N homogeneous classes of risk neutral agents. Each agent is assumed to
believe in a subjective probability measure (on a common measurable space) and
agents belonging to the same class share the same probability measure, but agents
belonging to different classes may exhibit different probability measures (opinions).
For each a 2 f1; : : : ; Ag, we denote by Pa the probability measure (opinion) of
the agents belonging to class a, with Ea Œ denoting the associated expectation.
Each class consists of an infinite number of agents. All the agents have a common
discount factor ı.
In the above setting, Harrison & Kreps [903] introduce the crucial hypothesis
that agents are prohibited from short selling the risky asset. Hence, together with the
assumption of risk neutrality, the analysis of the trading strategies at date t reduces
to the analysis of strategies consisting in selling one unit of the risky asset at a
future date T t (we also allow for the possibility of T D C1, i.e., the asset is
held forever). Following Harrison & Kreps [903], a price process .pt /t2N is said to
be consistent if
" T #
X ˇ
ˇ
pt D max sup Ea ı kt dk C ı Tt pT ˇFt ; for all t 2 N: (8.39)
a2f1;:::;Ag Tt
kDtC1
As explained in Harrison & Kreps [903, Section IV], condition (8.39) represents a
natural condition for a price process .pt /t2N to correspond to an equilibrium of the
economy. In fact, the quantity
" #
X
T ˇ
ˇ
sup E a
ı kt
dk C ı Tt
pT ˇFt
Tt kDtC1
represents the maximum present value that an agent belonging to class a can realize
by holding the risky asset at date t and selling it at some optimal future date T,
given the information available at date t. By the assumption of risk neutrality, this
represents the utility associated to owning the asset at date t for the agents belonging
to class a. Hence, the right-hand side of (8.39) is the maximum value of the risky
asset from the point of view of any agent in the economy. Indeed, if such an amount
was strictly greater than a candidate equilibrium price pt , then competition among
the (infinite number of) agents belonging to the class assigning the largest value
to the asset would make the price increase. A symmetric argument shows that the
maximum value of the risky asset from the point of view of any agent in the economy
cannot be smaller than pt , thus showing that an equilibrium price process must be
consistent in the sense of (8.39) (see also Harrison & Kreps [903, Section 4]).
The following proposition gives a characterization of consistent price processes
and should be compared with the fundamental no-arbitrage equation (6.85) (under
the assumption of risk neutrality). The proof of the following proposition is given in
Exercise 8.22.
Proposition 8.18 In the context of the model described above (see Harrison &
Kreps [903]), a price process .pt /t2N is consistent if and only if it satisfies the
following condition:
pt D max ıEa ŒdtC1 C ptC1 jFt ; for all t 2 N: (8.40)

a2f1;:::;Ag
Note also that, for any consistent price process .pt /t2N , since T D C1 is always
a feasible selling strategy, condition (8.39) implies that
" #
1
X ˇ
ˇ
pt max E a
ı kt
dk ˇFt for all t 2 N: (8.41)
a2f1;:::;Ag
kDtC1
In (8.41), the right-hand side represents the value of holding the asset forever.
Speculative behavior arises if the inequality is strict: in this case the option to
resell the risky asset at a future date has a strictly positive value and this option
will become viable when there are agents who are relatively more optimistic than
the current owner about the future dividends’ growth. This leads to a bubble
with a positive price wedge with respect to the fundamental no-arbitrage solution.
Summing up, the fundamental insight of Harrison & Kreps [903] is represented
by the observation that, in the presence of heterogeneous opinions and short sale
prohibition, agents can agree to disagree and prices may exceed fundamental values.
Related results have been also obtained in Scheinkman & Xiong [1502], consid-
ering the combined effect of short sale constraints and heterogeneous opinions in
the formation of speculative bubbles. In the context of a continuous time model,
Scheinkman & Xiong [1502] assume that overconfidence (i.e., the belief of an
agent that his own information is more accurate than what in fact it is) generates
disagreement among agents regarding asset fundamentals. More specifically, agents
disagree about the interpretation of publicly available signals, with different groups
of agents overestimating the informativeness of different signals. As information
flows into the market, investors’ forecasts change and, in turn, these changes
generate trading volume. Furthermore, an increase in the degree of overconfidence
induces an increase in the value of the bubble component (see Scheinkman & Xiong
[1502, Lemma 3]). Related results have been also obtained in Hong et al. [960] in
the context of a discrete time multi-period model. In this context, see also Dumas
et al. [601] for portfolio choice analysis and Xiong & Hongjun [1671] for models
on the term structure and asset float.
Hong & Stein [962] analyse the impact of differences of opinions and short sale
constraints on the possibility of market crashes. Their model considers two short
sales constrained investors A and B, each of whom can observe a private signal about
the stock’s terminal payoff, together with a group of rational arbitrageurs (not short
sales constrained). The model is able to generate market crashes and the rationale of
this result is that the presence of short sale constraints prevents pessimistic agents’
beliefs to be fully reflected in market prices. Indeed, suppose that at some date t D 1
agent B receives a more pessimistic signal than agent A and decides to stay out of
the market. In this case, at date t D 1, trading will only occur between agent A
and the arbitrageurs. The arbitrageurs (who cannot observe the private signals of the
two agents) will only infer that the signal of agent B is more pessimistic than that
of agent A, but cannot determine to what extent. Hence, market prices at t D 1 do
not fully reflect the pessimistic view of agent B. Suppose now that at a later date
t D 2 agent A receives a bad signal and wants to bail out the market. If the signal
received at the previous date t D 1 by agent B was very bad, then agent B won’t offer
buying support to the sell orders of agent A. In turn, this event will be interpreted
as a bad news by the arbitrageurs, beyond the direct bad news already represented
by the desire to sell of agent A. This behavior can generate a cascade effect. In this
model, large trading volume (due to the difference of opinions) forecasts a negative
skewness in asset returns.
In the context of a continuous time model with heterogeneous opinions, the role
of equilibrium prices in aggregating different opinions is analysed in Detemple &
Murthy [564]. They consider a model where agents have heterogeneous opinions
about the unobservable expected growth rate of the aggregate production (while
past realizations of the production growth rate are publicly observable). Agents
have different prior distributions, update their beliefs rationally and disagree on the
interpretation of economic news. As a consequence, there is persistent disagreement
among the agents (i.e., posterior distributions differ). The equilibrium risk free rate
and asset prices are weighted averages of those obtained in equilibrium for the
corresponding homogeneous economies (i.e., the economies where all the agents
share the same opinions of the respective agents of the original heterogeneous
economy), with the weights being determined by the wealth distribution among
the agents. In a similar continuous time setting for a pure exchange economy,
incomplete (symmetric) information and heterogeneous opinions about a process
which is unrelated to asset fundamentals can generate a low risk free rate and
excess volatility in equilibrium (see Basak [175]). In this setting, the perceived risk
premium of a security is not purely explained by the covariance with the aggregate
consumption (compare with Proposition 6.35), but also by the covariance with the
exogenous uncertainty (see also Abel [3] and Zapatero [1672] for related results).
8.6 Empirical Analysis
It is difficult to empirically test the informational efficiency of financial markets.

Such a difficulty is also due to the joint hypothesis problem, since any test of market
efficiency is a joint test on market efficiency as well as on the specific equilibrium
model on which market efficiency is being tested. In very general terms, the goal of
an empirical analysis of market efficiency consists in testing whether and to what
extent market prices transmit private information and aggregate the agents’ beliefs.
Information Aggregation
The empirical analysis of the role of market prices in aggregating private informa-
tion goes back to Figlewski [702], where it was investigated whether horse quotes
reflected the opinions of professional bidders. The empirical analysis reported in
Figlewski [702] provided positive evidence.
In Lang et al. [1166], a test has been developed based on the analysis of
trading volume and, in particular, of its correlation with market prices and some
variables representing agents’ expectations/beliefs. The test allows to distinguish
among four of the equilibrium models presented above: competitive equilibrium,
competitive equilibrium with random supply, Green-Lucas equilibrium and Green-
Lucas equilibrium with random supply. The test concerns the market behavior when
earning announcements are released by firms. The empirical evidence seems to
be consistent with a Green-Lucas equilibrium model with random supply: asset
prices reveal only part of the private information. Agents are rational, update their
beliefs by exploiting the private information and identify the informational content
of market prices. However, as explained earlier in this chapter, the presence of a
noise component prevents a perfect transmission/aggregation of private information
by market prices. Similar results on the capacity of financial markets to aggregate
dispersed information have been obtained in Huberman & Schwert [986] in the case
of information about inflation in a bond market.
In this regard, some interesting results come from experimental economics (see,
e.g., Camerer [327], Forsythe & Lundholm [728], Sunder [1581], Smith et al.
[1552], Plott & Sunder [1422], McKelvey & Page [1314], Bossaerts et al. [271]).
Noisy (i.e., in the presence of a noise component such as random supply) rational
expectations equilibria have been shown to be an adequate representation of the
market behavior if some of the agents are informed and some are not. However,
the same equilibrium concept does not perform well if every agent in the economy
observes a signal with a private noise component: in this case, the aggregation of
the information of the agents is not always reached in experimental markets (and,
even when the aggregation is reached, it may be reached after many iterations).
Furthermore, in the presence of a monopolistic position by the agent possessing
private information, market crashes and bubbles are typically observed before the
8.6 Empirical Analysis 459
convergence is reached. In this context, it is difficult to observe a market behavior

compatible with a rational expectations equilibrium, so that experimental results
confirm only in part the theoretical implications of rational expectations equilibrium
theory (see however Barner et al. [165] for results in favor of fully revealing
equilibria).
Volatility, Volume and Returns
In Chap. 7, we have reported some empirical evidence pointing out that classical
asset pricing theory fails to explain the volatility of asset prices. The empirical
literature surveyed in that chapter is mostly concerned with low frequency data.
High frequency data show that asset prices/returns are typically much more volatile
during trading hours than during non-trading hours (see French & Roll [741]).
Indeed, the variance of stock returns from the opening to the close of the market
is greater than the variance from the close to the opening (see also the related
discussion in Sect. 10.6). Three explanations of this phenomenon have been
proposed: more public information arrives during normal business hours; private
information affects prices only during trading hours; there exist pricing errors. Ross
[1470] has proposed a model relating asset price volatility to the rate of information
flow into the market. Examining returns during trading holidays, it turns out that a
large part of the variance difference is due to private information. Ex-post public
information about fundamentals allows to explain only a small part of the variance
and this finding has been confirmed by many other studies (see, e.g., Mitchell &
Mulherin [1347], Barclay et al. [160], Barclay & Warner [161], Berry & Howe
[208], Roll [1459, 1460], Cutler et al. [510]). An interpretation in favor of public
information as a major source of volatility has been proposed in Ederington &
Lee [628], Jones et al. [1040], Stoll & Whaley [1571]: after controlling for these
announcement effects, volatility is basically flat. Madhavan et al. [1278] show that
both public information and trading frictions are important factors in explaining
intraday price volatility.
Empirical studies have also shown the existence of a high volume puzzle: it
is difficult to explain the huge amount of trading observed in financial markets
by relying on a rational expectations equilibrium framework. The empirical liter-
ature has also reported a strong positive correlation between trading volume and
contemporaneous as well as delayed return volatility/price changes (price changes
lead to volume movements) and positive serial correlation in trading volume (see,
e.g., Karpoff [1075], Gallant et al. [751, 752]). In Jones et al. [1041], Easley et al.
[617], Chan & Fong [404], it is shown that the number of transactions rather than
trading volume is related to volatility. In general terms, these findings suggest
that volume and volatility are driven by the same factors. Note that the above
regularities represent empirical confirmations of some of the implications of the
models presented earlier in this chapter in the presence of random supply and
heterogeneous beliefs/opinions (see, e.g., Kim & Verrecchia [1095], Blume et al.
[257], Harris & Raviv [901], Foster & Viswanathan [730], Shalen [1522], Wang
[1644], He & Wang [920], Bernardo & Judd [203]). Many factors can contribute
to the trading volume observed in the market: differences in endowments, liquidity
trading, differences/shifts in preferences, heterogeneous information, heterogeneous
valuations/beliefs/opinions and the arrival of new information. While differences in
the endowments do not suffice to explain the large trading volume observed in real
markets, the other factors have been empirically investigated.
According to Kim & Verrecchia [1095], Kandel & Pearson [1066], Lang et al.
[1166], Harris & Raviv [901], Foster & Viswanathan [730], Hong & Stein [962], He
& Wang [920], the arrival of public information in the market generates trading.
Trading volume is positively related to price changes and the strength of the
relation is increasing with respect to the differential quality of private information,
the degree of disagreement (opinions/valuations/beliefs) and the market noise. To
this effect, some empirical evidence is reported in Bamber et al. [128], based on
data on analysts’ revisions of forecasts of annual earnings after the announcement
of quarterly earnings. The relation between news and volume has been tested
empirically with positive evidence in Berry & Howe [208], Kandel & Pearson
[1066], Bessembinder et al. [215] and with mixed results in Mitchell & Mulherin
[1347].
Trading volume is related to returns. Brennan & Subrahmanyam [297], Amihud
[57], Easley et al. [615], Kelly & Ljungqvist [1084] report a positive relation
between adverse selection proxies and risk premia. Campbell et al. [346] show
empirically (and theoretically) that return serial correlation (for both stock indexes
and individual large stocks) tends to decline with past volume (that is mainly due
to liquidity trading). This regularity has been empirically confirmed in Conrad et al.
[482] and Avramov et al. [94]: price reversals are typically observed for widely
traded stocks, while returns of less traded securities tend to be positively auto-
correlated. The implications derived in Wang [1644] on the relationship between
return dynamics and the effects on trading volume of informational/liquidity trading
have been confirmed empirically in Llorente et al. [1232]: informational trading
induces positively autocorrelated returns and liquidity trading induces negatively
autocorrelated returns. This relation determines whether returns accompanied by
trading volume exhibit negative or positive serial correlation (compare with the
discussion in Sect. 8.4). In Chae [385] it is shown that trading volume before public
announcements is decreasing with respect to adverse selection proxies (see also
Tetlock [1586] on public announcements and trading volume).
The relation between trading volume and the profitability of momentum strate-
gies has also been investigated: in Lee & Swaminathan [1175] and Chan et al.
[405] it is shown that assets with high trading volume in the past earn lower
returns in the future and exhibit faster price reversals than assets with low trading
volume (this result is aligned with the results reported in Campbell et al. [346]).
Chordia & Swaminathan [440] show that volume is a significant determinant of the
lead-lag patterns observed in stock returns: daily/weekly returns on high volume
portfolios predict returns on low volume portfolios and returns on low volume
portfolios respond more slowly to information in market returns. This effect is due to
8.6 Empirical Analysis 461
differentiated diffusion of market information and shows that the differential speed
of adjustment to information is a significant source of the cross-autocorrelation
patterns observed in short horizon returns.
Difference of Opinions
As pointed out in Hong & Stein [963], a nice feature of models with hetero-
geneous opinions is represented by the fact that they allow to capture the joint
behavior of returns and trading volume. In particular, as mentioned above, the
large trading volume observed in financial markets is difficult to reconcile with
classical asset pricing theory, even allowing for heterogeneous/asymmetric beliefs
and noise/liquidity traders. Indeed, while these models are able to relate volatility
and trading volume, they do not match other empirical regularities such as the fact
that overvalued assets (i.e., high market-to-book ratio assets) tend to have higher
volume than undervalued assets and that stocks with large volume tend to have
low future returns. In particular, it is difficult to explain the size of trading volume
observed in real financial markets and the excess volatility phenomenon by relying
on the hypothesis of heterogeneous beliefs. The presence of liquidity traders does
not suffice by itself to explain these phenomena. On the other hand, models with
difference of opinions are able to provide a theoretical explanation of these empirical
facts, being able to generate a large trading volume even when asset prices do not
move relative to the fundamentals.
As we have seen in Sect. 8.5, models with difference of opinions often assume the
existence of short sale constraints. In Figlewski [703], using observed short interest
rates as a proxy for the level of short sale constraints, some empirical evidence is
reported showing that more heavily shorted stocks underperform less shorted stocks.
Boehme et al. [262] find empirical evidence of significant overvaluation for stocks
characterized by a large dispersion of investors’ opinions and short sale constraints,
in line with the results of Propositions 8.16 and 8.17. Similarly, Diether et al. [577]
show that stocks with a larger dispersion in analysts’ earnings forecasts (that can
be regarded as a proxy for differences in opinions) earn lower future returns than
otherwise similar stocks (in this direction, see also Yu [1668], Sadka & Scherbina
[1489], Chordia et al. [432]). This evidence is consistent with the hypothesis that
prices reflect the optimistic view whenever investors with the lowest valuations
do not trade (see however Doukas et al. [583] for negative evidence). Diether
et al. [576] provide positive evidence to the short sales constraint overvaluation
hypothesis by showing that a strong short selling activity predicts negative future
returns. By means of an experimental analysis, Palfrey & Wang [1397] report
evidence of speculative overpricing both in complete and incomplete markets, where
the information flow is represented by a gradually revealed sequence of imperfect
public signals about the state of the world.
In recent years, a growing literature is providing positive empirical evidence
of the main findings of models with differences of opinions. For instance, Chen
et al. [417, 418] confirm the main implications of Hong & Stein [962]: negative
skewness in returns/negative returns are more likely to occur in periods with high
trading volume (which is in turn a proxy for the degree of disagreement). In general,
“glamour stocks” (i.e., stocks which are overvalued according to typical market-
based ratios) tend to have high volume and to be characterized by low future returns,
see Brennan et al. [293], Chordia et al. [439] and Chap. 5. Kandel & Pearson [1066]
empirically document the relationship between volume, disagreement and return
volatility in correspondence of public announcements.
Banerjee [129] provides a comparison of the hypotheses of heterogeneous
beliefs and heterogeneous opinions. When investors have rational expectations
and infer private information from the observation of market prices (as in the
case of heterogeneous beliefs), the degree of dispersion of beliefs is positively
related to expected returns, return volatility and correlation between volume and
absolute returns, but negatively related to return autocorrelation. On the contrary,
when investors do not condition on market prices (as in the case of heterogeneous
opinions) these relationships are reversed. In both cases, assets exhibiting a higher
disagreement have a higher expected volume as well as a higher variance of volume.
The empirical evidence reported by Banerjee [129] is in favor of the hypothesis that
agents condition on market prices (and, therefore, infer private information from
market prices) on the basis of quarterly and longer horizons, but reject both models
at the monthly horizon.
From a behavioral perspective, a crucial point in order to understand the
empirical findings reported above concerns the identification of a suitable proxy
of the investor sentiment (i.e., beliefs about future cash flows that are not related by
fundamentals), see Baker & Wurgler [111]. In line with the theoretical predictions
of several difference of opinions models, stocks or markets characterized by a
high trading volume exhibit lower subsequent returns. Moreover, when the investor
sentiment is high, future returns are low on stocks in which there are limits to
arbitrage and on stocks that are difficult to value (see Baker & Stein [108], Datar
et al. [525], Brennan et al. [293], Chordia et al. [439], Lee & Swaminathan [1175]
and also Baker & Wurgler [110], Stambaugh et al. [1563], Baker et al. [112] for
an analysis with a refined market sentiment indicator). We also want to mention
that the empirical evidence reported above on the relation between heterogeneity of
opinions and trading volume/asset returns has been also identified as an evidence of
information uncertainty (information risk, see Zhang [1677] and Johnson [1036]).
In Sect. 8.1, we have considered the role of information in financial markets. A more
detailed analysis of the value of information from the point of view of an individual
agent can be found in Gollier [800, Part VIII], relating information and risk aversion.
The existence of a Green-Lucas equilibrium and its fully revealing property have
been addressed in Radner [1438] in the context of an economy with a single good,
incomplete markets, real assets, a finite number of elementary events and finite
set of possible signals for the agents. Under these assumptions, it is shown that
a fully revealing Green-Lucas equilibrium exists for a generic set of economies.
The assumption on the finiteness of the states of the world and of the signals
observed by the agents has been relaxed in Allen [42, 43], where it is shown that
a fully revealing Green-Lucas equilibrium generically exists if the dimension of
the signal space is less than the number of relative prices (equal to N 1 if the
market comprises N securities). In this context, the dimension of the signal space
can be interpreted as the number of sources of information for each agent. This
result can also be extended to models with infinitely many states and/or with signals
having continuous distributions. In Jordan & Radner [1047] it is shown that, if the
number of relative prices is equal to the dimension of the signal space, then there
exists an open set of economies which do not admit an equilibrium and open sets of
economies admitting fully revealing equilibria. In particular, this result implies that
a fully revealing Green-Lucas equilibrium may not exist in an economy with a risky
asset, a risk free asset and a private signal taking values in R. If the dimension of
the signal space is larger than the number of relative prices, then the existence of a
Green-Lucas equilibrium is ensured for a dense set of economies, but full revelation
does not hold generically. As shown in Jordan [1045], a Green-Lucas equilibrium is
generically partially revealing and approximately fully revealing. Moreover, Jordan
[1046] shows that the equilibrium is fully revealing in three possible cases: (i) when
all investors are risk neutral; (ii) when all agents have identical constant relative risk
aversion; (iii) when all agents have constant absolute risk aversion.
Examples of the non-existence of fully revealing equilibria in incomplete markets
have been given in Kreps [1134], Radner [1438], Jordan & Radner [1047]. In
particular, a generic non-existence result has been established in Jordan & Radner
[1047]. The analysis has been extended in Pietra & Siconolfi [1419] to economies
with multiple numéraire assets and physical commodities, with finitely many states
of private information. In this context, it is shown that all rational expectations
equilibria are fully revealing for a generic dense subset of economies. Classes
of economies characterized by (robust) non-fully revealing equilibria have been
identified in Ausubel [90]. DeMarzo & Skiadas [556, 555] identify a class of
economies with private information (quasi-complete economies) which always
admit a unique fully revealing equilibrium and, under some conditions, partially
revealing equilibria. An economy is quasi-complete if a risk neutral probability
measure exists and, given the private information and the information revealed
by prices, equilibrium allocations are interim Pareto optimal (see Definition 8.3).
Under asymmetric information, assuming that all agents exhibit linear risk tolerance
(i.e., HARA utility functions) with a common slope and that their endowments are
tradable, demand aggregation holds even under partially revealing equilibria and the
economy is quasi-complete (see also Madrigal & Smith [1283]). In this case, when
agents exhibit quadratic utility functions, a conditional version of the CAPM holds.
Assuming that nominal assets are traded, it has been shown that partially
revealing rational expectations equilibria exist in an incomplete markets economy
with a finite dimensional signal and state space (see Polemarchakis & Siconolfi
[1425]). In Rahi [1440], it has been shown that any information flow revealed by
prices (from fully revealing to fully non-revealing equilibria) that is consistent with
the absence of arbitrage opportunities can be obtained in a rational expectations
equilibrium (n this context, see also Citanna & Villanacci [453]).
As we have mentioned in Sect. 8.1, Proposition 8.4 can be regarded as a no-
trade result. Considering an economy with a single asset, a no-trade result has been
established in Kreps [1134] and in Tirole [1594]. In particular, in Tirole [1594,
Proposition 1] it is shown that in a purely speculative market (in an economy where
the agents’ endowments are uncorrelated with the return of the asset, relative to the
information available to each agent) at equilibrium no trade occurs if agents are risk
averse. At equilibrium, risk neutral agents may trade, but they do not expect any
gain from trading. An early version of the no-trade theorem has been established in
Rubinstein [1486]. We also want to mention that, in the presence of both informed
and uninformed agents, the no-trade result may not hold at equilibrium even in
the absence of noise (see, e.g., Dow & Gorton [585], under the assumption that
uninformed agents cannot buy the market portfolio).
Orosel [1386] provides an analysis of the Hirshleifer effect in an overlapping
generations model, where at each period all agents receive information about the
future dividends of the assets. The author shows that there exists a unique stationary
equilibrium and, in the presence of market incompleteness, informationally efficient
prices do not lead to constrained Pareto optimal allocations. In this context, more
information may be socially harmful. On the efficiency of rational expectations
equilibria see also Green [821], Hankansson et al. [888], Schlee [1504], Eckwert
& Zilcha [625, 626], Gottardi & Rahi [813].
In Sect. 8.3, we have presented several models admitting a partially revealing
rational expectations equilibrium. In this context, it is important to remark that
the results of Grossman & Stiglitz [849] turn out to be very sensitive to the
hypotheses of the model. For instance, several of the comparative statics results
given in Proposition 8.15 do not hold in the case of CRRA preferences with
log-normally distributed returns (see Bernardo & Judd [203]). Moreover, strategic
complementarity in the acquisition of information can solve the Grossman-Stiglitz
paradox. Manzano & Vives [1301] and Chamley [398] show the robustness of the
substitution effect in the process of information acquisition. Models where a costly
information acquisition coexists with fully revealing equilibrium prices have been
proposed in Krebs [1133] and Muendler [1359]. Furthermore, in a model with risk
neutral agents and in the presence of no-borrowing constraints (i.e., agents cannot
spend more than their original endowments), Barlevy & Veronesi [163] show that
the paradox is solved and prices do not become more informative as more agents
acquire information, under different distributional assumptions. In the same model,
uninformed agents may generate a market crash (see Barlevy & Veronesi [164]): if
they observe a low price generated for example by pure liquidity trading, then they
can infer that informed agents have observed a bad signal and revise consequently
their expectations, thereby generating a selling panic among uninformed traders.
Garcia & Strobl [756] show that complementarities in the acquisition of information
arise if each agent’s satisfaction with his own consumption depends on how much
other agents are consuming (relative wealth concerns, see also Sect. 9.2). We also
want to mention that an extension of the model proposed by Grossman & Stiglitz
[849] to markets with multiple assets has been considered in Veldkamp [1615]: if
information is costly, rational investors only buy information related to a subset
of the available assets. This behavior induces strategic complementarities in the
acquisition of information and common movements in asset prices.
The model of Hellwig [931] presented at the beginning of Sect. 8.3 has been
extended in Admati [15] to a setting with multiple risky assets and infinitely many
agents. Kim & Verrecchia [1095] analyse a model with costly private information
similar to that proposed in Verrecchia [1618] (briefly discussed at the end of
Sect. 8.3), with the goal of understanding how anticipating a public announcement
through information gathering affects the market reaction to the announcement.
There are two trading periods: in the first one agents with common priors and
private information with different precision trade in the market, in the second there
is a public announcement about the dividend of the stock. Supply in the market is
noisy. The anticipated impact of the public announcement affects the acquisition
of private information. The main result of the analysis is that trading volume in
the second period is equal to the absolute price change from period one to period
two multiplied by an aggregate measure of the agents’ individual idiosyncrasy. This
measure is mainly due to the differential quality of private information across agents,
adjusted for the investors’ preferences for risk, and can be interpreted as a proxy
of information heterogeneity. Volume sensitivity to price changes is increasing in
the agents’ information heterogeneity. The model generates a positive relationship
between volume and absolute price changes (volatility). Information asymmetry
first increases with the precision of the public announcement and then decreases,
so that volume is sensitive to absolute price changes when the public announcement
precision is neither too high nor too low. The variance of price changes and expected
trading volume are decreasing in the quality of prior information and increasing in
market noise and in the precision of the public announcement. Similar results are
obtained in Lang et al. [1166], Foster & Viswanathan [730] by allowing agents to
acquire private information after a public announcement. Both volume and volatility
depend on the public announcement and volume is positively serially correlated.
On the relationship between public announcement, private information and trading
volume see also Kim & Verrecchia [1096, 1097].
The analysis on partially revealing equilibria developed in Sect. 8.3 refers to
a single trading period. Since markets are open for a single period, it is difficult
to detect agents’ private information through prices if the aggregate supply is
random. However, if agents trade over several periods, then they can learn more
information over time. In Brown & Jennings [310] and Grundy & Martin [851] it
is shown that, if agents are allowed to trade over several periods, then they will
eventually identify the information through market prices. This result provides a
theoretical foundation for the use of trading strategies based on technical analysis.
In other words, prices can be regarded as a noisy signal of private information: by
observing prices over several periods the underlying information can be identified
more precisely and, consequently, the agents’ trading strategies will depend on past
prices. A similar conclusion is drawn in Blume et al. [257] and Schmeidler [1506]
by allowing agents to observe market volume. In particular, in Blume et al. [257]
agents observe private signals of different precision and the precision of each agent’s
private signal is not common knowledge. Due to this assumption, trading volume
has a predictive power and agents can learn useful information by observing market
volume (technical analysis based on price and volume). In equilibrium, it is shown
that absolute price changes and volume are strongly correlated and, in the limit,
prices converge to the full information value but trading volume does not vanish.
See also Brunnermeier [319] on the role of technical analysis in models with partial
revelation of information.
In Sect. 8.3, we have considered the possibility of acquiring private information
from an equilibrium point of view. However, the possibility of acquiring costly
information also affects portfolio choices. Van Nieuwerburgh & Veldkamp [1608]
show that agents tend to acquire information on a limited set of assets and hold an
undiversified portfolio. Moreover, the information acquisition process can explain
the home bias (i.e., the fact that individual and institutional investors in most
countries typically hold modest amounts of foreign stocks). Investors overweight
the assets belonging to the domestic market in their portfolio and exhibit a
scarce international diversification, thus holding non-optimal portfolios (see Van
Nieuwerburgh & Veldkamp [1607]). Peress [1412] shows that the possibility of
acquiring costly information in conjunction with DARA preferences allows to
explain why wealthier people invest a large proportion of their wealth in risky
assets (see also the correction note Peress [1413]). The presence of asymmetric
information has been proposed as a possible explanation of the home bias puzzle,
by considering that foreign investors are less informed than domestic investors about
the domestic market and, as a consequence, domestic investors hold a large part of
their portfolio in domestic assets. A noisy rational expectations model to explain
this puzzle through asymmetric information (in the sense that domestic investors
have privileged information) has been proposed in Brennan & Cao [291] and Zhou
[1681] (see also Kang & Stulz [1072]). According to this model, there should be
a positive correlation between contemporaneous local equity returns and the net
purchase of equity made by foreign investors (trend-follower pattern). Foreigners
are more likely to buy when the price is increasing. This trend-following pattern of
foreign investors has been confirmed empirically.
At the end of Sect. 8.3, we have briefly discussed the model of Admati &
Pfleiderer [18], where there exists a market for information. In this direction, Garcia
& Sangiorgi [755] study information sales in financial markets with strategic risk
averse traders. The optimal selling strategy turns out to be similar to the one
obtained in Admati & Pfleiderer [18]: either sell a very imprecise information
to as many agents as possible, or sell a very precise information to a limited
number of agents. As risk sharing considerations prevail over the negative effects
of competition, the first strategy dominates the second one. Another approach to
the modeling of markets for information is represented by assuming that the agency
acts as a mutual fund by building a portfolio exploiting its own information and then
selling shares of the fund in the market. In this case, the information is being sold
indirectly. This case has been studied in Admati & Pfleiderer [22] by considering
a monopolistic mutual fund (see Garcia & Vanden [757] for an analysis under
imperfect competition). If information is being sold indirectly in an economy with
homogeneous agents, then there is no incentive to introduce a noise component.
Moreover, an indirect sale of information is more profitable than a direct sale if the
agency can establish a per share price and a fixed participation fee. However, when
agents are heterogeneous, an indirect sale can be either more or less profitable than
a direct one (see Biais & Germain [235] on the optimal contract for an indirect sale
of information by a mutual fund). Under some conditions, it is optimal for a broker
to sell information in return for a brokerage commission (see Brennan & Chordia
[292]). See also Fishman & Hagerty [711] for an analysis of the incentives for an
informed trader to sell private information.
In the present chapter, we have not discussed the role of information disclosure
by firms in the market (see Verrecchia [1620] for a survey on this topic and see
also Chap. 10). In Diamond [568], it is shown that public announcements by firms
can induce a Pareto improvement (more efficient risk sharing), lesser costs of
acquiring information and of issuing capital (see also Diamond & Verrecchia [570]).
In general, disclosure leads to liquid and more efficient markets, more efficient
investment decisions and reduces the cost of capital for firms. As a consequence,
firms should voluntarily disclose their private information to the market. In Fishman
& Hagerty [708] it is shown that, if firms need to compete in order to capture the
attention of traders, then an overdisclosure of information can occur. In a related
context, Boot & Thakor [268] show that voluntary disclosure is beneficial for
the economy and the firm, but when the disclosed information complements the
information which is already present in the market, then agents buy more private
information.
Despite the positive effects for the economy and for the companies, full voluntary
disclosure is quite rare. Admati & Pfleiderer [24] build a model on regulation of firm
disclosure based on an externality that leads individual firms not to internalize the
full social value of the information released. The externality is due to the presence
of correlation in the firms’ values and, as a consequence, the information disclosed
by one firm is used by investors to evaluate other firms. The Nash equilibrium
of a voluntary disclosure game is often inefficient: both underinvestment and
overinvestment in disclosure are possible, depending on the degree of correlation
among the firms. In this context, there is room for a regulation encouraging firm’s
disclosure. Under some conditions, high type (low type) firms withhold good
(bad) news and disclose bad (good) news (see Teoh & Hwang [1584]). Other
works studying the relations between corporate disclosure, corporate behavior and
capital markets are Leuz & Verrecchia [1202], Healy & Palepu [922], Verrecchia
[1619, 1621], Lambert et al. [1161].
In Sect. 8.4, we have briefly mentioned the phenomenon of higher-order beliefs,
which play a crucial role in the analysis of multi-period market models. In particular,
considering a multi-period economy with overlapping generations of risk averse
agents who can observe a private signal at each date, Allen et al. [50] show that the
law of iterated expectations does not hold for the average belief. Bacchetta & van
Wincoop [95] show that higher-order beliefs induce a wedge (called higher-order
wedge) between the equilibrium price of an asset and the equilibrium price of an
asset if higher-order expectations were replaced by first-order expectations (i.e., in
the case of homogeneous information). Furthermore, it is shown that the wedge at
each date t depends on the average expectational error at time t about the vector of
average private signals that remain informative about future dividends at t C 1 (see
Bacchetta & van Wincoop [95, Proposition 1]). Moinas & Pouget [1350] provide an
experimental analysis showing that higher-order beliefs may generate a bubble: the
probability to enter into a bubble is increasing with respect to the number of steps
of iterated reasoning needed to reach an equilibrium. More recently, Banerjee et al.
[131] have shown that, in a dynamic setting with homogeneous beliefs, higher-order
difference of opinions is a necessary condition for the existence of a drift in the price
process. This result stands in contrast to that of Allen et al. [50], since Banerjee et al.
[131] show that in a rational expectations equilibrium heterogeneous beliefs do not
lead to the existence of a drift.
In Sect. 8.5 we have introduced the possibility that agents exhibit heterogeneous
opinions. In that regard, Simsek [1547] analyse a model with difference of opinions
and borrowing constraints. Optimistic agents are assumed to have limited wealth and
need to borrow in order to take positions in line with their beliefs, borrowing from
more pessimistic agents using loans collateralized by the asset itself. However, due
to the fact that the collateral is provided by the asset itself, pessimistic agents are
reluctant to lend to the optimistic agents, thus inducing an endogenous constraint
on the ability of optimistic agents to influence market prices. This creates an
asymmetric disciplining effect, due to the asymmetry in the shape of the payoffs of
collateralized loans. In this setting, the level of opinion disagreement has ambiguous
effects on the equilibrium price.
It is important to mention that investors’ heterogeneity can arise from several
possible features of the economy. In a rather general setting, in Jackson & Peck
[1004] (compare also with Jackson [1003]) it is shown that a sunspot equilibrium
can exist in an overlapping generations economy where agents receive private
signals that are totally unrelated to fundamentals (i.e., variables not informative
about future dividends, like information about the psychology of the market).3 On
the basis of this information, agents trade with the aim of realizing a gain which is
not motivated by asset fundamentals. This shows the existence of speculative trading
generated by differences in information, even in the case of risk neutrality and of an
asset with no fundamental value. In equilibrium, it is shown that the asset price can
be larger than the fundamental value and bubbles can exist.
Among other possible sources of investors’ heterogeneity, an interesting aspect
is represented by tax heterogeneity. Indeed, agents can exhibit different evaluations
if they are subject to different taxation schemes. In particular, in Michaely et al.
3
Sunspots represent publicly observable extrinsic events that do not affect the fundamentals of
the economy, i.e., do not affect the agents’ endowments, preferences, budget constraints and
information.
8.8 Exercises 469
[1338] and Michaely & Vila [1337] it is shown that tax heterogeneity may generate
trading volume around ex-dividend dates. Another possible source of heterogeneity
is represented by the timing of information. In Berk & Uhlig [200], it is shown
that agents may choose to release information early (at a sufficiently small cost),
resulting in incompleteness of the market. Therefore, markets are unlikely to be
complete if the timing of information is endogenous and an equilibrium may not
even exist (see Berk [198]). As shown in Caplin & Lehay [364], the diffusion of
information is limited if agents change their behavior infrequently. In this setting,
the evolution of information is discontinuous and a small additional piece of
information can generate a market crash.
8.8 Exercises
Exercise 8.1 Let D .1 ; : : : ; S / be a vector of probabilities, i.e., 2 S , where

S is the simplex
( )
X
S
S WD 2 RSC W s D 1 :
sD1
Consider the optimal choice problem (8.2) and suppose that there exists an optimal
solution .x1 ./; : : : ; xS .// 2 RC
S , for each 2 S , and define
X
S
U ./ WD s u.xs .//:
sD1
Show that the map 7! U ./ from S onto R is convex in the vector of
probabilities.
Exercise 8.2 This exercise is taken from Eeckhoudt et al. [631, Section 8.2.2].
Consider the optimal saving problem of an agent over three dates t 2 f0; 1; 2g,
assuming that the agent is endowed with initial wealth w0 > 0, has a constant
discount factor normalized to one and is characterized by a strictly increasing and
strictly concave utility function u such that u000 > 0 (i.e., the agent is prudent). At
the final date t D 2, the agent receives the random income xQ . Consider the following
optimal saving problem, where ˛0 and ˛1 denote saving at dates t D 0 and t D 1,
respectively:

max u.w0 ˛0 / C EŒu.˛0 ˛1 / C u.Qx C ˛1 / : (8.42)
.˛0 ;˛1 /2R2
(i) Consider first the case where there is no early resolution of uncertainty, i.e.,
the random variable xQ is only observed at the final date t D 2. Show that in this
case the optimal saving decision satisfies ˛1 D 2˛0 w0 (i.e., it is optimal to
smooth consumption over the first two dates).
(ii) Consider then the case where the realization of the random variable xQ is
observed at the intermediate date t D 1. Show that in this case the optimal
saving ˛0i (where the superscript i stands for informed) from date t D 0 to
t D 1 satisfies the first order condition

i
0 0 ˛0 C x Q
u .w0 ˛0 / D E u
i
: (8.43)
2
(iii) Deduce that ˛0 ˛0i , i.e., an early resolution of uncertainty reduces the
optimal level of saving before the resolution of uncertainty.
Exercise 8.3 Let . a ; v a / and . b ; v b / be two information structures as described
before Proposition 8.2. Prove that the information structure . a ; v a / is finer than
the information structure . b ; v b / if there exists
P a .J J/-dimensional matrix K,
satisfying Kij 0, for all i; j D 1; : : : ; J, and JiD1 Kij D 1, for all j D 1; : : : ; J,
such that condition (8.6) holds.
Exercise 8.4 In the setting of Sect. 8.1, prove that ex-ante Pareto optimality implies
interim Pareto optimality and, in turn, interim Pareto optimality implies ex-post
Pareto optimality.
Exercise 8.5 (A No-Trade Result) This exercise is inspired by Scheinkman &
Xiong [1503, Proposition 2]. Consider a two-period (i.e., t 2 f0; 1g) economy
populated by I risk neutral agents i D 1; : : : ; I and where a single risky asset is
traded, delivering at date t D 1 the random dividend d. Q Denoting by zi the demand
of agent i (assumed to be bounded), the corresponding profits at t D 1 are given
by .dQ p/zi , with p denoting the price at date t D 0 of the asset. All the agents
are assumed to share the same prior probability distribution, but observe different
private signals yQ i . At equilibrium, every trader (by the assumption of risk neutrality)
maximizes his expected profit, conditioning on the information represented by his
private signal and the observation of the market price, as considered in Sect. 8.1.
Show that, in correspondence of a Green-Lucas equilibrium, the expected gains
from trading of each agent are null. Therefore, in correspondence of a Green-Lucas
equilibrium, there always exists another Green-Lucas equilibrium associated with
null optimal demands.
Exercise 8.6 Let zQ, xQ and yQ be three random variables. Show that zQ is a sufficient
statistic for the conditional density f . y; zjx/ if and only if the conditional distribution
of xQ given fQy D y; zQ D zg does not depend on y.
Exercise 8.7 This exercise is inspired by Laffont [1154, Section 9.2] (see also
Radner [1438] for a general analysis). Consider an exchange economy with two
goods, two agents and two possible states of the world. The preference relation of
8.8 Exercises 471
agent i, for i D 1; 2, is represented by the following state dependent utility function

ui .xi1 ; xi2 ; !/ D ˛!i log xi1 .!/ C .1 ˛!i / log xi2 .!/ ; for ! 2 f!1 ; !2 g;
with xin .!/ denoting the quantity of good n D 1; 2 consumed by agent i in

correspondence of the state of the world !. The initial endowments of the agents
are given by
ei1 .!1 / D ei1 .!2 / DW ei1 > 0;

ei2 .!1 / D ei2 .!2 / DW ei2 > 0; for i D 1; 2;
i.e., the endowments do not depend on !. Suppose that P.!1 / D , with 2 .0; 1/.
At date t D 0, the first agent receives a private signal yQ 1 which perfectly reveals the
state of the world, while the second agent receives an uninformative signal. At date
t D 0 (non-contingent) markets for the two consumption goods are open. Note that,
in the present setting, markets are incomplete.
(i) Suppose that the equilibrium prices of the two goods at date t D 0 are
normalized to .p ; 1 p /. Show that the price p corresponding to a
competitive (Arrow-Debreu) equilibrium at date t D 0 is given by
˛!1 1 e12 C ˛N 2 e22

p .!1 / D ; on f! D !1 g;
˛!1 1 e12 C .1 ˛!1 1 /e11 C ˛N 2 e22 C .1 ˛N 2 /e21
˛!1 2 e12 C ˛N 2 e22
p .!2 / D ; on f! D !2 g;
˛!1 2 e12 C .1 ˛!1 2 /e11 C ˛N 2 e22 C .1 ˛N 2 /e21
(8.44)
where ˛N 2 WD ˛!2 1 C .1 /˛!2 2 .

(ii) Deduce that, if p .!1 / D p .!2 /, then (8.44) corresponds to an Arrow-Debreu
equilibrium of the economy, where prices do not reveal any information. On
the contrary, if p .!1 / ¤ p .!2 /, then there cannot exist an Arrow-Debreu
equilibrium and the only candidate for an equilibrium of the economy is a
Green-Lucas equilibrium.
(iii) Show that, if the uninformed agent extracts information from the observation
of market prices, then the equilibrium prices associated to a Green-Lucas
equilibrium are characterized by
˛!1 1 e12 C ˛!2 1 e22

pQ .!1 / D ; on f! D !1 g;
˛!1 1 e12 C .1 ˛!1 1 /e11 C ˛!2 1 e22 C .1 ˛!2 1 /e21
˛!1 2 e12 C ˛!2 2 e22
pQ .!2 / D ; on f! D !2 g:
˛!1 2 e12 C .1 ˛!1 2 /e11 C ˛!2 2 e22 C .1 ˛!2 2 /e21
(8.45)
(iv) Deduce that, if pQ .!1 / ¤ pQ .!2 /, then the price functional obtained in part (iii)
corresponds indeed to a Green-Lucas equilibrium price functional. However, if
p .!1 / ¤ p .!2 / and pQ .!1 / D pQ .!2 /, then the economy admits no Green-
Lucas equilibrium.
Exercise 8.8 In the context of the model considered by Grossman [840] (as
presented in Sect. 8.2), prove that there exists a Green-Lucas equilibrium price
functional W Y ! RN which is a sufficient statistic if and only if there exists an
equilibrium price functional pa W Y ! RN of the artificial economy (where every
agent observes the full set of private signals) which is a sufficient statistic.
Exercise 8.9 This exercise is taken from Mas-Colell et al. [1310, Example 19.H.1].
Consider a one-period exchange economy with two agents i 2 f1; 2g, two possible
states of the world f!1 ; !2 g and two consumption goods. Suppose that !1 and
!2 have the same probability of occurrence, i.e., P.!1 / D P.!2 / D 1=2. The
endowments of the agents, expressed in terms of units of the two goods, are given
by e1 D .1; 0/ and e2 D .0; 1/, in correspondence of both states of the world. In
aggregate terms, there is one unit of each of the two goods in both states of the
world. The preferences of the two agents are characterized by the state dependent
utility function
q q
ui !; xi1 .!/; xi2 .!/ D ˇ.!/ xi1 .!/ C .1 ˇ.!// xi2 .!/; for ! 2 f!1 ; !2 g;
where ˇ.!1 / D 1 and ˇ.!2 / D 0, with xin .!/ denoting the demand by agent i
of good n in correspondence of state !. This specification means that agents do
not receive any utility from the second good in correspondence of the state of the
world !1 .
(i) In this economy, there are no security markets, but only a spot market (i.e.,
agents trade immediately before the realization of the state of nature). Suppose
first that there is no private information. Show that in equilibrium
p each agent
consumes 1=2 of each good and has an expected utility equal to 1=2.
(ii) Suppose now that the state of the world is perfectly revealed to the two agents
before the opening of the market. Show that in equilibrium each agent only
consumes his initial endowment, getting an utility of 1 in one state of the world
and 0 in the other state of the world.
(iii) Deduce that the revelation of the state of the world before the opening of the
market decreases the social welfare ( Hirshleifer effect, see Sect. 8.2).
Exercise 8.10 This exercise is inspired by Schlee [1504, Example 1]. Consider
an economy with I agents characterized by power utility functions ui .x/ D ˛1 x˛ ,
with common risk aversion parameter 0 ¤ ˛ < 1, and endowment .ei1 ; : : : ; eiS / 2
RSC , for all i D 1; : : : ; I. Assume that at t D 0 there are S markets for the
S Arrow securities paying in correspondence of the S states of the world, with
prices .p1 ; : : : ; pS /. Show that in correspondence of an equilibrium of this economy
the optimal expected utility of each agent is concave with respect to the vector
8.8 Exercises 473
of probabilities of the S states of the world. Deduce that additional information

will decrease the ex-ante expected utility of each agent as long as the individual
endowments are not proportional to the aggregate endowment, which we denote by
.e1 ; : : : ; eS /.
Exercise 8.11 In the setting of Proposition 8.10, show that at equilibrium the
optimal demand of every agent does not depend on the price of the risky asset.
Exercise 8.12 This exercise in inspired by Kreps [1134, Section 4]. Consider an
economy with two agents: an informed agent and an uninformed agent. The pref-
erences of both agents are characterized by negative exponential utility functions
with a common risk aversion parameter a. The possible states of the world are
partitioned into two events ˝1 and ˝2 , with P.˝1 / D P.˝2 / D 1=2. In the
economy, there is a single asset, which delivers at date t D 1 a random dividend
Q Moreover, agents can trade in t D 0 a future contract delivering one unit of the
d.
asset in t D 1. The price of the future contract at t D 0 is denoted by q and the
supply of the future contract is null. Conditionally on ˝1 , the dividend dQ at t D 1 is
normally distributed with mean 1 and variance 2 while, conditionally on ˝2 , it is
normally distributed with mean 2 and variance 2 , where 1 ¤ 2 . The first agent
is assumed to be informed, meaning that he observes a signal yQ taking the value 1
if ! 2 ˝1 and the value 2 if ! 2 ˝2 . The second agent receives no information
(uninformed agent). The uninformed agent receives an exogenous income at date
t D 1, defined in terms of units of the asset. More specifically, conditionally on
˝i , the uninformed agent receives ki 0 units of the asset, for i D 1; 2. In this
economy, an equilibrium price functional for the forward contract is represented by
a functional q D W f1; 2g ! R, which maps the realization of the signal yQ into
a price at t D 0 for the forward contract. The equilibrium price functional is fully
revealing if .1/ ¤ .2/.
Show that, if a D 1, 1 D 4, 2 D 5, k1 D 2, k2 D 4 and 2 D 1, then there
does not exist a fully revealing Green-Lucas equilibrium.
Exercise 8.13 Consider an economy as described at the beginning of Sect. 8.3,
under the additional assumptions that ai D a > 0 and "2i D 1, for all i D 1; : : : ; I.
Assume furthermore that each agent has a random initial endowment eQ i (in terms of
units of the risky asset), which is normally distributed with zero mean and variance
e2 , for all i D 1; : : : ; I. Show that this economy admits the partially revealing Green-
Lucas equilibrium price functional given in (8.23).
Exercise 8.14 (See Vives [1630], Proposition 4.1) Prove Proposition 8.12.
Exercise 8.15 In the setting of Proposition 8.12, show that the equilibrium price is
a weighted average of the agents’ conditional expectations of the future dividend
plus a noise term. Moreover, show that, if uninformed traders become risk neutral
Q
(i.e., aun ! 0), then the equilibrium price functional .d; u/ converges to EŒdjp.
Exercise 8.16 In the context of Proposition 8.14, prove that the function 7! ./
is strictly increasing and admits the representations given in (8.34).
Exercise 8.17 Prove part (iv) of Proposition 8.15.
Exercise 8.18 Prove parts (vii), (viii) and (ix) of Proposition 8.15.
Exercise 8.19 (The Equilibrium in the Hirshleifer et al. [947] Model) Consider
an economy with two trading dates t 2 f1; 2g and two traded assets: a riskless asset
with constant rate of return rf D 1 and a risky asset whose liquidation value after
date t D 2 is given by the random variable
fQ D fN C Q C "Q;
where EŒQ D EŒQ" D 0 and the couple .; Q "Q/ follows as a bivariate normal
distribution with independent components, with Var./ Q DW 2 and Var.Q"/ DW 2 .
"
The economy is populated by a continuum of agents indexed in the interval
Œ0; N. All agents are assumed to have negative exponential utility functions with
a common risk aversion parameter a. In this economy, there are early informed
and late informed agents. More specifically, a mass Œ0; M of the total population
Œ0; N of agents is early informed, meaning that at date t D 1 they can observe
the realization of the random variable . Q On the other hand, late informed agents
can only observe the realization of Q at the successive date t D 2. The error term "Q
remains unobservable to both classes of agents. Furthermore, every agent is assumed
to have an initial endowment of e0 units of the riskless asset.
Besides early and late informed agents, there are also liquidity traders, whose
net trades (demand shocks) at dates t D 1 and t D 2 are represented by the random
variables zQ1 and zQ2 , respectively. The random variables zQ1 and zQ2 are normally
distributed, with zero mean and variance Var.Qz1 / D Var.Qz2 / DW z2 and are mutually
independent as well as independent of .Q ; "Q/.
Finally, besides early and late informed agents and liquidity traders, there is a
group of competitive risk neutral market makers, who possess no information on
the fundamental value of the risky asset. Market makers are willing to absorb the
net demand of the other traders at competitive prices.
If we denote by G1 and G2 the information sets of the early informed agents
at date t D 1 and t D 2, respectively, then the information set of the late informed
agents at date t D 2 coincides with G2 , while the information set of the late informed
agents at t D 1 coincides with the information set of the risk neutral market makers
(which contains only the observation of the aggregate demand or, equivalently, the
equilibrium price functional). As usual, all the agents are assumed to rationally
extract information from the observation of equilibrium prices. Letting p1 and p2
denote the price of the risky asset at dates t D 1 and t D 2, respectively, we denote
by x1 .; p1 / and x2 .; p2 / the demand of an early informed agent at t D 1 and t D 2,
respectively, and by y1 .p1 / and y2 .; p2 / the demand of a late informed agent at
t D 1 and t D 2, respectively. Taking into account the effect of the liquidity traders,
the net aggregate demands are then given by
D1 .p1 / WD Mx1 .; p1 / C .N M/y1 .p1 / C zQ1

D2 .p2 / WD Mx2 .; p2 / C .N M/y2 .; p2 / C zQ1 C zQ2 ;
8.8 Exercises 475
at t D 1 and t D 2, respectively. Since market makers are risk neutral and

competitive, they set prices which are equal to the expectation of the terminal value
of the risky asset conditionally on their information set (which only consists of the
aggregate demand function of all other traders), so that
p1 D EŒ fQ jD1 ./ and p2 D EŒ fQ jD1 ./; D2 ./: (8.46)
(i) Let us conjecture that the equilibrium price functional is of the form
p2 D 2 . fN ; ; z1 ; z2 / D fN C ˛ C ˇz1 C z2
p1 D 1 . fN ; ; z1 / D fN C C z1 ;
where ˛; ˇ; ; ; are suitable coefficients to be determined. Show that in

correspondence of the equilibrium prices the optimal demands of the early and
late informed agents are given by
fN C p2
x2 .; p2 / D y2 .; p2 / D (8.47)
a"2
!
Q zQ1 ; zQ2 /jG1 p1
EŒ2 . fN ; ; 1 1
x1 .; p1 / D N Q
C 2
a Var.2 . f ; ; zQ1 ; zQ2 /jG1 / "
Q zQ1 ; zQ2 /jG1

fN C EŒ2 . fN ; ;
C (8.48)
a"2
y1 .p1 / D 0: (8.49)
(ii) As shown in Hirshleifer et al. [947, Lemma 1], the conjectured price functional
is indeed the equilibrium price functional and the coefficients ˛; ˇ; ; ; can
be explicitly determined. Moreover, it holds that 2 .0; 1/ and ˛ > . In this
case, show that

Q zQ 1 / fN ; Q > 0
Cov 1 . fN ; ; and Q zQ1 ; zQ 2 / . fN ; Q ; zQ 1 /; Q > 0:
Cov 2 . fN ; ; 1
(8.50)
Exercise 8.20 Consider the model proposed in Campbell et al. [346, Section III]
and discussed in Sect. 8.4. In this exercise, following Campbell et al. [346,
Appendix A], we compute the equilibrium price of the risky asset.
The economy is supposed to have an infinite time horizon (i.e., the trading dates
are t 2 N) and there are two traded assets: a riskless asset paying the constant rate
of return rf > 1 and a risky asset paying a random dividend dt at each trading
date t 2 N. For each t 2 N, the random variable dt is supposed to be of the form
dt D dN C dQ t , where dN represents the average dividend and EŒdQ t D 0, for all t 2 N.
The sequence of random variables fdQ t gt2N is assumed to follow the process
dQ t D ˛d dQ t1 C uQ t ; for all t 2 N;
where ˛d 2 Œ0; 1 and .Qut /t2N is a sequence of i.i.d. normally distributed random
variables with zero mean and constant variance u2 . The per capita supply of the
risky asset is fixed and normalized to one.
There are two classes A and B of agents. The preferences of both types of agents
are represented by negative exponential utility functions, where the risk aversion
parameter of class A agents is a constant a, while the risk aversion parameter of class
B agents is time-varying and denoted by bt , for t 2 N. Let denote the proportion
of type A agents in the economy.
At each trading date t 2 N, every agent observes a signal yQ t , which is supposed
to be of the form
yQ t D uQ tC1 "QtC1 ;
where .Qyt ; "QtC1 / are jointly normally distributed, for all t 2 N, with
EŒQ"t D EŒQyt D 0; Var.Q"t / D "2 ; Var.Qyt / D y2 and EŒQutC1 jQyt D yQ t :
For each t 2 N, define the following variable zt that can be interpreted as the risk
aversion of the marginal investor:
abt
zt WD ;
.1 /a C bt
and assume that zt D zN C zQt , for all t 2 N, where
zQt D ˛z zQt1 C Q t ; for all t 2 N;
where ˛z 2 Œ0; 1 and .Q t /t2N is a sequence of i.i.d. normally distributed random
variables with EŒQ t D 0 and Var.t / D 2 , for all t 2 N, independent of all other
random variables introduced so far.
Prove that the equilibrium price process .pt /t2N of the risky asset is given by
pt D ft dt C .0 C z zt /; (8.51)
where
.1 ˛z /z zN rf ˛z q
0 D <0 and z D 2
1 1 .2 = 2; / ;
rf 1 2
8.8 Exercises 477
and, for all t 2 N, ft denotes the cum-dividend fundamental value of the risky asset
in the hypothetical case of risk neutral agents and is given by
" #
dQ tCs ˇˇ Q
1
X rf dN rf Q 1
ft D E Qt D
s ˇd t ; y C dt C yQ t ;
sD0
rf rf 1 rf ˛d rf ˛d
under the assumption that 2 2; WD .rf ˛z /2 =.4f2 /, where
rf2 1
f2 WD 2 C 2
.rf ˛d /2 " .rf ˛d /2
denotes the innovation variance of ft (see Campbell et al. [346, Theorem 1]).
Chapter 9
Uncertainty, Rationality and Heterogeneity
If you consider men as infinitely egoistic and infinitely

farsighted. The first hypothesis can be accepted in first
approximation, but the second would maybe need a few reserves.
Poincaré to L. Walras (letter of October 1st, 1901)
Investment based on genuine long-term expectation is so

difficult as to be scarcely practicable. He who attempts it must
surely lead much more laborious days and run greater risks
than he who tries to guess better than the crowd how the crowd
will behave; and given equal intelligence, he may make more
disastrous mistakes.
Keynes (1936)
The results presented in the previous chapters rely on a set of classical assumptions
concerning the agents’ preferences, the economy and the modeling of randomness.
More specifically, we have assumed the validity of several fundamental hypotheses,
which can be summarized as follows:
• classical probability theory;
• expected utility theory;
• substantial rationality;
• rational expectations;
• homogeneous agents.
In this chapter, as well as in the following one, we will critically analyse
the role played by these assumptions and present several alternative approaches
which go beyond the classical paradigm. In our analysis, we adopt the perspective
illustrated by the two quotations by Solow and Kuhn at the beginning of Chap. 1.
We will consider the relation between hypotheses and results: each hypothesis
will be evaluated both for its own realism as well as for its implications in
generating theoretical results or in explaining empirical facts. Alternative behavioral
hypotheses will be evaluated not only on the basis of their capability to explain
the asset pricing anomalies illustrated in the previous chapters but also in terms of
their capability to provide an alternative paradigm compatible with what classical
asset pricing theory is already able to explain. Indeed, one of the main problems of
going beyond the classical framework of full rationality-expected utility theory is

480 9 Uncertainty, Rationality and Heterogeneity
represented by the fact that, typically, each alternative theory is only able to explain
a specific puzzle arising in the context of classical asset pricing theory but cannot
reproduce other facts, thus making new puzzles arise. Moreover, the robustness of
the results of each alternative theory should be carefully evaluated with respect to
different behavioral assumptions.
In this chapter, we shall focus our attention on the assumptions made on the
modeling of randomness, on the agents’ behavior and on the existence of market
imperfections. For the time being, we continue to make the standing assumption of
perfect competition, which will be relaxed in the following chapter.
This chapter is structured as follows. In Sect. 9.1, we discuss the fundamental
notions of probability, risk, uncertainty and information ambiguity, pointing out
their implications for the modeling of preferences. In Sect. 9.2, we present several
alternative preference functionals which generalize the classical time additive
expected utility representation, including behavioral models, recursive utilities,
habit formation preferences and alternative forms of discounting. In Sect. 9.3,
we relax the perfect rationality hypothesis and discuss models with noise traders,
overconfident traders, feedback traders and the role of market selection in the
survival of irrational agents. In Sect. 9.4, we provide an overview of the concepts of
bounded rationality, learning and distorted beliefs. Section 9.5 contains a discussion
of the impact of several market imperfections. At the end of the chapter, we provide
a guide to further readings as well as a series of exercises.
9.1 Probability, Risk and Uncertainty
In the analysis developed in the previous chapters, we have always assumed that
randomness is described through a probability space and that this probability
space is common knowledge among the agents. More specifically, we adopted the
classical axiomatic approach to probability theory and we assumed the existence of
a probability space .˝; ; P/, where the set ˝ is the set of all elementary states of
the world, is a -algebra on ˝ and P represents an objective probability measure
(see the introduction to Chap. 2). In this section, we will discuss the relevance of
this hypothesis and we present several alternative paradigms for the modeling of
randomness, also on the basis of experimental evidence.
Even though it is a very natural assumption, representing agents’ beliefs-opinions
by means of a probability space .˝; ; P/ does not come without consequences. In
particular, while the couple .˝; / seems to be a necessary tool to describe the set of
possible realizations of randomness, the assumption of the existence of an objective
probability measure P is more delicate. In particular, recall that, according to the
standard axiomatic approach to probability theory, the measure P is assumed to be
countably additive. Many philosophers, economists and mathematicians (including
Ramsey, Keynes, Knight, De Finetti, Savage amongst others) have pointed out
the limits of the classical axiomatic probability theory and proposed alternative
approaches.
9.1 Probability, Risk and Uncertainty 481
The objectivist approach to probability theory has a long tradition, dating back
to the origins of probability theory with the study of gambling games. In this
context, the likelihood of an event can be defined as the ratio of favorable outcomes
over all possible outcomes of a random experiment, with all elementary events
being assumed to have an equal probability of realization. This approach led to the
frequentist approach to probability theory, where the probability of occurrence of an
event is given by the limit of its relative frequency of occurrence. As in the classical
axiomatic probability theory, this interpretation refers to an objective probability
measure.
The frequentist approach to probability is particularly well suited to random
phenomena for which experiments can be repeated and reproduced in order to
assess the probability of an event, like gambling games. However, this is clearly
not the case in economic problems. In this direction, the subjectivist approach
seems more promising. According to this approach, the likelihood of an event is
given by the confidence that an agent places on its realization, thus providing an
estimate of the probability of occurrence without resorting to repeated experiments.
In particular, according to the subjectivist interpretation of probability, dating back
to contributions of De Finetti, Ramsey and Savage (we refer to Gilboa & Marinacci
[779] for a survey of their approaches), the probability of an event E 2 from
the point of view of an agent is given by the price that he considers fair for the
contract delivering one unit of a reference good whenever the event E occurs (see
De Finetti [538]). According to this interpretation, the probability is subjective since
it depends on the agent’s assessment of the likelihood of the event. The probability
of an event can therefore be associated with the certainty equivalent of a specific
gamble and, in the case of an elementary event, it corresponds to the price of an
Arrow security. De Finetti also required that the probability evaluation be coherent,
meaning that no combination of bets should allow for a surely positive gain. Note
that the classical and the subjectivist approaches to probability share some features:
indeed, in both approaches, it holds that P.˝/ D 1, P.;/ D 0 and the probability of
two incompatible events is equal to the sum of the probabilities of the two events.
However, according to De Finetti’s approach, the probability measure is only finitely
additive and is not assumed to be countably additive. This represents a significant
departure from the classical axiomatic approach to probability theory.
Expected utility theory is based on the existence of a probability measure known
ex-ante by the agents. As discussed in Chap. 2, this reflects the riskiness of the
economic environment, where agents do not know which state of the world will be
realized at a future date but have a complete knowledge of the underlying probability
distribution. Savage has extended expected utility theory to a setting with unknown
probabilities, i.e., a setting where agents do not know ex-ante the probabilities of
realization of the events (see Savage [1499]). According to Savage, even though
their beliefs cannot be represented in terms of a probability measure, agents still
have preferences over gambles and can formulate choices among different gambles.
Denoting by R the preference relation over gambles, if R satisfies suitable axioms,
then it is possible to infer a subjective probability distribution and a utility function
such that R can be represented by the expectation of this utility function with
respect to the subjective probability distribution (compare with Sect. 2.1). In this
setting, it is important to remark that the subjective probabilities are derived from
the agent’s preferences and, hence, the preference relation R embeds both an agent’s
preferences with respect to wealth under certainty as well as his beliefs. We refer
to Kreps [1136, Chapter 9] for a detailed analysis of the subjectivist approach to
decision theory.
According to Savage’s subjectivist approach, even though an agent does not know
ex-ante the probability distribution, the decision making problem can be analysed
ex-post in the context of expected utility theory, with the agent behaving as if the
probability distribution was known and where the utility function and the probability
distribution are implied by the preference relation R. However, this type of behavior
is not confirmed empirically. In particular, a deep criticism was expressed by Knight,
who distinguished between uncertainty and risk (or uncertainty in a strong sense and
in a weak sense, respectively), see Knight [1108]. As explained in the introduction
to Chap. 2, uncertainty arises when agents are unable to assign a probability to some
event, typically because of missing information (information ambiguity), and do not
behave (ex-post) according to the expected utility paradigm, with respect to some
objective or subjective probability measure.
The Ellsberg paradox illustrated in Ellsberg [636] provides an enlightening
example of a violation of the expected utility theory in the presence of uncertainty-
information ambiguity. Consider an agent who has to draw a ball from an urn
containing nine colored balls: three balls are red and each of the remaining six balls
is either yellow or black and the agent does not know how many of the six balls are
yellow. This example cannot be described in the classical setting considered in the
previous chapters. Indeed, while the agent knows the probability of extracting a red
ball (1=3), he does not know the probability of extracting a yellow (or a black) ball.
The probability of such an event is comprised between 0 and 2=3. This situation
provides an example of information ambiguity, since the agent does not know how
many balls in the urn are yellow and/or black.
Continuing the description of the Ellsberg paradox, let us consider four gambles
A, B, C and D, whose payoffs depend on the colour of the first ball drawn from
the urn (we write fRg, fYg and fBg to denote respectively the event where a red, a
yellow or a black ball is drawn):
• gamble A: payoff 1 if fRg, payoff 0 if fYg or fBg;
• gamble B: payoff 0 if fRg, payoff 1 if fBg and payoff 0 if fYg;
• gamble C: payoff 1 if fRg, payoff 0 if fBg and payoff 1 if fYg;
• gamble D: payoff 0 if fRg and payoff 1 if fYg or fBg.
Observe that the payoffs of gambles A and D are not affected by the information
ambiguity. On the contrary, the payoffs of gambles B and C are affected by
information ambiguity. When confronted with the above four gambles, most of
the agents will typically choose according to the following order: A P B and
D P C. However, this preference ordering is not compatible with the expected
utility theory à la Savage. As a matter of fact, if an agent’s preferences can be
represented by some expected utility function, then A P B implies that the agent
considers the probability of the event fBg less than 1=3 and the probability of
the event fYg bigger than 1=3. Instead, D P C implies that the agent estimates
the probability of the event fBg to be bigger than 1=3 and the probability of
the event fYg to be smaller than 1=3. This experimental result shows that the
expected utility theory, even in its subjectivist version à la Savage, is unable to
describe the agents’ behavior in a context of information ambiguity-uncertainty (see
Camerer & Weber [329] for a survey on the experimental evidence on decision
making under information ambiguity). Typically, it is shown that agents are averse
towards gambles characterized by payoffs with uncertain probabilities, meaning that
agents dislike information ambiguity (see Epstein [640] and Ghirardato & Marinacci
[774]).
This empirical evidence motivated the development of new decision theories
aiming at explaining the agents’ behavior under information ambiguity (see again
Camerer & Weber [329] for a survey). In this section, we briefly present two
approaches, both based on an axiomatic description of agents’ preferences: the
maxmin expected utility theory and the Choquet expected utility theory. By relaxing
the independence axiom (see Assumption 2.3), these two approaches incorporate the
distinction between risk and uncertainty and, in the special case of no information
ambiguity, they reduce to the classical expected utility theory as presented in
Chap. 2. In a nutshell, the maxmin expected utility theory is based on the idea
that agents formulate their decisions with respect to a set of probability measures
compatible with the available information, while the Choquet expected utility
theory relies on non-additive probabilities. On the axiomatization of decision theory
under information ambiguity see also Chateauneuf et al. [413], Epstein & Zhang
[648], Maccheroni et al. [1262, 1263], Seo [1516], Klibanoff et al. [1105].
In the seminal paper Gilboa & Schmeidler [780], the maxmin expected utility
(or expected utility in the worst-case scenario) has been proposed to represent
preferences under information ambiguity. In order to illustrate the concept, let
us consider an economy with a finite number S 2 N of elementary states of
the world. Under information ambiguity, there does not exist a single probability
measure commonly known to all the agents, but rather a set of probability measures
which are compatible with the available information. For instance, in the case
of the Ellsberg paradox, there are seven probability measures compatible with
the available information, each one corresponding to a specific number of yellow
(or black) balls in the urn. We denote the set of these probability P measures by
˘ D f 1 ; 2 ; : : : ; K g, for some K 2 N, where k 2 RSC and SsD1 sk D 1, for all
k D 1; : : : ; K. Similarly as in (2.1), we represent a gamble as a measurable mapping
xQ W ˝ ! RC , with x.!s / denoting the realization of the gamble in correspondence
of !s , for s D 1; : : : ; S. Following Schmeidler [1505], the preference relation R is
said to exhibit uncertainty aversion if, for any three gambles xQ 1 , xQ 2 and xQ 3 ,
xQ 1 R xQ 3 and xQ 2 R xQ 3 ) ˛Qx1 C .1 ˛/Qx2 R xQ 3 ; for all ˛ 2 Œ0; 1:
Assuming that the preference relation R satisfies suitable axioms and that the agent
is uncertainty averse (see Gilboa & Schmeidler [780, Axiom A.5]), there exists a
utility function u W RC ! R such that, for every two gambles xQ 1 and xQ 2 , it holds that
xQ 1 R xQ 2 ” min E Œu.Qx1 / min E Œu.Qx2 /; (9.1)

2˘ 2˘
where, for each 2 ˘ , we denote by E Œ the expectation with respect to the
probability distribution (see Gilboa & Schmeidler [780, Theorem 1]). Note that,
in Gilboa & Schmeidler [780, Theorem 1], the probability space is general (i.e., it
is not necessarily finite) and the existence of a non-empty, closed and convex set ˘
of (finitely additive) probability distributions such that (9.1) holds is actually part
of the result. Moreover, similarly to the result of Theorem 2.1, the utility function
u is unique up to positive linear transformations. Relation (9.1) captures the fact
that the knowledge of the probability structure is incomplete and agents are averse
to uncertainty-information ambiguity. This corresponds to evaluating a gamble
with respect to the probability distribution associated with the lowest possible
level of expected utility (worst-case scenario), among all probability distributions
compatible with the available information. As a consequence, if X denotes a given
set of gambles, an agent will optimally choose a gamble xQ 2 X by solving the
problem
max min E Œu.Qx/;

xQ2X 2˘
where u is the utility function appearing in the representation (9.1) of the preference
relation R. In particular, it can be shown that the maxmin expected utility is
consistent with the Ellsberg paradox.
An axiomatization of subjective expected utility theory different from the
maxmin expected utility has been formulated in Gilboa [778] and Schmeidler
[1505] by relying on non-additive probabilities (capacities). The expected utility
with respect to a non-additive probability can be computed by using the Choquet
integral, making the expected utility criterion applicable to cases where classical
expected utility theory is not applicable. By weakening the independence axiom
(and adopting the notion of comonotonic independence, see Schmeidler [1505]),
it is shown that preferences can be represented through the expected utility where
the expectation is computed with respect to a monotonic non-additive subjective
probability (Choquet expected utility). The fact that the probability is monotonic
and non-additive means that, for any two events A; B 2 such that A B, it holds
that .A/ .B/, together with the basic properties .˝/ D 1 and .;/ D 0.
Note that, for a non-additive probability (also known as capacity), it holds that
.A/ C .B/ ¤ .A [ B/ C .A \ B/; (9.2)
for two events A; B 2 . In particular, a non-additive probability is said to be

convex if the left-hand side of (9.2) is less or equal than the right-hand side. In
particular, is convex if and only if an agent is uncertainty averse (see Schmeidler
[1505]). The inequality reduces to an equality when the agent faces a standard risky
situation, as considered in Chap. 2. Note that does not represent an objective

measure, but only allows for a representation of an agent’s preferences in the form of
non-additive expected utility. Similarly as above, the utility function is unique up to
linear transformations. As in the case of the maximin expected utility, the approach
based on the Choquet expected utility is consistent with the Ellsberg paradox. Note
also that the Choquet expected utility theory has a partial overlap with the maxmin
expected utility theory. Indeed, when the non-additive probability is convex, the
Choquet expected utility can be described in terms of a suitable maxmin expected
utility (see Gilboa & Schmeidler [780]).
Information Ambiguity: Portfolio Choices, Equilibrium and

Asset Pricing Implications
The literature on the implications of information ambiguity on the modeling of

financial markets is quite large. In the remaining part of this section, we limit
ourselves to a brief overview of some important results.
An interesting effect of information ambiguity on the optimal demand of a
risky asset has been shown in Dow & Werlang [588]. The authors consider a
two-period economy (i.e., t 2 f0; 1g) and an agent characterized by a utility
function u such that u0 > 0 and u00 0 having the possibility of investing in
a risky asset delivering the random dividend dQ at date t D 1. The risk free rate
is normalized to one. In the classical setting, i.e., when the probability measure
in the expected utility representation of the agent’s preferences is additive (no
information ambiguity), then the agent invests a positive amount of wealth in the
risky asset if and only if the expected dividend is higher than the current price (see
Proposition 3.1) and the optimal demand decreases with respect to the agent’s risk
aversion (see Proposition 3.2). On the contrary, if the agent’s preferences imply
a non-additive probability measure and the agent is uncertainty averse, then these
results do not hold anymore. More precisely, we have the following proposition,
which corresponds to Dow & Werlang [588, Theorem 4.2].
Proposition 9.1 Consider a risk averse or risk neutral agent (with a constant initial
wealth w0 > 0) who is uncertainty averse and has the possibility of investing in a
risky asset delivering the random dividend dQ for a price p > 0. Denote by w the
agent’s optimal demand of the risky asset. Then the following hold:
(i) w > 0 if p < EŒdQ and only if p EŒdQ ;
(ii) w < 0 if p > EŒdQ and only if p EŒdQ ,
where EŒ denotes the expectation with respect to a non-additive probability.
Proof We only provide an outline of the proof, referring to Dow & Werlang
[588] for full details. In the present context (as in Schmeidler [1505]), the agent’s
preference relation can be represented in expected utility terms with respect
to a non-additive probability. Moreover, as discussed in Dow & Werlang [588,

Appendix], the expectation defined with respect to a non-additive probability is
such that Jensen’s inequality holds and the expectation of an affine transformation
of a random variable is given by the affine transformation of the expectation of the
random variable. The claim then follows by arguments similar to those used in the
proof of Proposition 3.1. t
u
As a consequence of the above proposition, there exists an interval of prices,
given by .EŒdQ ; EŒdQ /, in correspondence of which the optimal demand of a risk
averse or risk neutral agent is null. Note that, in the axiomatic approach proposed by
Schmeidler [1505], it holds that EŒdQ < EŒdQ . This result has been extended to
the case of multiple beliefs characterized by a set of normal distributions, see Easley
& O’Hara [621]: in this case, ambiguity averse agents do not trade in the market for
a set of prices. Microstructure models showing that illiquidity in the market may
arise from ambiguity aversion have been analysed in Easley & O’Hara [622, 623]
and Routledge & Zin [1475], while the implications of ambiguity on portfolio
and equilibrium analysis in a mean-variance framework have been analysed in
Maccheroni et al. [1264], Garlappi et al. [759], Boyle et al. [276].
The above argument has been further investigated in Mukerji & Tallon [1361],
where the authors show that uncertainty aversion may lead to market incompleteness
(in a related context, see also Rinaldi [1454]). Mukerji & Tallon [1361] adopt the
Choquet expected utility framework of Schmeidler [1505], where the agents’ sub-
jective beliefs are represented by a convex non-additive probability, thus introducing
uncertainty aversion in the economy. In a complete market economy, in the presence
of a severe information ambiguity, the agents can decide not to trade financial
assets which are affected by idiosyncratic risk (in the sense that their payoffs do
not only depend on realized endowments). Therefore, uncertainty aversion reduces
the risk sharing opportunities offered by the financial market, with the consequence
that agents cannot fully diversify the idiosyncratic risk. In turn, this leads to an
equilibrium allocation that does not correspond to the allocation obtained in a
complete market economy. Note that this no-trade result holds under information
ambiguity as long as the initial allocation is Pareto optimal. This result can also
explain the limited use of indexed debt (i.e., loan contracts indexed to the inflation
rate) in real financial markets: if agents are uncertainty averse, then under some
conditions there is no trade in indexed bonds in equilibrium (see Mukerji & Tallon
[1362]).
Pareto optimal allocations and sharing rules under uncertainty aversion have been
investigated in Chateauneuf et al. [412]. When agents believe in the same convex set
of probability distributions, then the set of Pareto optimal allocations is independent
of that set and coincides with the set of Pareto optimal allocations obtained in
a classical economy where agents have a standard expected utility function and
homogeneous beliefs. Pareto optimal allocations depend only on aggregate risk
and are comonotonic (compare with Theorem 4.3). Moreover, in the absence of
aggregate risk, Chateauneuf et al. [412] provide sufficient conditions in order to
obtain an allocation with full insurance. In Epstein [641], considering a simple
two-period economy with heterogeneous agents and ambiguity, it is shown that

agents’ consumption levels are only positively correlated with aggregate shocks, but
are not necessarily perfectly correlated, unlike in the classical setting considered in
the previous chapters (see Epstein & Miao [643] for a continuous time extension
of this result). Dana [513], Rigotti & Shannon [1452], Rigotti et al. [1453] provide
general characterizations of equilibrium and Pareto optimality, also analysing the
determinacy of equilibria.
In Epstein & Wang [646, 647] the maxmin expected utility approach of Gilboa
& Schmeidler [780] has been applied to a multi-period economy in an infinite time
horizon by relying on the recursive utility framework (see Sect. 9.2 below). In a
dynamic setting with information ambiguity, in correspondence of an event At 2 Ft
an agent does not believe in a unique conditional probability distribution for the
events belonging to FtC1 , but rather in a set of conditional probability distributions
forming a probability kernel correspondence. In particular, this reflects both the
presence of uncertainty and the agents’ aversion to uncertainty. In this discrete time
setting, the utility function is represented by a utility process solving a recursive
relation (the existence of a utility process is established in Epstein & Wang [646,
Theorem 1]). Referring to the next section for a presentation of the recursive utility
framework, we just mention here that, in the setting considered in Epstein & Wang
[646], the utility is computed at each date t 2 N by taking into account the certainty
equivalent of the future utility, which is computed by taking the minimum among
all certainty equivalents associated to the probability distributions belonging to the
kernel. In a pure exchange economy with a representative agent, the existence
of an equilibrium is established and asset prices are characterized by means of
an Euler inequality. Equilibria may be indeterminate and non-unique. Necessary
and sufficient conditions for equilibrium indeterminacy can be found in Epstein
& Wang [646]. More precisely, there may exist uncountably many equilibria, with
the consequence that equilibrium asset prices are not uniquely defined but rather
only belong to an interval. Equilibrium prices may be discontinuous with respect
to the previous realizations, thus inducing high volatility. Moreover, in Epstein &
Wang [647] the presence of uncertainty aversion is linked to the possibility of abrupt
changes in asset prices and may be incompatible with the existence of a risk neutral
representation of asset prices.
A continuous time utility functional incorporating information ambiguity with
multiple probability distributions has been proposed in Chen & Epstein [426]. Under
information ambiguity, it is shown that the equity premium is given by the sum of
a risk premium and an ambiguity premium. Epstein & Schneider [645] show that
excess returns are higher when future information quality is more uncertain (see also
Kogan & Wang [1117] for a two-factor CAPM extension). It is important to remark
that the presence of an ambiguity premium represents an interesting perspective to
address the equity premium and interest rate puzzles, the home bias puzzle and the
participation puzzle, see Epstein & Miao [643], Easley & O’Hara [621], Cao et al.
[363], Maenhout [1284], Trojani & Vanini [1600], Leippold et al. [1179], Uppal &
Wang [1605], Ju & Miao [1053], Barillas et al. [162], Gollier [802]. In general, the
demand of the risky asset under information ambiguity is lower than in the classical
setting (see also Epstein & Schneider [644] for an analysis of portfolio choices under
information ambiguity and learning).
Asset pricing results under information ambiguity have been obtained by relying
on robust-risk sensitive control theory in Anderson et al. [63], Hansen & Sargent
[893]. More precisely, in Hansen & Sargent [893] the authors refer to model
uncertainty rather than ambiguity and consider a continuous time perturbed Markov
process in an infinite horizon. Each perturbation corresponds to an absolutely con-
tinuous measure and represents a specification error on the model. The optimization
criterion is formulated by taking the infimum over all such measures (worst case
scenario), in line with the maxmin approach of Gilboa & Schmeidler [780] (on this
point see also Maccheroni et al. [1262]). Agents facing model uncertainty are averse
to specification errors and make decisions that are robust to this type of uncertainty.
Risk sensitivity and preference for robustness contribute (positively) to generate
a high risk premium together with a positive model misspecification premium and
induce a precautionary saving motive, see Hansen et al. [894] and Maenhout [1284].
Portfolio choices with ambiguity averse investors and (partly) predictable returns
have been analysed in Ait-Sahalia & Brandt [31] (see also Chen et al. [415] and Liu
[1223]), showing that the stock demand is decreasing with respect to the ambiguity
aversion, similarly as in the classical case of risk aversion (in this context, see also
Epstein & Schneider [644]).
9.2 On Expected Utility Theory
In the previous chapters, we have always assumed that an agent’s preference relation
can be represented by the expectation of an utility function. In particular, as shown in
Sect. 2.1, if an agent’s preference relation satisfies a set of axioms, then there exists
a utility function and a probability distribution (objective or subjective, as in the
Savage framework) such that the corresponding expected utility induces the same
ordering of the preference relation. In this section, we will discuss the relevance of
the set of axioms introduced in Sect. 2.1 and present several alternative approaches
that have been proposed in the literature in order to overcome the limitations of
classical expected utility theory. Moreover, we will also discuss the structure of
the utility functional (6.6) in a multi-period economy and the assumptions of time
additivity and of constant preferences over time.
As before, we denote by R the preference relation of an agent. The rationality
axiom introduced in Assumption 2.1 implies that R is complete, reflexive and
transitive. The completeness of R means that, for every couple of gambles, an
agent is always able to express a preference between the two gambles (either a strict
preference or an indifference relation), while the transitivity property implies that
the agent is able to rank different gambles according to his preference criterion.
The continuity axiom (Assumption 2.2) mainly plays a technical role: the main
inconvenient of this hypothesis is that some preference relations do not satisfy
it (e.g., lexicographic preferences). The independence axiom (Assumption 2.3)
9.2 On Expected Utility Theory 489
represents probably the most crucial and controversial assumption. Indeed, the
independence axiom requires that the preference relation between two arbitrary
gambles is not affected if each of the two gambles is mixed (with the same weight)
with a third arbitrary gamble. In other words, an agent whose preferences satisfy
the independence axiom evaluates two gambles by only taking into account what is
different among the two gambles. As can be seen from the proof of Theorem 2.1, the
independence axiom implies that the functional representing the agent’s preferences
is linear with respect to the probabilities.
The validity of the independence axiom has been deeply discussed in the decision
theory literature, the driving force of the debate being experimental (laboratory)
evidence showing that agents do not behave according to the independence axiom.
In this context, the most famous example violating the independence axiom is the
Allais paradox presented in Sect. 2.5. Starting from the choices over two couples
of gambles and applying repeatedly the independence axiom, we have shown that
agents’ choices are inconsistent with a representation of preferences by means of
the expectation of a utility function. A second set of experiments contradicting the
expected utility paradigm has been illustrated in Kahneman & Tversky [1060]. For
instance, Problems 3 and 4 in Kahneman & Tversky [1060] consider the case of
an agent who has to choose between a lottery offering a 25% chance of winning
3:000 and a lottery offering a 20% chance of winning 4:000. It is empirically
observed that 65% of the agents will choose the second lottery. On the contrary,
if agents have to choose between a 100% chance of winning 3:000 and an 80%
chance of winning 4:000, then 80% of the agents will choose the first lottery. In
the context of this example, according to classical expected utility theory, agents
should not choose differently in the two situations, since the probabilities associated
by the two lotteries to the two scenarios have been only multiplied by the same
factor. This distortion has been called certainty effect in Kahneman & Tversky
[1060], i.e., agents prefer certain outcomes over random outcomes and overweight
outcomes considered certain, with the consequences that preferences are in general
non-linear with respect to the probabilities. There are by now many experimental
paradoxes of this type: see Machina [1267, 1268], Camerer [328], Hens & Rieger
[938], Starmer [1564] for some surveys of the most significant ones. However, it
has to be noted that most of these experiments (including the Allais paradox) are
obtained by considering gambles with extreme probabilities (probabilities close to
zero or one). On the contrary, the application of expected utility theory seems to be
more appropriate when considering gambles without extreme probabilities.
In the decision theory literature, the independence axiom has been widely
discussed and several weaker formulations have been proposed, mainly by intro-
ducing non-linearities (with respect to the probabilities) in the expected utility
through different preference axiomatizations. Quiggin [1436] has introduced the
concept of rank dependent expected utility, i.e., a utility functional defined as the
expectation of a utility function with respect to a transformation of the probability
distribution based on the rank of the outcome. This allows to relax the validity
of the independence axiom, preserving at the same time the completeness, the
reflexivity and the transitivity of the preference relation. In Machina [1266], the
independence axiom has been replaced by a smoothness condition of the utility

functional (i.e., the utility functional is assumed to be Fréchet differentiable on
the space of probability distribution functions) and it is shown that most of the
basic tools and results of classical expected utility theory can be derived under
this weaker assumption, notably first and second order stochastic dominance, risk
aversion and the Arrow-Pratt risk aversion comparison (compare with Chap. 2).
Moreover, the utility functional introduced in Machina [1266] is consistent with
Allais-type paradoxes and other experimental evidence (in a related context, see
also Chew et al. [428]).
In a seminal paper on behavioral finance, Kahneman & Tversky [1060] have pro-
posed a utility functional that is additive but non-linear with respect to probabilities,
such that small probabilities are overweighted and, conversely, large probabilities
are underweighted (prospect theory). The utility functional is defined over changes
(gains/losses) with respect to a reference point representing the status quo and is
steeper for losses than for gains, concave for gains and convex for losses. This
type of utility functional is consistent with the certainty effect and captures several
behavioral aspects of decision making, like loss aversion and different risk attitudes
in the case of losses or gains. Prospect theory preferences will be discussed in more
detail below in this section.
In Chap. 6, we have assumed that an agent’s preferences can be represented
by a time additive expected utility function, with the implicit assumption that
the utility at each date t only depends on consumption at date t and not on
consumption at different dates. Moreover, we have typically assumed that the utility
functions associated to consumption at different dates coincide, thus implying that
the agent’s preferences do not change over time. These assumptions, however, have
a weak behavioral justification (see Machina [1268]). In a multi-period setting,
besides the inherent limitations of classical expected utility theory discussed above,
a major problem is represented by the fact that, assuming a utility functional
of the form (6.6), the function u determines both the degree of risk aversion
and the degree of intertemporal substitution of consumption. For instance, if the
function u is nearly linear, then the risk aversion is very low and, at the same
time, the rate of intertemporal substitution is almost insensitive to consumption
levels (large elasticity of intertemporal substitution of consumption), indicating a
large propensity to substitute consumption intertemporally. In the special case of a
power utility function, the relative risk aversion coefficient is exactly the inverse
of the elasticity of the intertemporal rate of substitution (see also Sect. 7.3 and
Exercise 7.8). Therefore, a utility functional of the form (6.6) can only capture
high risk aversion together with low intertemporal substitution or low risk aversion
together with high intertemporal substitution. It turns out that there is no behavioral
motivation for a connection of this type between risk aversion and elasticity of
intertemporal substitution of consumption. Indeed, risk aversion represents the
willingness to diversify over different states of the world, while intertemporal
substitution represents the willingness to diversify over different periods in time.
Experimental results in Barsky et al. [169] have shown that there is no relation
between risk aversion and the intertemporal rate of substitution.
In the literature, two main approaches have been proposed for relaxing this
relation between risk aversion and intertemporal substitution. A first approach is
based on recursive preferences, where the utility at every date is also affected by
the expected utility from future consumption. The second approach is represented
by habit formation utilities, where the utility also depends on past consumption or
other variables. These two approaches will be discussed in more detail below.
Prospect Theory, Loss Aversion and Disappointment Aversion
As put forward in Kahneman & Tversky [1060], the loss aversion phenomenon
refers to the tendency of individuals to be more sensitive to reductions/losses in
their levels of consumption or wealth than to increases/gains (for experimental
evidence on loss aversion see Bateman et al. [180], Abdellaoui et al. [2] and Levy
& Levy [1206]). In the context of prospect theory, this asymmetry has been taken
into account by considering utility functions defined over changes (gains and losses)
with respect to a reference point representing the status quo. The fact that utilities
are defined with respect to a reference point also captures the so-called framing
effect, meaning that individuals tend to compare alternatives to a given frame or
reference point. The reference point can be defined with respect to consumption,
wealth, capital gains, etc. (on reference-dependent preferences see also Koszegi &
Rabin [1121], where the reference point is determined endogenously by rational
expectations). The utility function is assumed to be steeper for losses than for gains
and, typically, it is concave in the region of gains and convex in the region of losses.
This last feature captures a risk averse behavior with respect to gains and a risk
seeking behavior with respect to losses.
According to prospect theory, the utility functional U./ representing preferences
over gambles xQ D fx1 ; : : : ; xS I 1 ; : : : ; S g, with xs being a gain/loss with respect to
some reference point, for s D 1; : : : ; S, is provided by
X
S
U.Qx/ D u.xs /w.s /; (9.3)
sD1
where u./ is the value function and w./ the weighting function. The value function
u./ W R ! R is assumed to be continuous and increasing, strictly concave for gains
and strictly convex for losses and, at zero, it is steeper for losses than for gains. The
weighting function w./ W Œ0; 1 ! Œ0; 1 is continuous and increasing, with w.0/ D
0 and w.1/ D 1, satisfies w./P > for small values of > 0 and w./ < for
large values of < 1 and SsD1 w.s / D 1. The value function plays an analogous
role to the classical utility function in the expected utility representation (2.2), while
the weighting function w transforms probabilities into subjective probabilities. We
call the functional U appearing in (9.3) the prospect utility functional.
A drawback of prospect theory preference functionals of the form (9.3) involves

potential violations of first order stochastic dominance (see Sect. 2.3). To see this,
as explained in Kahneman & Tversky [1060], it suffices to consider a probability
space with two possible states of the world and two gambles xQ 1 and xQ 2 given by
xQ 1 D fx; yI ; 1 g and xQ 2 D fx; yI 0 ; 1 0 g;
with 0 < y < x and 1 > > 0 > 0. Clearly, xQ 1 dominates xQ 2 according to the first
order stochastic dominance criterion (compare with Proposition 2.8). For a given
value function u./ and a weighting function w./, xQ 1 is preferred to xQ 2 if and only if
w./u.x/ C w.1 /u.y/ w. 0 /u.x/ C w.1 0 /u.y/;
i.e., if and only if
w./ w. 0 / u.y/

0
:
w.1 / w.1 / u.x/
Since u./ is assumed to be continuous, the right-hand side of the last inequality
converges to 1 as y ! x. Hence, w./ w. 0 / approaches w.1 0 / w.1 / as
y converges to x. Since C .1 / D 0 C .1 0 / D 1, this means that the above
inequality can be satisfied only if the weighting function w./ is essentially linear.
In this way, for suitable non-linear weighting functions, one can derive explicit
examples violating first order stochastic dominance.
A suitable notion of stochastic dominance has been formulated in the context
of prospect theory by Levy & Wiener [1208] introducing the prospect stochastic
dominance criterion. Similarly as in Sect. 2.3, let us consider gambles xQ defined on
a general (not necessarily discrete) probability space taking values in the bounded
interval Œ1; 1 (recall that the realizations of xQ represent changes with respect to a
status quo).
Definition 9.2 Let xQ 1 and xQ 2 be two gambles taking values in Œ1; 1 and let w./
be a weighting function. We say that xQ 1 dominates xQ 2 according to the prospect
stochastic dominance criterion, i.e., xQ 1
PSD xQ 2 , if U.Qx1 / U.Qx2 / holds for every
continuous and increasing value function u./ such that u00 .x/ 0 for x 0 and
u00 .x/ 0 for x 0, where the preference functional U./ is defined as in (9.3).
Similarly to the results presented in Sect. 2.3, the prospect stochastic dominance
criterion admits a characterization in terms of the distribution functions of two
random variables, as shown in the following proposition. The distribution functions
can be defined with respect to either the objective probability measure (i.e., the
probability measure represented by .1 ; : : : ; S /) or the subjective probability (i.e.,
the probability transformed by the weighting function w./). We denote by Fi ./ the
cumulative distribution function of the random variable xQ i , for i D 1; 2.
(i) RxQ 1
PSD xQ 2 ;
x
(ii) y .F1 .z/ F2 .z//dz 0, for all 1 y 0 x 1.
Moreover, if xQ 1
FSD xQ 2 , then it also holds that xQ 1
PSD xQ 2 .
Proof The proof of the equivalence between properties (i)–(ii) is given in Exer-
cise 9.1 and follows arguments similar to those used in the proof of Proposition 2.10.
In view of Proposition 2.8, xQ 1
FSD xQ 2 holds if and only if F1 .x/ F2 .x/, for
every x 2 Œ1; 1. The last claim of the proposition then follows directly from the
implication (ii))(i) above. t
u
In Proposition 9.3, the weighting function (which in turn determines the distribu-
tion functions F1 and F2 of the two random variables xQ 1 and xQ 2 ) is given. However,
Levy & Wiener [1208, Theorem 8] shows that prospect stochastic dominance is
stable with respect to all monotonic transformations of the distribution function that
are concave for gains and convex for losses.
Portfolio choices with loss averse preferences have been analysed in several
papers (see, e.g., Jarrow & Zhao [1022] for an analysis of optimal portfolios
considering the lower partial moment as a measure of downside risk). In particular,
despite the fact that prospect theory is in contrast with the foundations of mean-
variance theory (see Sect. 2.4), it has been shown in Levy & Levy [1207] that, when
diversification is allowed, the set of mean-variance efficient and prospect theory
efficient portfolios (i.e., portfolios that are not dominated by any other portfolio
according to the prospect theory stochastic dominance criterion) almost coincide.
More precisely, we have the following proposition, which corresponds to Levy &
Levy [1207, Theorem 2].
Proposition 9.4 Let us consider a financial market with normally distributed
returns and such that there does not exist a couple of assets with perfectly
correlated returns. Then, in the absence of portfolio restrictions and for any
prospect utility functional that does not violate first order stochastic dominance, the
set of prospect theory efficient portfolios is contained in the set of mean-variance
efficient portfolios.
Proof To prove the claim, it suffices to prove that any mean-variance inefficient
portfolio is also prospect theory inefficient. Let rQ denote the return of an arbitrary
mean-variance inefficient portfolio. This means that there exists another portfolio
such that its return rQ 0 satisfies 2 .Qr/ D 2 .Qr0 / and EŒQr0 > EŒQr . Due to the
assumption of normal distribution, this can be seen to imply that rQ0
FSD rQ
under the objective probability distribution. By assumption, rQ 0
FSD rQ also holds
with respect to the subjective probability distribution. The result then follows by
Proposition 9.3. t
u
In the above proposition, the prospect stochastic dominance criterion can refer to
subjective probabilities (probabilities suitably transformed by a weighting function).
The result can be refined in the case where the objective probabilities are employed
(see Levy & Levy [1207, Theorem 1]), with a more precise description of the set
of mean-variance efficient portfolios that do not belong to the set of prospect theory
efficient portfolios.
In order to solve the inconsistency between (classical) stochastic dominance and
prospect theory, Tversky & Kahneman [1603] introduce cumulative prospect theory.
Consider a gamble xQ D fx1 ; : : : ; xS I 1 ; : : : ; S g, with outcomes x1 < x2 < : : : < xS .
According to cumulative prospect theory, the preference functional (9.3) is replaced
by the following functional
!
X
S X
s X
s1
U.Qx/ D u.xs / w k w k ; (9.4)
sD1 kD1 kD1
where u./ still plays

P the role of the value function and w./ of the weighting
function. Note that skD1 k represents the cumulative probability distribution of
the outcome xs , for s D 1; : : : ; S. According to cumulative prospect theory, only
probabilities associated to extreme outcomes are overweighted, while probabilities
associated to outcomes in the middle of the distribution are underweighted. As an
example, the following specification has been proposed in Tversky & Kahneman
[1603]:
u.x/ D x˛ for x 0 and u.x/ D .x/ˇ for x 0;
with ˛; ˇ 2 .0; 1/ and > 1 and w./ D . C .1 / /1= . As shown

in Exercise 9.2, cumulative prospect theory is compatible with first order stochastic
dominance. Moreover, unlike prospect theory in the form (9.3), cumulative prospect
theory can be easily extended to gambles on general (not necessarily discrete)
probability spaces.
Adopting the cumulative prospect theory of Tversky & Kahneman [1603],
Benartzi & Thaler [191] have shown that a utility functional taking into account
loss aversion allows for a solution to the equity premium and the risk free rate
puzzles. More specifically, Benartzi & Thaler [191] assume a short evaluation
period together with loss aversion. The combined effect of loss aversion and a short
evaluation period has been named myopic loss aversion. It is shown that a one-year
evaluation period allows to explain the equity premium puzzle (see also Thaler et al.
[1588], Gneezy & Potters [792], Gneezy et al. [791] for experimental evidence on
myopic loss aversion and on the interplay between risk aversion and the length of
the evaluation period). In this context, the agents’ behavior shows a framing effect,
in the sense that agents are sensitive to changes in financial wealth with respect to
some status quo, as discussed above.
In Barberis et al. [150] and Barberis & Huang [148], several asset pricing
anomalies are addressed by assuming loss aversion with respect to changes in
financial wealth, with a loss after prior gains being less painful than a loss after
previous losses. More specifically, Barberis et al. [150] assume a multi-period
economy in an infinite time horizon, with a single risky asset (in unit supply) paying
a dividend stream .dt /t2N satisfying
log dtC1 D log dt C gd C d "tC1 ; for all t 2 N; (9.5)
where ."t /t2N is a sequence of i.i.d. standard normal random variables, gd represents
an average logarithmic growth rate and d a volatility coefficient. Moreover, the
market also contains a risk free asset (in zero net supply) paying the constant return
rf (a priori allowed to be time-varying in Barberis et al. [150]). The aggregate per
capita consumption process .Nct /t2N is modeled as a separate process (i.e., it does not
coincide with the dividend process) satisfying the following dynamics:
log cN tC1 D log cN t C gc C c tC1 ; for all t 2 N; (9.6)
where .t /t2N is a sequence of i.i.d. standard normal random variables such that
Cov.t ; "t / D , for all t 2 N. The difference between the dividend and the
aggregate consumption process can arise from the fact that agents have other sources
of income besides dividends. Indeed, it is assumed that each agent in the economy
receives an income stream .yt /t2N such that cN t D dt C yt , for all t 2 N, with
..dt ; yt //t2N being a joint Markov process.
Investors’ preferences are described by the following functional, defined with
respect to self-financing trading-consumption strategies .w; c/ D ..wt ; ct //t2N ,
where wt denotes the amount of wealth invested in the risky asset in the period
Œt 1; t:
"1 #
X
U.c; w/ WD E ı u.ct / C bt ı v.XtC1 ; wtC1 ; zt / ;
t tC1
(9.7)
tD0
where u.x/ D x1 =.1 /, with > 0 being the coefficient of relative risk
aversion, and where ı 2 .0; 1/ is the discount factor. The processes appearing in
(9.7) have the following interpretation:
• .ct /t2N denotes the consumption process;
• .wt /t2N represents the trading strategy parameterized in terms of the amount of
wealth invested in the risky asset between date t 1 and date t (recall also that
.wt /t2N is a predictable stochastic process);
• .Xt /t2N is a process describing the gains or losses realized between date t 1 and
date t;
• .zt /t2N measures the gains or losses realized prior to date t as a fraction of wt , for
each t 2 N;

• .bt /t2N is a process of scaling factors specified as bt D b0 cN t , with b0 0.
Letting b0 D 0 allows to recover the classical expected utility functional in an
infinite horizon economy (compare with Sect. 6.5). The term v.XtC1 ; wtC1 ; zt /
appearing in the preference functional (9.7) represents the utility derived from gains
or losses. In particular, the presence of wtC1 and zt implies that this utility does
not only depend on the gains/losses realized at the current date (captured by XtC1 ),
but also on the previous performance, which affects how current gains/losses are
perceived by the investor.
Let us explain in detail how the processes .Xt /t2N and .zt /t2N are defined.
Denoting by .rt /t2N the return process of the risky asset, the process .Xt /t2N
describing the gains/losses realized at each trading date is defined by
XtC1 WD wtC1 .rtC1 rf /; for all t 2 N:
This corresponds to assuming that the reference point (status quo) at date t C 1 is the
amount of wealth invested at the preceding date t in the risky asset (measured by the
quantity wtC1 ) multiplied by the risk free rate rf . The process .zt /t2N captures the
fact that the pain of a loss also depends on the investment performance prior to the
loss, with zt representing a benchmark level. In this sense, zt < 1 or zt > 1 means
that the agent has realized gains or losses, respectively, prior to date t, while zt D 1
means that the agent has realized neither prior gains nor losses. The benchmark
process .zt /t2N is modeled via

rN
ztC1 D ˇ zt C .1 ˇ/; for all t 2 N;
rtC1
where ˇ 2 Œ0; 1 and rN is a parameter which will be endogenously determined by

requiring that, in equilibrium, the median value of zt is equal to one.
The function v./ is then defined as follows, if zt 1:
(
wtC1 .rtC1 rf /; for rtC1 zt rf ;
v.XtC1 ; wtC1 ; zt / WD
wtC1 .zt 1/rf C wtC1 .rtC1 zt rf /; for rtC1 < zt rf ;
(9.8)
where > 1. On the other hand, if zt > 1, the function v./ is defined as
(
wtC1 .rtC1 rf /; for rtC1 rf ;
v.XtC1 ; wtC1 ; zt / WD (9.9)
.zt /wtC1 .rtC1 rf /; for rtC1 < rf ;
where < .zt / WD C k.zt 1/, for k > 0 (this represents the fact that the
larger zt is the more painful subsequent losses will be). Referring to the original
paper Barberis et al. [150] for a more detailed explanation of the above specification,
we just mention that the specification (9.8)–(9.9) captures three main behavioral
aspects:
• If zt D 1, then the investor is more sensitive to reductions in financial wealth than
to increases (loss aversion). This can be seen from (9.8), noting that for zt D 1
losses are amplified by the factor > 1.
• If zt < 1 (meaning that the investor has accumulated prior gains), then small
losses are not heavily penalized, but once the loss exceeds a certain amount (i.e.,
if rtC1 < zt rf ) then it is penalized at a more severe rate. This is reflected in (9.8).
• If zt > 1 (meaning that the investor has previously realized losses), then losses are
more heavily penalized, capturing the idea that losses following previous losses
are more painful to the investor.
In the context of the above model, Barberis et al. [150] build a one-factor Markov
equilibrium, in the sense that the Markov state variable zt determines the distribution
of future stock returns and the risk free rate is constant. Specifically, Barberis et al.
[150] assume that the equilibrium price-dividend ratio of the risky asset is given by
pt
D f .zt /; for all t 2 N;
dt
for a suitable function f ./. The following proposition, the proof of which is given
in Exercise 9.3, characterizes the equilibrium of the economy.
Proposition 9.5 In the context of the above model (see Barberis et al. [150]),
suppose that
2 c2 2c d C d2

log ı gc C gd C < 0:
2
Then, in equilibrium the risk free rate rf satisfies
1 gc 2 c2
rf D e 2 (9.10)
ı
and the function f ./ determining the equilibrium price of the risky asset satisfies,
for all t 2 N,

ˇ
2 c2 .12 / 1 C f .ztC1 / .d c /"tC1 ˇˇ
1 D ıegd gc C 2 E e ˇFt
f .zt /

ˇ (9.11)
1 C f .ztC1 / gd Cd "tC1 ˇ
C b0 ıE vO e ; zt ˇˇFt ;
f .zt /
where, for zt 1,
(
x rf ; for x zt rf ;
v.x;
O zt / D
.zt 1/rf C .x zt rf /; for x < zt rf I
and, for zt > 1,

(
x rf ; for x rf ;
v.x;
O zt / D
.zt /.x rf /; for x < rf :
The above model permits to rationalize several asset pricing anomalies discussed
in Sects. 7.2 and 7.3. In particular, it is consistent with a low volatility of
consumption growth together with a high volatility and a large risk premium
for stock returns. Moreover, a low risk free rate can be easily generated by the
model. The model predicts a low correlation between consumption growth and
stock returns and generates long horizon predictability in stock returns. The crucial
feature is that, depending on the previous investment performance, agents exhibit
a changing risk aversion (low risk aversion after a market run up and high risk
aversion after a market fall), so that expected returns change over time and a
risk premium for downside risk is generated. Note that this model differs from
habit formation models (see below), since the changes in risk aversion are not
due to changes in the level of consumption. Barberis & Huang [149] analyse asset
pricing implications of cumulative prospect theory, focusing on the effects of the
specification of the weighting function. They show that the CAPM can hold when
securities are normally distributed. However, a positively skewed security can be
overpriced and earn low average returns.
Prospect theory also allows us to rationalize the so-called disposition effect,
i.e., the tendency to hold loser assets too long and to sell winner assets too
soon (see Shefrin & Statman [1531], Ferris et al. [684], Odean [1380], Gomes
[805], Grinblatt & Keloharju [829], Barberis & Xiong [156, 155]). The rationale of
this behavior is that, in the case of a winner asset, the investor is in the concave
region and, therefore, he is risk averse. On the contrary, in the case of a loser
asset, the investor is in the convex region and, therefore, he is risk lover. Intraday
empirical evidence supporting loss aversion is provided in Coval & Shumway
[503].
The disposition effect may generate underreaction to information and a momen-
tum effect in stock returns, as shown in Grinblatt & Han [827]. More specifically,
Grinblatt & Han [827] consider a multi-period equilibrium model with a single risky
asset (with supply normalized to one). The fundamental value . ft /t2N of the risky
asset is assumed to follow the random walk
ftC1 D ft C "tC1 ; for all t 2 N; (9.12)
where ."t /t2N is a sequence of i.i.d. random variables with zero mean. At each date
t 2 N, public news about the fundamental value ft arrive just prior to trading. The
economy is populated by two classes of agents: rational traders and prospect theory
traders. It is assumed that prospect theory traders constitute a proportion 2 .0; 1/
of the overall population. The optimal demand of the risky asset by the two classes
of agents is assumed to be of the following form:

tC1
PT
D 1 C bt . ft pt / C .xt pt / ;
(9.13)
tC1
RA
D 1 C bt . ft pt /;
where tC1PT
and tC1
RA
denote the optimal demand of a prospect theory agent and
of a rational agent, respectively, for t 2 N. In (9.13), . pt /t2N denotes the market
price of the risky asset, .bt /t2N is allowed to be an adapted stochastic process and
depends on the choice of the utility function as well as on the structure of the
economy. For the purposes of the present discussion, we do not need to specify
.bt /t2N further. The process .xt /t2N , supposed to be a predictable stochastic process,
represents a reference price of the risky asset with respect to which prospect theory
investors measure their gains and losses. Finally, the coefficient > 0 represents the
relative importance of the capital gain component in the demand of prospect theory
investors. The perturbation term appearing in the demand function of prospect
theory investors is inversely related to the unrealized profit on the risky asset. As
we are going to show, this specification allows us to obtain an explicit description of
the deviation of the risky asset market price from the equilibrium price that would
prevail if all the investors were rational.
The reference price process .xt /t2N is assumed to satisfy the following relation:
xtC1 D vt pt C .1 vt /xt ; for all t 2 N; (9.14)
capturing the fact that the reference price gets updated as shares are exchanged
between investors at each trading date, so that new reference price is a weighted
average of the old reference price and the current market price. The process .vt /t2N ,
taking values in the interval Œ0; 1, represents the weighting factors and captures the
mental accounting of prospect theory traders.
Proposition 9.6 In the context of the above model (see Grinblatt & Han [827]), the
following hold:
(i) the equilibrium market price . pt /t2N of the risky asset is given by
pt D wft C .1 w/xt ; for all t 2 N;
where w WD 1=.1 C /;

(ii) at each date t 2 N, the expected return of the risky asset satisfies

ptC1 pt ˇˇ p t xt
E ˇFt D .1 w/vt : (9.15)
pt pt
Proof Part (i) simply follows by aggregating the optimal demands given in (9.13)
and equating the aggregate demand with the market supply (which is equal to one
at all dates t 2 N). In order to prove part (ii), note that
EŒ ptC1 pt jFt D wEŒ ftC1 ft jFt C .1 w/EŒxtC1 xt jFt

D wEŒ"tC1 jFt C .1 w/vt EŒ pt xt jFt
D .1 w/vt . pt xt /;
where we have used part (i) of the proposition together with relations (9.12) and
(9.14). t
u
Part (i) of Proposition 9.6 shows that the risky assets underreact to information
about the fundamental value. The degree of underreaction is measured by the term
w, which is in turn determined by the proportion of prospect theory traders in the
economy and by the parameter characterizing the optimal demand of prospect
theory traders. The implication is that in equilibrium stocks that have performed
well in the past tend to be undervalued and, on the contrary, stocks that have badly
performed tend to be overvalued. The smaller the number of prospect theory traders
the closer the market price will be to the fundamental value. The right-hand side of
(9.15) is related to the (percentage) unrealized capital gain. Relation (9.15) implies
that any variable which proxies for the unrealized capital gain (for instance, past
returns) has a predictive power on future returns. In this sense, this model is able to
explain the profitability of momentum strategies.
In the decision theory literature, disappointment or anticipation effects have also
been considered. Disappointment occurs when an agent assigns more importance
to the bad outcomes than to the good outcomes of a lottery, while anticipation
refers to the fact that current decisions are affected by suspense and anxiety about
the future. Agents may experience disappointment by comparing an outcome with
its past expectation. On the other hand, agents may anticipate future utility and,
therefore, the current utility will be affected by the expectation of future utility.
An axiomatic framework for disappointment aversion has been first proposed in
Gul [863], including classical expected utility as a special case and rationalizing
the Allais paradox. More specifically, preferences are characterized in Gul [863] in
terms of a utility function u and a parameter ˇ > 1 (the case of classical expected
utility corresponding to ˇ D 0). The case ˇ 0 captures disappointment aversion
and it is shown that risk aversion implies disappointment aversion. The effects of
disappointment will also be discussed below in the context of recursive preferences.
The asset pricing results obtained in Ang et al. [64], Routledge & Zin [1476],
Bonomo et al. [267], Liu & Miao [1225], Fielding & Stracca [700], assuming a util-
ity function capturing disappointment aversion (good outcomes are underweighted
with respect to bad outcomes), confirm that disappointment may help to explain the
equity premium and the risk free rate puzzles. Furthermore, disappointment aversion
can also explain the low participation rate in the market with a counter-cyclical risk
aversion. There is evidence that disappointment is priced in the market: Ang et al.
[67] show that returns are increasing in downside risk, i.e., the risk that asset returns
are more correlated with the market when the latter is falling rather than when it is
rising. The impacts of anticipation effects on asset prices have been investigated in
Caplin & Lehay [365], where the authors show that anxiety (the agent is anxious
for future wealth-consumption) contributes to explain both the risk free rate and
the equity premium puzzle, due to the time inconsistency in investors’ preferences
caused by anticipation.
Recursive Preferences
In Chap. 6, we have considered time additive utility functions. As explained in

Sect. 7.3, one of the drawbacks of this specification consists in the fact that
the relative risk aversion coefficient determines the elasticity of intertemporal
substitution of consumption. Time additive preferences have been generalized by
introducing recursive utilities (also known as generalized expected utility), see
Epstein & Zin [649] and Weil [1648]. In particular, one of the main advantages of
this preference structure consists in the possibility of separating risk aversion from
the elasticity of intertemporal substitution of consumption.
In order to illustrate the theory of recursive preferences, we adopt the same
setting introduced in Sect. 6.1 for multi-period optimal consumption-investment
problems. We consider an economy with T 2 N trading dates t 2 f0; 1; : : : ; Tg and
N traded securities. Trading-consumption strategies are defined as in Definition 6.1
and denoted by .; c/, with .t /tD0;1;:::;T being an RN -valued predictable stochastic
process representing the trading strategy and .ct /tD0;1;:::;T a non-negative adapted
stochastic process representing the consumption stream. As in Sect. 6.1, we define
by Ct .x/ the set of all consumption processes .cs /sDt;:::;T which can be financed by
some trading strategy starting at date t with wealth x. We shall always assume that
trading is done in a self-financing way and, for simplicity, we do not consider the
presence of an exogenous income stream.
Extending the classical time additive expected utility function, we assume that
preferences are defined with respect to self-financing trading-consumption strategies
.c; / and are characterized by a time-dependent preference functional Ut .c; /
satisfying the following recursive relation:

Ut .c; / D v ct ; t UtC1 .c; / ; for all t D 0; 1; : : : ; T 1; (9.16)
with UT .c; / D u.cT /, for some utility function u./. In (9.16), the term ct
represents the current consumption at date t, while t .UtC1 .c; // represents the
certainty equivalent, as measured at date t, of the future utility at the following date
t C 1. More precisely, the functional t is specified as

t UtC1 .c; / D uQ 1 E uQ UtC1 .c; / jFt ; for all t D 0; 1; : : : ; T 1;
(9.17)
for some increasing and concave function uQ ./. The function v W RC R ! R

appearing in (9.16) is also assumed to be increasing and concave. The role of the
function v consists in aggregating the utility derived from current consumption and
the future utility. With this specification, risk aversion and intertemporal substitution
of consumption can be easily separated. Indeed, risk aversion is encoded in the
functional t , while the propensity of substituting consumption intertemporally is
determined by the function v. Note that (9.16) implies that the utility associated to
a consumption stream .ct /tD0;1;:::;T can be computed as the solution to a backward

stochastic difference equation. As shown in Exercise 9.4, the time additive expected
utility of the form (6.6) represents a special case of (9.16).
In Epstein & Zin [649, 651], the following specification of recursive preferences
has been proposed (in the original formulation, an infinite horizon economy was
considered):
1%
1
x1˛
v.c; y/ D .1 ı/c1% C ıy1% and uQ .x/ D ; (9.18)
1˛
for % 2 .0; 1/, ı 2 .0; 1/ and ˛ 2 .0; 1/. With this specification, the recursive
relation (9.16) can be rewritten in the following form:
1
1% 1% 1%
Ut .c; / D .1 ı/ct C ı EŒUtC1 .c; /1˛ jFt 1˛ ; (9.19)
for all t D 0; 1; : : : ; T 1. We also assume that u.cT / D c1˛

T =.1˛/. The parameter
˛ represents the coefficient of relative risk aversion, while the parameter 1=%
represents the elasticity of intertemporal substitution of consumption (hence, the
preference functional (9.19) exhibits constant elasticity of intertemporal substitution
of consumption and constant relative risk aversion). As shown in Exercise 9.4,
this specification allows to recover the classical time additive expected utility
representation (6.6) with a power utility function if ˛ D % (i.e., relative risk
aversion coincides with the inverse of the elasticity of intertemporal substitution
of consumption). Moreover, taking the limit for ˛ ! 1, a logarithmic utility
specification is obtained (in a time additive form if ˛ D % D 1). Finally, the
parameter ı plays the role of a (subjective) discount factor.
An important feature of the recursive utility functional (9.19) is that it allows to
model different preferences with respect to the resolution of uncertainty. Indeed, if
˛ D % then the agent is indifferent between early and late resolution of uncertainty,
while, if ˛ > %, then early resolution of uncertainty is preferred (as it seems
intuitively plausible) and, if ˛ < %, then late resolution of uncertainty is preferred
(see Epstein & Zin [649]).
In order to analyse optimal investment-consumption problems under recursive
preferences, proceeding similarly as in Sect. 6.1, let us introduce the value function
V defined by

V.x; t/ WD sup Ut .c; / D sup v ct ; t UtC1 .c; / ; (9.20)
c2Ct .x/ c2Ct .x/
for all t D 0; 1; : : : ; T 1 and for an arbitrary initial wealth x 2 RC , with V.x; T/ D

u.x/. Note that, if the function v is assumed to be homogeneous of degree one,
as in (9.18), then the value function V is proportional to wealth (see Epstein & Zin
[649, 651]). The value function satisfies a suitable version of the Bellman optimality
principle, as shown in the following proposition, which can be proved by relying on

the same arguments given in the proof of Propositions 6.3 and 6.4.
Proposition 9.7 Consider the recursive preference functional of the general form
(9.16) and the corresponding value function V defined in (9.20). Then, for every
.x; t/ 2 RC f0; 1; : : : ; T 1g, the value function V satisfies the following recursive
relation:

V.x; t/ D sup v ct ; t V.WtC1 ./; t C 1/ ; (9.21)
where the supremum is taken with respect to all Ft -measurable random variables
> >
.tC1 ; ct / 2 RN RC satisfying tC1 St C ct D x and where WtC1 ./ D tC1 .StC1 C

DtC1 /. Moreover, in correspondence of an optimal solution . ; c /, it holds that

V.Wt ; t/ D v ct ; t V.WtC1

; t C 1/ ; for all t D 0; 1; : : : ; T 1; (9.22)
with Wt WD Wt . /, and the following envelope condition holds:
@

v ct ; t V.WtC1 ; t C 1/ D V 0 .Wt ; t/; for all t D 0; 1; : : : ; T 1;
@ct
with V 0 .; t/ denoting the first derivative of the value function with respect to its first
argument, for t D 0; 1; : : : ; T.
Formula (9.21) corresponds to the equation of the dynamic programming
principle in the context of general recursive preferences. According to (9.22), in
correspondence of the optimal trading-consumption strategy . ; c /, the optimal
utility at each date t corresponds to the utility derived from current consumption
at date t and from the certainty equivalent of the future date’s optimal utility,
aggregated by the function v.
Proposition 9.7 implicitly characterizes the optimal trading-consumption strategy
. ; c / maximizing the recursive preference functional (9.16). In particular, in the
case of recursive preferences of the form (9.19), explicit Euler conditions can be
formulated, as shown in the following proposition, the proof of which is given in
Exercise 9.5.
Proposition 9.8 Consider the recursive preference functional defined in (9.19),
with parameters % 2 .0; 1/, ı 2 .0; 1/ and ˛ 2 .0; 1/ and with UT .c; / D u.c/ WD
c.1 ı/1=.1%/ . Denote by .Wt /tD0;1;:::;T the optimal wealth process associated
to the trading-consumption strategy . ; c / maximizing the recursive preference
functional U0 .c; / defined in (9.19). Then the following Euler condition holds:
" !%˛ % ˇ #

V.WtC1 ; t C 1/ ctC1 n ˇ
ˇ
1 D ıE
rtC1 ˇFt ; (9.23)
t V.WtC1 ; t C 1/ ct
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N. In particular, assuming that there

exists a risk free asset with constant rate of return rf > 0, it holds that
" !%˛ % ˇ #

V.WtC1 ; t C 1/ ctC1 ˇ
E
.rtC1
n
rf /ˇˇFt D 0; (9.24)
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N. Moreover, the optimal wealth process

.Wt /tD0;1;:::;T satisfies
V.Wt ; t/1% %
Wt D .ct / ; for all t D 0; 1; : : : ; T; (9.25)
1ı
and the return process .rt /tD1;:::;T on the optimal wealth process satisfies
!1% %
W 1
V.WtC1 ; t C 1/ ctC1
rtC1 WD tC1 D
; (9.26)
Wt ct ı t V.WtC1 ; t C 1/ ct
for all t D 0; 1; : : : ; T 1. Therefore, assuming the existence of a risk free asset with
constant rate of return rf > 0, the Euler condition (9.24) can be rewritten as

% ˇ
ctC1 ˇ
E
.rtC1 / 1 .rtC1
n
rf /ˇˇFt D 0; (9.27)
ct
for all t D 0; 1; : : : ; T and n D 1; : : : ; N, with WD .1 ˛/=.1 %/.

It is interesting to compare Proposition 9.8 with the results obtained in the context
of optimal investment-consumption problems for classical time additive expected
utility functions, as considered in Sect. 6.1, in the case of a power utility function.
Starting with the Euler conditions (9.23)–(9.24), note first that, in the special case
where ˛ D %, then the first term inside the conditional expectation vanishes and the
conditions thus reduce to the classical Euler conditions obtained for a time additive
power utility function (compare with Exercise 6.6). This is a consequence of the fact
that (9.19) reduces to a classical time additive power utility function if ˛ D %, as
shown in Exercise 9.4.
While Proposition 9.8 has been established for ˛ 2 .0; 1/ and % 2 .0; 1/, it can
be shown that if ˛ D 1 and % ¤ 1, the Euler condition (9.27) reduces to
1 n ˇ
E .rtC1 / .rtC1 rf /ˇFt D 0; (9.28)
for all t D 0; 1; : : : ; T and n D 1; : : : ; N (see Epstein & Zin [651] and Attanasio &
Weber [86]). When ˛ D 1, Giovannini & Weil [785] have shown that the portfolio
choice is myopic (Samuelson’s result) for any value of %. On the other hand, if % D 1
then ct =Wt is constant over time for any value of ˛ (see Giovannini & Weil [785]
and compare also with relation (9.25)).
Relation (9.25) gives a precise link between the optimal wealth process and the
optimal consumption process. In particular, due to the specification (9.18), it can
be shown that the value function is linear with respect to x (see Epstein & Zin
[651]). As a consequence, relation (9.25) implies that the optimal consumption ct
is proportional to the current optimal wealth Wt . This result is analogous to the
case of a classical time additive power utility function and has to be compared with
relation (6.26) in Proposition 6.8.
In the general case where ˛ ¤ %, the Euler conditions (9.23)–(9.24) show that
the stochastic discount factor .Mt /tD0;1;:::;T associated to the recursive preferences
(9.19) is given by
% Y !%˛
t1
ct V.WsC1 ; s C 1/
Mt D ı t
; for all t D 1; : : : ; T;
c0 sD0

s V.WsC1 ; s C 1/
(9.29)
with M0 D 1 (see also Exercise 9.5). Recall that the parameter % represents the recip-
rocal of the elasticity of intertemporal substitution of consumption. In comparison
with the classical structure (6.64) of the stochastic discount factor obtained with a
time additive utility function, (9.29) shows that the stochastic discount factor under
recursive preferences depends not only on the optimal consumption process but also
on the certainty equivalent of future’s utility. Moreover, as a consequence of (9.26)–
(9.27), the stochastic discount factor admits the following alternative representation:
% Y
t
ct
Mt D ı t .rs / 1 ; for all t D 1; : : : ; T; (9.30)
c0 sD0
with M0 D 1. This shows that the stochastic discount factor depends on the
return process .rt /tD1;:::;T associated to the optimal portfolio in the maximization
of the recursive preference functional (9.19). Relation (9.30) also shows that the
stochastic discount factor associated to the recursive preference functional (9.19) is
a geometric average of the stochastic discount factor associated to a standard time
additive power utility function and the stochastic discount factor associated to the
logarithmic case (compare with relation (9.28) above). The weight between these
two components is determined by the parameter .
As a consequence of the Euler condition (9.27), asset risk premia depend on
the covariance between asset excess returns and a geometric average of the growth
rate of the optimal consumption process and of the return on the optimal portfolio.
In particular, when ˛ D % (so that D 1), asset risk premia only depend
on the covariance between the excess returns and the growth rate of the optimal
consumption process, thus recovering the classical CCAPM result with a time
additive power utility function (see Sect. 6.4). On the other hand, if ˛ D 1 (so
that D 0), then asset risk premia are determined similarly as in the CAPM,
with the reciprocal of the return on the optimal portfolio playing the role of the
single risk factor. For other values of , both the consumption growth rate and
the optimal return are necessary for determining asset risk premia. More precisely,
% ˇ
ctC1 n ˇ
ˇ
EŒrtC1
n
jFt rf D rf Cov .rtC1 / 1 ; rtC1 ˇFt ; (9.31)
ct
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N. Moreover, by approximating the

geometric average with the arithmetic average, we obtain
ˇ
ctC1 n ˇ
EŒrtC1
jFt rf rf .1 / Cov.rtC1 ; rtC1 jFt / C rf % Cov ; ˇ
tC1 ˇFt :
n n
r
ct
(9.32)
Once more, if D 1, then we recover the classical asset pricing relation of the
CCAPM, while, if D 0, relation (9.32) reduces to the CAPM relation. Moreover,
under suitable assumptions on the joint conditional distribution of the optimal
consumption growth rate and of the returns, relation (9.32) can be shown to hold
in an exact form, as shown in the following proposition, the proof of which is given
in Exercise 9.6 (compare also with Munk [1363, Theorem 9.4]).
Proposition 9.9 Under the same assumptions of Proposition 9.8, let us denote by
.rt /tD1;:::;T and by .ct /tD0;1;:::;T the return process on the optimal portfolio and the
optimal consumption process, respectively, associated to the trading-consumption
strategy . ; c / maximizing the recursive preference functional (9.19). Suppose
that, for every t D 0; 1; : : : ; T 1, conditionally on Ft , it holds that

ctC1
log N .c ; c2 / and
log rtC1 N .r ; r2 /;
ct
with Cov.log.ctC1 =ct /; log rtC1

jFt / D cr , for all t D 0; 1; : : : ; T 1. Then, the
equilibrium risk free return rf is given by
r2 %2 c2
log rf D log ı C %c C . 1/ : (9.33)
2 2
Suppose furthermore that .log.ctC1 =ct /; log rtC1

1
; : : : ; log rtC1
N
/ is jointly normally
distributed conditionally on Ft , for all t D 0; 1; : : : ; T 1. Then the asset risk
premia in equilibrium satisfy
n
Var.log rtC1 jFt /

EŒlog rtC1
n
jFt log rf D . 1/ Cov log rtC1 ; log rtC1
n
jFt
2
ˇ
c ˇ
C % Cov log tC1 ; ˇ
tC1 ˇFt ;
/
n

log.r
ct
(9.34)
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N.
In particular, if ˛ D % (so that D 1), then the result of the above proposition is
in line with Proposition 6.38 (compare the asset pricing relations (6.78) and (9.34)).
In the case of general distributions for the return and consumption processes, the
asset pricing relations stated in Proposition 9.9 hold in an approximate form (see
Campbell & Viceira [354, Section III]).
As we have already remarked, an important feature of the recursive preference
functionals (9.16)–(9.19) consists in the separation between risk aversion and
the elasticity of intertemporal substitution of consumption. This feature seems to
be promising in order to provide an explanation for the equity premium puzzle
and the risk free rate puzzle (see Sect. 7.3). In particular, relation (9.33) shows
that the equilibrium risk free rate is determined not only by the degree of risk
aversion (encoded in the parameter ), but also by the average growth rate of the
consumption process and by the elasticity of intertemporal substitution (measured
by the parameter %). As a consequence, a high degree of risk aversion is no longer
incompatible with a low risk free rate. This allows us to rationalize the risk free
rate puzzle. Moreover, a high degree of risk aversion is no longer the only way
of explaining the equity premium puzzle, since asset premia depend on both the
covariance with consumption growth and the covariance with the return on a market
portfolio. Unfortunately, as we are going to discuss below, recursive preferences
have been empirically shown to provide only a partial resolution to the equity
premium puzzle.
Assuming i.i.d. aggregate consumption growth rates, Weil [1648], Kocherlakota
[1110], Epstein & Melino [642] show that a representative agent economy with
a utility functional of the type (9.19) is observationally equivalent to a classical
economy with a time additive expected utility function. The parameters ˛ and %
cannot be identified separately. On the other hand, assuming a Markovian process
for the evolution of the aggregate consumption growth rates, Wang [1645] obtains a
non-equivalence result.
It has been empirically observed that the equity premium puzzle cannot be easily
explained by simply assuming recursive preferences, see Weil [1648], Kandel &
Stambaugh [1069], Kocherlakota [1110, 1112]. Unlike the risk free rate puzzle, the
equity premium puzzle does not depend on the relation between risk aversion and
the elasticity of intertemporal substitution of consumption imposed by the classical
time additive expected utility framework. The equity premium puzzle originates
from the fact that the consumption process is too smooth and, therefore, the degree
of risk aversion needs to be dramatically high in order to reproduce in equilibrium
the risk premia observed historically. Instead, recursive preferences may explain the
risk free rate puzzle, as shown in Kocherlakota [1112]. Indeed, as we have already
mentioned, the essential ingredient of the risk free rate puzzle is the connection
between risk aversion and the intertemporal substitution of consumption. Consider-
ing recursive preferences of the form (9.19), both risk aversion and the elasticity of
intertemporal substitution can be simultaneously high. As a consequence, in order
to generate a low risk free rate in equilibrium, it suffices to choose an appropriate
value of the elasticity of intertemporal substitution of consumption. However, the
case of the equity premium puzzle is different. In order to fit both the risk free rate
and the risk premium, Weil [1648] need to choose a relative risk aversion coefficient
around 45 and an elasticity of intertemporal substitution equal to 0:10 (higher than
1=45). As shown in Kandel & Stambaugh [1069], a low elasticity of intertemporal
substitution generates predictable asset returns and a high volatility of returns.
Volatility is determined primarily by the elasticity of intertemporal substitution.
Kandel & Stambaugh [1069] manage to explain the risk free rate puzzle as well
as the equity premium puzzle, but find it difficult to reproduce first and second
moments of returns.
A more positive evidence on the possibility of explaining the above phe-
nomena by relying on recursive preferences has been provided in Attanasio &
Vissing-Jørgensen [85] and Guvenen [867, 868] considering limited stock market
participation and heterogeneous agents. In Bansal & Yaron [139], it is shown that
a representative agent economy with recursive preferences of the type (9.19) can
simultaneously explain the equity premium puzzle, the risk free rate puzzle and
the volatility of equity returns. In their model, the consumption and the dividend
growth rates contain a small long run predictable component and a fluctuating
economic uncertainty (consumption volatility). As argued in Bansal & Yaron [139],
in order to reproduce the above phenomena, it is important to consider both recursive
preferences and suitable dynamics for the consumption and dividend growth rates
(see however Constantinides & Ghosh [497] for some negative evidence concerning
the estimation of the Bansal & Yaron [139] model). In a related context, positive
evidence for models with recursive preferences has been provided in Ai [30] and
Lettau et al. [1197], showing that a plausible level of risk aversion is consistent
with the observed risk premia. Moreover, long run risk and recursive preferences
allow also to explain risk premia anomalies cross-sectionally (see Bansal et al.
[134, 135], Hansen et al. [889], Malloy et al. [1291]).
The implication of the Euler conditions stated in Proposition 9.8 have been tested
in Epstein & Zin [651] on consumption and monthly returns time series by applying
a methodology similar to that employed in Hansen & Singleton [895]. It is shown
that the elasticity of intertemporal substitution is typically small (thus confirming the
result of Hall [880]), risk preferences do not differ statistically from the logarithmic
case and investors prefer a late resolution of uncertainty. The empirical evidence
reported in Epstein & Zin [651] is in favor of recursive preferences and leads to a
rejection of the hypothesis of a time additive utility function. However, Epstein &
Zin [651] report that the performance of the model and of the tests is sensitive to
the choice of the instrumental variables. In Jorion & Giovannini [1048], it has been
empirically shown that a recursive utility function does not improve the performance
of a time additive utility function.
Empirical tests of intertemporal asset pricing models require consumption time
series. Consumption data are typically measured with errors and are aggregated.
This may have serious consequences on asset pricing tests, with a bias towards
rejection. Campbell [332], assuming that asset returns and news about future returns
are jointly log-normally distributed and employing a log-linear approximation of
the budget constraint (see Exercise 9.7), replaces consumption with wealth in the
asset pricing formulae obtained above. The approximation is accurate when the
variation of the consumption-wealth ratio is small. Similarly as in Proposition 9.9,
the logarithmic risk premium of an asset is related to the variance of the logarithmic
return of the asset and to its covariances with the logarithmic return of invested
wealth (market portfolio) and with news about future returns on invested wealth
(changing investment opportunities). Under suitable conditions (notably in the case
˛ D 1 or in the presence of constant investment opportunities), the last term can be
ignored, thus obtaining an approximate asset pricing relation where the market risk
is the single risk factor. Otherwise, a multi-factor asset pricing model is obtained,
with the market portfolio (and not consumption) being one of the risk factors. This
result can also be derived in the presence of heteroskedasticity.
In Epstein & Zin [650], a recursive utility characterized by first order risk
aversion has been proposed. This type of utility function, introduced in Segal &
Spivak [1514], is such that the risk premium for a small gamble is proportional to
its standard deviation rather than to its variance (compare with Sect. 2.2). Since
the aggregate consumption process is smooth, its standard deviation is typically
larger than its variance. This type of utility functional seems to be promising in
order to explain the equity premium and the risk free rate puzzles. Indeed, unlike
the case of a classical utility function, a high equity premium is compatible with
a low risk free rate. By calibrating the model, the authors show that a low risk
free rate is compatible with a risk premium larger than the one associated with
a standard utility function but is still significantly lower than the historical value,
thus providing only a partial resolution of the equity premium puzzle. In Bekaert
et al. [187], an equilibrium model with a utility function characterized by first
order risk aversion has been adopted to model time varying risk premia and to
reproduce the observed predictability of asset returns. Increasing first order risk
aversion substantially increases the variance of the risk premia, but the effect on
return predictability does not suffice to fit the observed time series.
Portfolio choices with recursive preferences have been analysed by Campbell
& Viceira [354], Campbell et al. [340], Chacko & Viceira [387], Gomes &
Michaelides [808], Cocco et al. [456] in various settings, also taking into
account the presence of mean reversion and stochastic volatility. In particular,
as shown in Exercise 9.7, in Campbell & Viceira [354] an approximate
formula for the optimal demand of the risky asset has been obtained by taking
a log-linear approximation of the Euler condition and of the self-financing
condition, assuming recursive preferences of the form (9.19). The approximate
formula derived in Campbell & Viceira [354] is composed of two terms:
the first term is related to the risk premium of the asset, while the second
term reflects the intertemporal hedging demand (see Exercise 9.7 for more
details).
Habit Formation and Durable Goods
Besides recursive preferences, another generalization of the classical time additive

expected utility theory consists in relaxing the separability of the preference
functional with respect to time. In other words, we can consider utility functionals
where the current utility at date t depends not only on consumption ct at date t, but
also on a process zt related to consumption at the previous dates. More specifically,
let us consider a preference functional of the following form, defined with respect
to a consumption process c D .ct /tD0;1;:::;T :
X
T
ı t EŒu.ct ; zt /: (9.35)
tD0
The process .zt /tD0;1;:::;T can be specified as a function of the previous consumption
or can also play the role of an exogenous factor influencing the preferences (this
is for instance the case of the “Keeping up with the Joneses” preferences discussed
below). For the sake of the present discussion, let us suppose that .zt /tD0;1;:::;T is
given by a weighted average of past consumption, i.e.,
X
t
zt D bt z0 C a bs cts ; for all t D 1; : : : ; T; (9.36)
sD1
with a > 0; 0 < b < 1 and z0 a given initial value. Equivalently, for a given
consumption process .ct /tD0;1;:::;T , the process .zt /tD0;1;:::;T can be characterized as
the solution to the recursive relation
zt D bzt1 C abct1 ; for all t D 1; : : : ; T;
with initial value z0 . In a continuous time formulation,

Rt the specification (9.36) can
be naturally generalized to zt D ebt z0 C a 0 eb.ts/ cs ds, for some a; b > 0.
The utility function u appearing in (9.35) now depends on two arguments.
Depending on the behavior of u with respect to the second argument, two different
situation arise (in the following, we denote by @z u.c; z/ the first partial derivative of
the function u with respect to its second argument):
• If @z u.c; z/ < 0, then the preference functional (9.35) represents habit formation
(or habit persistence): an agent compares his current consumption with the level
of past consumption. If the agent has experienced in the past a high level of
consumption, then he continues to desire a high level of consumption and the
agent will be disappointed if he has to reduce the current level of consumption. In
other words, we can think of the process .zt /tD0;1;:::;T as representing a standard
of living. The discounting of past consumption in (9.36) captures the fact that
the effect of past consumption is declining over time and that the habit mostly
depends on recent rather than remote consumption.
• If @z u.c; z/ > 0, then the preference functional (9.35) represents the presence of
durable goods and we can think of the process .zt /tD0;1;:::;T as modeling durable
goods that an agent accumulates over time (a typical example being housing).
According to these two interpretations, we can have complementarity or substi-
tutability effects between consumption at different points in time. Indeed, if there
exists a complementarity (substitutability, resp.) effect between consumption at
different dates, then the second cross derivative of the function u will be negative
(positive, resp.). When goods are durable, consumptions at nearby dates are almost
perfect substitutes (see also below).
As suggested in Constantinides [493], a preference functional of the form (9.35)
with a habit formation process .zt /tD0;1;:::;T allows to generate a wedge between the
coefficient of relative risk aversion and the reciprocal of the elasticity of intertempo-
ral substitution of consumption, thus making possible the resolution of several asset
pricing puzzles, notably the risk free rate puzzle and the equity premium puzzle
(see Constantinides [493]). The utility function proposed in Constantinides [493] is
of the following form, for all t D 0; 1; : : : ; T:
(
.ct zt /1˛
1˛
; if ct zt ;
u.ct ; zt / D (9.37)
1; if ct < zt ;
with .zt /tD0;1;:::;T representing the habit formation process, given by a weighted
average of past consumption, and ˛ > 0. By adopting this specification, in a
continuous time formulation, Constantinides [493] shows that the risk premium
observed historically can be fitted with ˛ < 3 (see however Chapman [410] and
Otrok et al. [1388] for a critical analysis of this result). The specification (9.37) is
sometimes called the difference model. A similar result was obtained in Abel [4]
assuming

.ct =zt /1˛
u.ct ; zt / D ; (9.38)
1˛
with 2 Œ0; 1. Specification (9.38) is sometimes called the ratio model. Note that
the difference model generates a time-varying coefficient of relative risk aversion,
while the ratio model generates a constant coefficient of relative risk aversion.
Indeed, in the difference model it holds that
@cc u.ct ; zt / ˛ct

rur .ct ; zt / D ct D ;
@c u.ct ; zt / ct zt
while in the ratio model it holds that
@cc u.ct ; zt /
rur .ct ; zt / D ct D ˛;
@c u.ct ; zt /
for all t D 0; 1; : : : ; T, where we denote by @c and @cc the first and the second partial
derivatives, respectively, of the function u with respect to its first argument.
For the sake of the present discussion, let us consider a preference functional
of the form (9.35), defined with respect to a habit formation process satisfying
(9.36), with a general utility function u. Consider a multi-period economy with
N C 1 traded assets, where asset 0 is a risk free asset paying the constant rate
of return rf and the remaining N assets are risky and deliver random dividends.
Similarly as in Sect. 6.1, we consider self-financing trading-consumption strategies
.; c/. As shown in Exercise 9.8, the Euler condition associated to the optimal
investment-consumption problem under habit formation preferences can be written
in the following form (compare also with Proposition 6.4):
" PTt1 ˇ #
@c u.c
tC1 ; ztC1 / C a

sD1 .ıb/ EŒ@z u.ctC1Cs ; ztC1Cs /jFtC1 n
s
ˇ
E PTt .rtC1 rf /ˇˇFt D 0;
@c u.c
t ; zt / C a sD1 .ıb/s EŒ@z u.c
tCs ; ztCs /jFt
(9.39)
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N, where we denote by .zt /tD0;1;:::;T the

habit formation process associated to the optimal consumption stream .ct /tD0;1;:::;T .
The above Euler condition is rather complicated, mainly due to the endogeneity of
the habit formation process .zt /tD0;1;:::;T .
A simpler Euler condition can be obtained by assuming that the habit formation
process .zt /tD0;1;:::;T is simply given by the level of consumption at the previous date,
i.e., zt D ct1 , for all t D 1; : : : ; T. Following Kocherlakota [1112], suppose that
the utility function u.c; z/ appearing in (9.35) is given by
.ct ct1 /1˛

u.ct ; zt / D ; (9.40)
1˛
for > 0. The economic intuition is that an agent having consumed ct1 at
date t 1 has become familiar with that level of consumption and, therefore, he
compares his current consumption at date t with the previous level of consumption.
In other words, lagged consumption endogenously determines a subsistence level.
Note that the instantaneous utility at date t is a decreasing function of consumption
at the previous date t 1 (provided that ct ct1 > 0). See also Munk [1363,
Example 6.3] for a simple two-period example of habit formation preferences of the
type (9.40). As shown in Exercise 9.8, in the case of the utility function (9.40), the
Euler condition (9.39) reduces to

ˇ
.ctC1 ct /˛ ıEŒ.ctC2 ctC1 /˛ jFtC1 n ˇ
E ˛ ˛
.rtC1 rf /ˇˇFt D 0;
.ct ct1 / ıEŒ.ctC1 ct / jFt
(9.41)
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N. Observe that both Euler conditions

(9.39) and (9.41) depend on the investor’s ability to predict future consumption
growth. This also implies that the stochastic discount factor depends on the
individual information.
In Sect. 7.3, we have observed that the aggregate consumption time series is
typically smooth, an observation which does not agree with several asset pricing
results obtained with a time additive utility function. Sundaresan [1580] has shown
that a utility function with habit persistence generates a consumption process which
is less sensitive to wealth shocks than the one obtained with a time additive utility
function (excess smoothness of consumption). This result has been confirmed in
Detemple & Zapatero [567] for a generic utility function u.c; z/ decreasing with
respect to the second argument (habit formation), under suitable conditions. Similar
results are also obtained assuming that the instantaneous utility depends negatively
on the current value of the habit process and on the conditional expectation of future
habit, see Antonelli et al. [73].
In Detemple & Zapatero [567] and Sundaresan [1580], two-factor and multi-
factor versions of the CCAPM are derived. In general, the presence of habit
formation leads to an increase of the risk premium in comparison to the classical
case. Moreover, the elasticity of intertemporal substitution becomes more volatile.
However, a resolution of the risk premium puzzle comes at the expense of a rather
poor performance in fitting the variance of asset returns and, in particular, of the
short term interest rate, see Heaton [924], Heaton & Lucas [927], Cochrane [463].
Habit formation preferences help to resolve the risk free rate puzzle, since the intro-
duction of habit persistence allows to disentangle risk aversion from the elasticity of
intertemporal substitution, see Daniel & Marshall [519] and Kocherlakota [1112].
Consumption and portfolio choices under habit formation preferences have
been analysed in Gomes & Michaelides [808], Polkovnichenko [1426], Díaz
et al. [572]. In Alessie & Lusardi [39], a closed-form solution is derived, with
optimal current consumption also depending on past consumption. The optimal
consumption problem in the presence of durable goods has been investigated in
several papers, see Dunn & Singleton [602], Eichenbaum & Hansen [633], Cuoco
& Liu [508], Grossman & Laroque [844], Hindy & Huang [943], Campbell &
Cocco [341], Cocco [455], Yao & Zhang [1664], Yogo [1665], Villaverde & Krueger
[1624], Bertola et al. [211], Flavin & Nakagawa [716]. In some of these models,
there exists a single durable good (typically housing), while in other models there
are both durable and non-durable goods. In the first case, the utility function can be
a function only of zt .
Similarly as in the habit formation case, durability introduces a wedge between
risk aversion and the elasticity of intertemporal substitution. However, durable
goods and habit formation have different (and sometimes contrary) asset pricing
implications. In Hindy & Huang [943], the risk premium in the presence of durable
goods turns out to be smaller than in the classical case. As a matter of fact, when
the levels of consumption at nearby dates are substitutes, agents tend to be less risk
averse. In Grossman & Laroque [844], it is shown that, in the presence of an illiquid
durable good, the optimal consumption is not a smooth function of wealth. Indeed,
it is optimal to wait until a large shock in wealth before adjusting the consumption.
As a consequence, the CCAPM clearly fails but the CAPM still holds (see also
Hindy & Huang [943], Cuoco & Liu [508]). Yogo [1665] show that a model with
non-separable utility in non-durable and durable consumption can explain both the
cross-sectional variation in expected stock returns and the time variation in the
equity premium.
The empirical research on the presence of a habit persistence, durable goods
and intertemporal complementarity/substitutability of consumption is quite large.
Testing Euler conditions on monthly time series, Dunn & Singleton [602], Eichen-
baum & Hansen [633], Eichenbaum et al. [634], Gallant & Tauchen [753] find
evidence of the presence of durable good effects (substitutability over short periods)
rather than of habit persistence. On the other hand, limited evidence of a habit
formation effect has been detected in Dynan [613]. The analysis in Ferson &
Constantinides [689] and Braun et al. [282] on monthly, quarterly and annual
data shows that habit persistence effects dominate over durability effects. The
empirical evidence reported in Ferson & Harvey [692] leads to a rejection of time
separable preferences, even when the model incorporates seasonality and allows
for seasonal heteroskedasticity. A form of seasonal habit persistence, rather than
goods durability, is empirically detected. The habit persistence model performs
better than the classical model with a time separable utility function and is not
rejected by Euler condition tests. However, the equity premium puzzle cannot
be explained. The mean-variance bound on the intertemporal rate of substitu-
tion is more likely to be satisfied by a habit persistence preference functional
showing intertemporal complementarity than by a utility with durable goods
characterized by intertemporal substitution, see Hansen & Jagannathan [891], Gal-
lant et al. [750], Cecchetti et al. [380], Ferson & Harvey [692]. Consumption
durability helps to satisfy Euler conditions tests, while habit persistence helps
to match the first two moments of asset returns. Consumption durability lowers
the volatility of the marginal rate of substitution. Chen & Ludvigson [420] find
evidence of a (non-linear) internal habit rather than external habit (keeping up
with the Joneses preferences, see below). A model with habit persistence can
also explain a cross section of returns on size and book-market sorted portfolio
better than the CAPM and several of the classical multi-factor models discussed in
Sect. 5.3.
An empirical analysis of a continuous time representative agent economy with
non-separable preferences has been proposed in Heaton [923]. It is shown that
the empirical performance of an intertemporal consumption model depends on
the frequency of consumption data as well as on their aggregation. Monthly
observations of seasonally adjusted consumption and quarterly observations of
seasonally unadjusted consumption are at odds with the hypothesis that aggre-
gate consumption follows a martingale process. However, the martingale hypoth-
esis and its implications are typically stated under the assumption of a rep-
resentative agent economy with time separable preferences. In contrast, non-
separable preferences can account for different dynamics of consumption. Using
observations of consumption on non-durable goods, there is strong evidence in
favor of a model where consumption is substitutable over time (consumption
is durable). No evidence is found for habit persistence alone. Using seasonally

unadjusted data, there is also evidence for habit persistence at seasonal frequen-
cies.
A similar analysis has been carried out in Heaton [924], examining the asset
pricing implications of a representative agent economy with non-separable prefer-
ences. Evidence is found in favor of a preference structure where consumption at
nearby dates is substitutable (local substitution) and where habit over consump-
tion develops slowly. Moreover, there is evidence of long run habit persistence.
The two effects (local substitutability and long run habit persistence) have an
important interaction, in the sense that habit persistence substantially improves
the performance of the model only if local substitutability is also present. These
results suggest that there is consumption substitution over short periods and
consumption complementarity over longer periods. A preference structure of this
type is consistent with the Hansen & Jagannathan [891] bounds, generates a low
volatility of the risk free rate as well as positively autocorrelated returns but it is
rejected using a simulated methods of moments approach and cannot reproduce
the historical equity premium as well as the volatility of stock returns. A utility
function capturing local substitution (durability) over short periods and long run
complementarity (habit formation) has been also proposed in Hindy et al. [944] and
Detemple & Giannikos [562]. Habit and durability also provide an explanation of
the excess sensitivity of consumption to past income changes.
Keeping Up with the Joneses and External Habit
So far, we have considered preference functionals exhibiting an internal habit

formation, in the sense that the habit process .zt /tD0;1;:::;T is defined as in (9.36)
as a function of past consumption and, hence, is endogenously determined. In
Abel [4, 5] and Gali [749], preference functionals with an external habit process
have been proposed. Typically, the external habit process .zt /tD0;1;:::;T is assumed to
represent the aggregate consumption per capita (as in the case of the keeping up with
the Joneses preferences described below) or simply a fully exogenous process (see
Clark et al. [454] for a review).
The following preference functional has been proposed in Kocherlakota [1112]
(originally in an infinite horizon economy):
X
T

ct1˛
ıt E cN t cN t1 ; (9.42)
tD0
1˛
for ˛ 0 and ; < 1, where c D .ct /tD0;1;:::;T denotes as usual the consumption
process of an agent and cN D .Nct /tD0;1;:::;T represents the aggregate per capita
consumption in the economy. The parameters and model the agent’s evaluation
of his own consumption at each date t with respect to the per capita consumption
at the same date and the one-period lagged per capita consumption. In other words,
an agent with preferences (9.42) is concerned about how well he performs with
respect to the average agent in the economy. This utility function captures an
externality effect in consumption: if an agent is jealous, then and will be
negative (negative externality, the agent is unhappy when other individuals perform
better than him). On the other hand, if an individual is altruistic, then and will
be positive (positive externality). When solving the optimal consumption problem,
an individual takes the per capita aggregate consumption as an exogenous datum,
while in equilibrium the agent’s consumption will coincide with the per capita
consumption in a homogeneous economy. In this case, as shown in Kocherlakota
[1112], the Euler condition for the optimal consumption-investment problem can be
written as
" ˇ #
cN tC1 ˛ cN t n ˇ
E .rtC1 rf /ˇˇFt D 0;
cN t cN t1
for all n D 1; : : : ; N and t D 0; 1; : : : ; T 1. This condition can be derived along

the lines of Exercise 9.5 (compare also with Proposition 6.4).
In Abel [4], the author introduces a preference functional that nests three classes
of preferences: classical time separable utility, habit formation preferences and
external habit preferences. More specifically, Abel [4] considers the following
preference functional:
X
T
ı t EŒu.ct ; zt /; (9.43)
tD0
where c D .ct /tD0;1;:::;T denotes the consumption process and where the habit
process .zt /tD0;1;:::;T depends both on an external and on an internal habit:

ˇ 1ˇ
zt D ct1 cN t1 ; for all t D 1; : : : ; T; (9.44)
where 0 and ˇ 0. Clearly, if D 0, then (9.43) reduces to the classical

time additive utility function. If > 0 and ˇ D 0, then the habit process (9.44)
is purely external, since it only depends on the per capita aggregate consumption,
as considered in (9.42). Finally, if > 0 and ˇ D 1, then (9.43) reduces to the
preference functional (9.35) with an endogenous habit process. Abel [4] considers
a utility function of the form
.ct =zt /1˛

u.ct ; zt / D ; for all t D 0; 1; : : : ; T;
1˛
with ˛ > 0, corresponding to the ratio model discussed above. In the context of the
Abel [4] model, the Euler condition and the equilibrium price-dividend ratio of a
stock are derived in Exercise 9.9.
Considering a consumption process with i.i.d. growth rates calibrated to the U.S.
economy, Abel [4] shows that both the risk premium and the risk free rate observed
empirically can be reproduced with ˛ D 6 and ¤ 0 (non-separable preferences).
Similarly to the case of habit persistence preferences, the main empirical drawback
is that the model generates a return variance (in particular of the risk free rate)
which is much larger than that observed historically. In Abel [5], it is shown that
this drawback can be avoided by introducing a leverage factor in the wealth of the
economy. Setting D 0 in the preferences (9.42), Gali [749] shows that there exists
an inverse relation between and the risk premium, so that sufficiently negative
values of allow to solve the equity premium puzzle. However, large negative
values of also induce a large risk free rate.
Campbell & Cochrane [343] consider a utility function with an external habit,
where each agent compares his consumption with past aggregate consumption.
More specifically, the preference functional is defined by the utility function
.ct st /1˛ 1 ct zt
u.ct ; zt / D ; where st WD ; (9.45)
1˛ ct
for all t D 0; 1; : : : ; T and with ˛ 2 .0; 1/. The process .st /tD0;1;:::;T represents the
surplus consumption ratio and .zt /tD0;1;:::;T the external habit process. It is assumed
that the logarithm of the surplus process satisfies the following dynamics:
log stC1 D sN C '.log st sN/ C .st /.log ctC1 log ct g/; (9.46)
for all t D 0; 1; : : : ; T 1, where ', sN and g are given parameters. According

to (9.46), the logarithm of the surplus process fluctuates around the value sN, thus
capturing business cycle fluctuations. The function ./ is called the sensitivity
function and will be specified further below. The aggregate endowment process
(expressed as the average endowment across all the agents of the economy) satisfies
the following dynamics:
log ctC1 WD log ctC1 log ct D g C vtC1 ; (9.47)
where .vt /tD1;:::;T is a sequence of i.i.d. normal random variables with zero mean and
variance 2 . Note that, due to the specification (9.46)–(9.47), the surplus process is
perfectly correlated with the consumption growth, in the sense that
Cov. log stC1 ; log ctC1 / D 1; for all t D 0; 1; : : : ; T 1:
In equilibrium, identical individuals choose the same level of consumption. We

can then establish the following proposition, the proof of which is given in
Exercise 9.10.
Proposition 9.10 Consider the Campbell & Cochrane [343] model introduced
above, where the surplus process and the consumption growth process satisfy the
dynamics (9.46) and (9.47), respectively. In equilibrium, the intertemporal marginal

rate of substitution process .mt /tD1;:::;T is given by
mtC1 D ıe˛.gC.1C.st //vtC1 C.'1/.log st Ns// ; (9.48)
and it holds that
EŒmtC1 jFt p
p D e˛2 .1C.st //2 2 1 ˛ 1 C .st / ; (9.49)
Var.mtC1 jFt /
for all t D 0; 1; : : : ; T 1. Moreover, the equilibrium risk free rate rf is given by
1 ˛.gC.'1/.log st Ns//˛2 .1C.st //2 2

rf D e 2 ; for all t D 1; : : : ; T. (9.50)
ı
Note that mt corresponds to the one-period stochastic discount factor introduced
in Sect. 6.4. In particular, relation (9.49) represents the Hansen & Jagannathan
[891] bound for the maximal Sharpe ratio among all traded assets (compare with
Proposition 5.1). In the expression for the equilibrium risk free rate (which is a
priori time dependent), the term log st sN reflects intertemporal substitution, while
the last term appearing in the exponential in (9.50) represents a precautionary saving
term. As pointed out in Campbell & Cochrane [343], the approximation obtained in
(9.49) is helpful for the specification of the model. To generate time-varying Sharpe
ratios, the term .st / must vary with st and it seems desirable that the function ./
is decreasing. In Campbell & Cochrane [343], the function ./ is chosen such that
the equilibrium risk free rate is constant, the habit process is predetermined at the
steady state sN and habit moves non-negatively with consumption. This leads to the
specification
( p
eNs 1 2.log st sN/ 1; if log st smax ;
.st / D
0; otherwise;
where smax is given by smax WD sN C .1 e2Ns /=2 and sN D log C 12 log.˛=.1 '//.
With this specification of ./, formula (9.50) reduces to
1 ˛gC ˛ .'1/
rf D e 2 :
ı
The Campbell & Cochrane [343] model allows to reproduce a wide variety of
asset pricing phenomena. In particular, it is possible to reproduce the historically
observed risk free rate, the average excess stock return and its variance. In particular,
the model fits the equity premium without generating a high volatility in the risk
free rate, unlike most of the habit formation models, and generates high volatility.
Moreover, the model explains business cycle movements of asset prices and risk
premia (high risk premia during recessions and small risk premia during booms,
i.e., counter-cyclical variations in the Sharpe ratios) as well as the presence of

serial correlation in asset returns and predictability via the price-dividend ratio. The
rationale is that during a recession the consumption of an agent decreases towards
the level of the habit, thereby increasing risk aversion and the risk premium. This
type of utility function generates high and time-varying risk aversion. According to
this model, there is an inverse relation between consumption surplus and expected
returns. The model also generates a low correlation between asset returns and con-
sumption growth, as observed empirically. As pointed out in Campbell & Cochrane
[343], the main rationale for these implications is the presence of a slowly time-
varying, counter-cyclical risk premium. On asset pricing models under this type of
preferences see also Cochrane [463], Tallarini & Zhang [1582], Wachter [1635], Li
[1217], Yogo [1666]. As argued in Campbell & Cochrane [344], this type of model
may explain the poor performance of the CCAPM relative to the CAPM. The
model also performs well in explaining the cross-sectional variation in conditional
expected returns, showing that asset characteristics can play a role (growth premium
rather than value premium), see Santos & Veronesi [1498], Wachter [1635], Lettau
& Wachter [1201].
Time Preferences and Hyperbolic Discounting
In most of the analysis developed so far, we have assumed that multi-period prefer-
ences are modeled by assuming a constant discount factor that grows exponentially
with respect to time. However, experimental studies have shown that this way of
representing preferences over time is typically violated (see Frederick et al. [736]
for a review of time discounting and time preferences). In order to overcome
the shortcomings of this specification of the discount factor, alternative forms of
discounting have been proposed. In particular, hyperbolic discounting has been
introduced in order to capture the empirical fact that discount rates should be
declining with respect to time (see, e.g., Angeletos et al. [71]). Loewenstein & Prelec
[1248] enumerate a series of anomalous findings on time preferences and introduce
an hyperbolic discount function ı W RC ! RC of the following form:
ı.t/ D .1 C ˛t/ˇ=˛ ; for all t 0;
with ˛; ˇ > 0, thus yielding the preference functional
X
T
ı.t/EŒu.ct /;
tD0
where c D .ct /tD0;1;:::;T denotes as usual a consumption stream. This alternative

specification of time preferences is consistent with several empirical facts and with
intertemporal choice policies which are incompatible with the classical assumption
of exponential discounting. In Laibson [1156], a quasi-hyperbolic discounting

function has been proposed, leading to the preference functional:
" #
X
T
E u.c0 / C ˇ ı u.ct / :
t
tD1
When ˇ < 1, quasi-hyperbolic discounting mimics the qualitative properties of

hyperbolic discounting, while preserving the analytical tractability of exponential
discounting.
Non-exponential discounting introduces time inconsistency in optimal choice
problems: a decision taken at date t D 0 about something that is going to happen
at a future date t > 0 may be evaluated in a different way at a later date (e.g., at
date t D 1). This phenomenon, which is not observed in the classical setting of
exponential discounting, induces personal strategic conflicts, since a future option
may be valued in different ways as time goes. Gul & Pesendorfer [864, 865] provide
an axiomatization of this attitude in the context of a two-period model where an ex-
ante inferior choice may tempt the decision-maker in the second period (temptation
and self-control preferences). These axioms yield a representation that identifies the
agent’s commitment ranking, the temptation ranking and the cost of self-control.
By analysing household data, Huang et al. [974] have found an empirical support
for temptation and self-control preferences. A life-cycle model including temptation
and self-control preferences has been studied in Laibson et al. [1157]. Building on
Gul & Pesendorfer [864], Krusell et al. [1139] analyse a general equilibrium model
where a small subset of the agents exhibits a negative short run discount factor
(short run “urge to save”), at the same time avoiding time inconsistency. Agents
are fully rational, but are subject to temptation and self-control. It is shown that, in
equilibrium, the model can generate a low risk free rate together with a high risk
premium. However, DeJong & Ripoll [542] show that the self-control preferences
introduced by Gul & Pesendorfer [864] do not allow for a resolution of the asset
pricing puzzles.
9.3 Beyond Substantial Rationality
In the preceding sections of this chapter, we have focused our attention on how
agents behave in the presence of risk and uncertainty and we have introduced
alternative preference functionals that depart from the classical time additive
expected utility paradigm. However, we have always assumed that the agents’
behavior is characterized by substantial rationality, i.e., agents pursue their goals
in the most appropriate way under the constraints imposed by the environment, as
already discussed in Sect. 1.1. In a perfectly competitive economy, the assumption
of substantial rationality implies that the decision problem of each agent can be
represented by the maximization of a preference functional. In this section (as
9.3 Beyond Substantial Rationality 521
well as in the following one), we will depart from the substantial rationality
assumption, entering the domain of non (substantial) rationality. The notion of
non-rationality (as well as bounded rationality and near-rationality, as discussed in
the following section) means that agents do not maximize an objective functional
representing their preferences and/or that agents do not handle risk/uncertainty in an
optimal way (i.e., by updating their beliefs via Bayes’ rule). Alternative behavioral
assumptions are mainly motivated by two arguments: first, the assumption of
substantial rationality (together with the rational expectations hypothesis) seems
rather implausible to psychologists and economists; second, the hypothesis of
perfect rationality leads to very complex decision problems and, in some cases, no
explicit solution to an optimal choice problem exists. Moreover, as discussed in the
previous sections, there is a consistent experimental evidence reporting empirical
regularities that are incompatible with classical asset pricing theory. Based on this
observation, Simon [1545] proposed a satisfying approach: agents possess limited
computing skills and the inherent complexity of typical decision problems leads
agents to adopt what is improperly defined as a non-fully rational behavior (e.g.,
adopting rules of thumb or mental accounting rules).
We will address the impact of non-rationality in a setting with heterogeneous
agents. Note that heterogeneity can produce interesting results going beyond
classical asset pricing theory only if it is considered in conjunction with non-
rationality. Indeed, under suitable assumptions, if agents are rational and markets
are complete, then even under heterogeneity an aggregation result holds, so that the
prices of the economy with heterogeneous agents are observationally equivalent to
an economy with a single representative agent (compare with Sects. 1.3 and 4.3).
In what follows, heterogeneity will typically not refer to differences in the agents’
resources, preferences, or information, but rather in their goals, in their motives to
trade and in the way they pursue their goals.
Before presenting alternative asset pricing models based on a non-rational
behavior, we want to point out that classical asset pricing theory based on full
rationality (and rational expectations) will always play an important role, both
in a normative and in a descriptive sense. Indeed, classical asset pricing theory
represents the main reference point, because it describes what happens in the
economy when agents fully exploit the available information and act in an optimal
way. Alternative asset pricing theories will always be compared to classical asset
pricing theory: recalling Kuhn’s point of view (quoted at the beginning of Chap. 1),
alternative behavioral theories will be evaluated not only for their capability to
explain anomalies appearing within classical asset pricing theory but also for their
capability to provide alternative paradigms which can explain what classical asset
pricing theory is already able to explain.
Apart from the normative interpretation, Friedman [743] proposed two argu-
ments in favor of full rationality and rational expectations. The first argument is that
a theory should be evaluated on the validity of its implications and not necessarily
on the validity of the assumptions (this line of thought is well exemplified by the
two statements by Solow and Kuhn quoted at the beginning of Chap. 1). The second
argument is that, in a competitive market, non-rational agents will tend to perform
poorly when compared to rational agents, with the consequence that in the long
run non-rational agents will be wiped out of the market by rational agents. In this
section, we will carefully discuss both arguments. We do not aim at providing
an exhaustive survey on the topic, limiting ourselves to highlighting some of the
most important results. More precisely, we shall consider economies characterized
by the presence of the following types of non-rational agents: a) noise traders; b)
overconfident agents; c) feedback traders. At the end of the present section, we will
also discuss the possibility that non-rational agents survive in the long run and the
role of rational agents in stabilizing the market. This section is also related to the
analysis developed in Sect. 8.5 on the difference of opinions.
Noise Traders
De Long et al. [549, 550] and Shleifer & Summers [1540] analyse an economy with
noise traders. The underlying idea is that noise traders are characterized by incorrect
beliefs on the fundamental value of an asset and underestimate or overestimate
the future expected price of the asset. From the point of view of rational traders,
the presence of noise traders creates an additional risk in the economy and prices
can diverge significantly from fundamental values. In this way, several anomalies
appearing in the context of classical asset pricing theory can be addressed.
In De Long et al. [549], the authors consider an overlapping generations model
with agents living for two periods, with no consumption occurring in the first period
and where agents are endowed with exogenous resources to invest in the market.
Agents have only to choose a portfolio in the first period (when agents are “young”),
while in the second period (when agents are “old”) all the wealth generated by the
chosen portfolio will be consumed. There are two available assets. The first asset is a
safe asset yielding the dividend rf in each period and is available in perfectly elastic
supply, with a price fixed at one. The second asset is interpreted as the aggregate
equity available in the financial market and, since it is not in elastic supply, it is a
risky asset. The risky asset yields the constant dividend rf , equal to the dividend of
the safe asset, in each period, but is available in unitary fixed supply. We denote by
pt the price of the risky asset at date t, for t 2 N. Note that the safe asset and the
risky asset pay identical dividends.
If all agents were rational (so that the prices of both assets would be equal to
the net present value of the corresponding future dividends), then the safe and the
risky assets would be perfect substitutes and would be traded for a price equal to one
at every date. However, in the presence of noise traders, the equality pt D 1 does
not necessarily hold. De Long et al. [549] assume that the economy is populated
by two types of agents: there are sophisticated investors with rational expectations
and there are noise traders. It is assumed that noise traders represent a proportion
2 Œ0; 1 of the economy. All agents of a given type are identical and every agent
is allowed to take unlimited short positions in the assets when choosing a portfolio
in the first period. At each date t 2 N, every young agent chooses his portfolio by
maximizing an expected utility function given his own beliefs at date t and the ex-
ante distribution of the price ptC1 of the risky asset at the subsequent date t C 1.
All the agents know the economic model and are characterized by an exponential
utility function of the form u.x/ D e2 x , with 2 being the coefficient of absolute
risk aversion. The utility function of every agent is defined with respect to wealth
when the agent is old. The non-rational behavior of noise traders is modeled by
assuming that they misperceive the future expected price of the risky asset by a
random variable %t . More specifically, at date t a noise trader believes that the
expected future price of the risky asset is given by
EŒ ptC1 jFt C %t ;
where .Ft /t2N represents the information flow of the economy. It is assumed that the
sequence .%t /t2N is composed of i.i.d. random variables with common distribution
N .%;N %2 /. The mean value %N represents the average “bullishness” of noise traders,
while %2 represents the variance of noise traders’ misperceptions. On the contrary,
sophisticated investors correctly perceive the conditional distribution of the future
price.
In the context of the economy described above, we can characterize the equi-
librium price process . pt /t2N of the risky asset. Moreover, we can compute the
expected relative returns of noise and rational traders in equilibrium. This is the
content of the following proposition, the proof of which is given in Exercise 9.11.
Proposition 9.11 In the context of the above model (see De Long et al. [549]),
consider a steady-state equilibrium, where the unconditional distribution of the
equilibrium price pt does not depend on the date t 2 N. Then in a steady state
equilibrium the following hold:
(i) the price process . pt /t2N of the risky asset satisfies
.%t %/
N %N 2 2 %2
pt D 1 C C ; for all t 2 NI (9.51)
1 C rf rf rf .1 C rf /2
(ii) the expectation of the difference R between noise traders’ and rational
investors’ total returns (given equal initial wealth) is given by
.1 C rf /2 %N 2 C .1 C rf /2 %2
EŒ RtC1 D %N ; (9.52)
2%2
for all t 2 N.
As shown in Exercise 9.11, in comparison to the optimal demand of a rational
trader, the optimal demand of the risky asset by a noise trader contains an additional
term. This additional term is due to the noise trader’s misperception of the future
expected price of the risky asset: when noise traders overestimate expected returns,
they demand more of the risky asset than rational traders do. The variance appearing
in the optimal demand of rational and noise traders is purely due to the presence of
noise traders, since there is no aggregate risk in the economy.
The equilibrium price process . pt /t2N characterized in part (i) of Proposi-
tion 9.11 admits an interesting interpretation. Note first that, if the distribution of
%t converges to a point mass at zero, then the equilibrium price is constant and
always equal to one (the fundamental value), as previously mentioned. If this is not
the case, then the presence of noise traders affects the equilibrium price in three
distinct ways, each of them corresponding to one of the three terms appearing on
the right-hand side of (9.51) (and observe that the relevance of these three terms
increases as the proportion of noise traders increases):
• the term .%t %/=.1Cr
N f / captures the variability in the equilibrium price due to
the variations in the misperception of noise traders. Even though the risky asset
does not have any fundamental risk (since it delivers a constant dividend), its
price varies depending on the noise traders’ misperceptions of its future price. In
particular, if the young generation of noise traders at date t is characterized by a
misperception greater than its average (%t > %), N then this leads to an increase of
the equilibrium price;
• the term %=r
N f captures deviations of the equilibrium price from the fundamental
value of the asset due to the fact that the average misperception of noise traders is
different from zero (%N ¤ 0). If noise traders are bullish on average (%N > 0), then
this price pressure effect makes the equilibrium price of the risky asset increase;
• the last term 22 %2 =.rf .1 C rf /2 / plays a crucial role: because of the presence
of noise traders in the market, the future price of the risky asset is random and,
therefore, its return is risky. This fact induces rational and noise traders to require
a positive risk premium in order to invest in the risky asset and this risk premium
drives the price down and the return up, with the risk premium being proportional
to the conditional variance of the future price (see Exercise 9.11). Note that this
effect plays a role despite the fact that both rational and noise traders always hold
portfolios without any fundamental risk. The last term in (9.51) means that noise
traders create their own space in the market and their trading activity is the only
source of risk in the economy.
It has to be remarked that the results described above depend strongly on the
assumptions of the model, as pointed out in Loewenstein & Willard [1249]. In
particular, the assumptions that agents can take unlimited short positions, that
the risk free asset is in perfectly elastic supply and the two-period overlapping
generations structure are crucial for the main results of De Long et al. [549] to
hold. If rational agents live for two periods, then they care about tomorrow’s
price and require a risk premium in order to compensate the risk generated by
noise traders. On the contrary, if rational agents are long-lived (in particular,
if they have an infinite time horizon), then they will not care too much about
tomorrow price and they will adopt strategies (buy and hold strategies) aimed at
exploiting the mispricing opportunities created by noise traders. However, in a
mean-variance setting, Bhushan et al. [230] have shown that “trading myopia” is
neither a necessary nor a sufficient condition for noisy prices. They show that
a unique noisy equilibrium (in the sense of noise traders affecting prices) exists
only if agents are rational regarding volatility and irrational regarding expected
returns and noise traders are allowed to take unbounded positions. If these two
crucial hypotheses (which are satisfied in De Long et al. [549]) are not verified,
then multiple noise equilibria can coexist with a classical rational equilibrium.
Furthermore, noisy equilibria exist only on a subset of the parameter space, while
the rational equilibrium exists for all parameter values.
Let us now comment on part (ii) of Proposition 9.11. As already mentioned
in the first part of the present section, the presence of non-rational traders in the
economy has often been denied on the basis of an evolutionary argument (see
Friedman [743]): non-rational agents tend to have a poor investment performance, in
comparison to rational agents, and, therefore, only rational agents will survive and
dominate the economy in the long run. In other words, non-rational agents should
earn lower returns on average and, therefore, economic selection will make rational
agents prevail in the market. In the model of De Long et al. [549], this conjecture
is not necessarily true. Indeed, in correspondence of a steady-state equilibrium, the
expected difference in the returns on the optimal portfolios of noise and rational
traders is given by formula (9.52) in Proposition 9.11. This formula shows that a
necessary condition for the expected return of noise traders to be greater or equal
than that of rational agents is %N > 0. The economic intuition behind this fact is
rather simple: if %N > 0 then noise traders will take larger positions in the risky
asset and, therefore, they will earn a larger share of the rewards to risk bearing. On
the other hand, if %N < 0 then the expected return of noise traders is lower than
that of rational agents. The first term in the numerator on the right-hand side of
(9.52) captures the “price pressure” effect, i.e., the fact that the demand of bullish
noise traders drives up the equilibrium price of the risky asset, thus decreasing the
return to risk bearing and, therefore, the expected return difference EŒ RtC1 . The
second term in the numerator captures the fact that noise traders have the worst
possible market timing: they buy the risky asset when all the other noise traders
are buying it (high price and low return). The more variable %t is, the stronger
the bad timing effect is. This effect goes against the investment performance of
noise traders and reduces their expected return, thus supporting the evolutionary
conjecture of Friedman [743]. The denominator on the right-hand side of (9.52)
captures the fact that noise traders create their own space in the market and, as
the variability of the misperceptions of noise traders increases, the risk premium
increases. In order to exploit the mispricing created by the misperceptions of noise
traders, rational investors must bear a greater risk. Since rational investors are risk
averse, they reduce the extent to which they bet against noise traders as a response
to this increased risk. Summing up, the first term as well as the denominator on the
right-hand side of (9.52) lead to an increase of the expected return of noise traders,
while the numerator tends to a decrease. A priori neither of these two effects clearly
dominates on the other. The expected return difference EŒ RtC1 is quadratic in %, N
so that noise traders will have an expected return lower than that of rational agents
if on average they are bearish or if they are excessively bullish (for a large value of
%).
N On the contrary, for intermediate values of %, N the expected return of noise traders
can be greater than that of rational agents. However, the optimal expected utility of
noise traders is always less or equal than that of rational agents (see De Long et al.
[549, Section II.B]).
So far, we have considered a fixed proportion of noise traders. De Long
et al. [549] extend the basic model by introducing population dynamics. More
specifically, they assume that new generations of investors enter into the economy
and decide on which investment strategy to follow (rational or non-rational)
depending on the past performance of those strategies, without accurately assessing
the ex-ante risks involved in each strategy. In this context, the impact of noise traders
on market prices does not necessarily diminish over time. In De Long et al. [549,
Section III.A], the authors introduce the following population dynamics for the noise
traders, based on imitation among agents:
˚
tC1 D max 0; minf1; t C
t .rtn rtr /g ; for all t 2 N; (9.53)
where rtn and rtr denote respectively the wealth returns of noise and rational traders
at date t and where
t represents the rate at which new investors entering into the
economy choose to be noise traders (per unit difference in realized returns), with
limt!1
t D 0. According to the population dynamics (9.53), the proportion of
noise traders increases if they overperform rational investors. The dynamics can be
interpreted as the result of an imitation process, in the sense that the most profitable
behavior (between being rational or being non-rational) tends to be adopted by new
agents entering the market (success breeds imitation), with some inertia represented
by the parameter
t .
The model introduced above with the population dynamics (9.53) can be easily
solved if
t is close to zero. In this case, returns can be computed under the hypoth-
esis that the proportion of noise traders in the economy remains approximately
constant over two consecutive dates. Formula (9.51) under the population dynamics
(9.53), with changed to t , characterizes the equilibrium price of the risky asset in
the limit as
t converges to zero. Relation (9.52) (in the conditional expectation form,
as given in Exercise 9.11), with changed to t and in the limit as
t converges to
zero, has an analogous interpretation and can be stated as follows:
.1 C rf /2 %2t
EŒ RtC1 jFt D %t : (9.54)
2t %2
Over time, the proportion t of noise traders will grow or shrink according to
whether the right-hand side of (9.54) is greater or lower than zero. Equivalently,
if t is below the critical value
.%N 2 C %2 /2 .1 C rf /2
WD ;
N %2
2%
then t tends to shrink. On the contrary, if t > , then noise traders induce
so much risk to make rational investors reluctant to speculate against them, with the
consequence that the proportion of noise traders grows and rational traders disappear
in the long run.
The model is further extended in De Long et al. [549, Section III.B] considering,
together with the population dynamics (9.53), also the possibility that the risky asset
delivers at each date t 2 N a random dividend equal to rf C "Qt , where .Q"t /t2N is a
sequence of i.i.d. normal random variables with zero mean and constant variance,
independent of the noise traders’ misperceptions .%t /t2N . In particular, as shown in
Exercise 9.12, in correspondence of a steady-state equilibrium and in the limit as
t
converges to zero, the difference between the expected return of noise and rational
traders can be greater than zero and the “create space” effect is more pronounced,
as rational investors are less willing to trade in the risky asset, due to the increased
riskiness. In the long run, either noise traders dominate the economy or t converges
(in expectation) to an equilibrium where both types of agents coexist. On the other
hand, if the imitation process is based on the comparison of the utility obtained at
the previous date, then in the long run noise traders disappear from the economy.
The model of De Long et al. [549] provides several insights that allow us to
address some of the anomalies encountered in the context of classical asset pricing
theory. First, noise traders induce an excess variability of the price process . pt /t2N
with respect to what would be implied by a fully rational equilibrium. Moreover,
the volatility of market prices is not due to variations in the asset fundamentals,
but only to the presence of noise traders. Hence, the introduction of noise traders
in the economy helps to explain the excess volatility phenomenon discussed in
Sect. 7.2. Furthermore, as considered in De Long et al. [549, Section IV.A], if
the noise traders’ misperceptions .%t /t2N follow a first-order autoregressive process
instead of being a sequence of i.i.d. random variables, then the risky asset price will
exhibit mean reversion. In this setting, a contrarian strategy can earn excess profits.
Moreover, if misperceptions are serially correlated, then noise traders can cause
larger deviations of asset prices from the fundamental values. Finally, the model
can also shed some light on the equity premium puzzle (see Sect. 7.3), since the
presence of noise trader risk drives down the price of the risky asset and, therefore,
the risky asset yields a higher return than the risk free asset.
In De Long et al. [550], in the context of a slightly different model, the dynamics
of the population is analysed. The economy is assumed to be populated by noise
and rational traders, but noise traders do not affect market prices. Noise traders
misperceive the distribution of the asset return (in particular, they misperceive its
mean and variance). It is shown that, if noise traders only misperceive the variance,
then they earn expected returns higher than those of rational traders (see De Long
et al. [550, Proposition 1]). Moreover, if the noise traders’ utility function exhibits
a larger risk aversion than a logarithmic utility function and they misperceive the
variance in a limited way, then noise traders survive in the market, in the sense that
the share of the aggregate wealth held by noise traders does not converge almost
surely to zero as time passes. Under some conditions, noise traders even dominate
the market, meaning that after a sufficient time the probability that noise traders hold
a share of the aggregate wealth larger than that of rational traders is greater than 1=2
(see De Long et al. [550, Proposition 2]). On the other hand, if noise traders strongly
underestimate the variance, then they are more likely to default.
In De Long et al. [547], the welfare consequences of the presence of noise
traders are evaluated in the context of an overlapping generations model. It is
shown that the presence of noise traders typically reduces the stock of capital and
consumption in comparison to an economy populated only by rational agents and
that the consumption process is more volatile. The driving force behind these results
is the risk generated by the presence of noise traders. In many cases, the wealth
of rational agents is negatively affected by the presence of noise traders and the
benefits generated by profitable trading opportunities for skillful rational traders
are not compensated by the costs of bearing the additional risk generated by noise
trading. In particular, the welfare costs of noise trading can be significant if the
magnitude of noise in aggregate stock prices is large.
Overconfidence
As we have explained above, models allowing for the presence of noise traders allow
to explain mean reversion phenomena in asset prices. However, models with noise
traders typically fail to explain the momentum effect (see Sect. 7.2). In the literature,
models with overconfident agents have also been considered and, in particular, it has
been shown that these models allow to generate both overreaction and underreaction
of asset prices, thus addressing the momentum effect.
Overconfidence can account for many different behavioral patterns: agents tend
to overestimate the precision of their information, agents tend to overestimate their
ability to perform well and overestimate their own role when they have obtained a
positive performance in the past, agents consider themselves above the average of
the population as well as several other related phenomena. In the financial literature,
it is typically assumed that agents are overconfident about information, i.e., agents
overestimate the precision of the information available to them. More specifically,
agents tend to overestimate the precision of their private information, while they
are not overconfident concerning the publicly available information (see Odean
[1379] and Daniel et al. [516]). Moreover, agents can also be overconfident on
their knowledge of the private information of the other agents, as considered for
instance in Benos [193]. Overconfidence dynamically changes over time, as it tends
to increase after a positive performance (see Daniel et al. [516] and Gervais & Odean
[773]). In these models, the posterior beliefs of overconfident agents are too precise
and overweight (underweight, respectively) private information (public information,
respectively) in comparison to a rational Bayesian updating. As a consequence,
overconfidence leads to suboptimal portfolio choices and, since posterior beliefs
are excessively precise, overconfident agents tend to exhibit a larger risk exposure
in comparison to rational agents.
A model based on overconfidence and variations in confidence arising from

biased self-attribution of investment outcomes has been proposed in Daniel et al.
[516]. Agents are overconfident about their private information and, therefore,
they overreact to private information and underreact to public information. This
overreaction-underreaction phenomenon can generate positive serial correlation of
returns in the short run together with negative serial correlation in the long run
(agents overreact to their private information and then gradually revise their beliefs
according to the public information) and excess volatility. Biased self-attribution of
investment outcomes means that agents are biased in updating their confidence on
their own ability: they tend to attribute their success to their own ability and, on
the contrary, bad events to external forces. When agents receive a public signal
that confirms the validity of their actions, then the agents’ confidence rises. On
the contrary, the agents’ confidence is not affected by a public signal that does not
confirm their actions. This process is consistent with a momentum effect in the short
run together with long run reversals in asset prices.
Let us present in more detail the model of Daniel et al. [516], first considering
the case of a static confidence (meaning that overconfidence does not change over
time). The economy is populated by two classes of agents: uninformed risk averse
agents and informed risk neutral agents. Apart from the different attitude towards
risk, informed agents differ from uninformed agents since they can observe a private
signal. Each individual is endowed with a basket of security shares and units of the
risk free asset paying one unit of wealth at the terminal date. The model consists of
dates t 2 f0; 1; 2; 3g. At the initial date t D 0, agents start with their endowments,
identical prior beliefs and trade only for risk sharing purposes. At date t D 1, the
informed agents observe a common noisy private signal about the fundamental value
of the security and they trade with uninformed agents. At date t D 2, everyone
observes a noisy public signal and further trade occurs. Finally, at the terminal date
t D 3, all the uncertainty is resolved, the security pays its liquidating dividend and
every agent consumes. All the random variables appearing in the model are assumed
to be mutually independent and normally distributed.
The dividend delivered by the risky security at the terminal date t D 3 is
represented by the random variable dQ which is supposed to be a normal random
variable with mean dN and variance d2 (without loss of generality, we shall assume
from now on that dN D 0). The private signal observed by the informed agents at date
t D 1 is a realization of
sQ1 D dQ C ";
Q
where "Q N .0; "2 /. Uninformed agents correctly estimate the variance of the
noise component "Q (even though they do not observe the signal), whereas informed
agents underestimate it believing that the variance of "Q is given by c2 , with c2 <
"2 . The different beliefs about the variance of the noise of the signal are common
knowledge. The public signal released at date t D 2 is a realization of
sQ2 D dQ C ;
Q
where N .0; 2 /. Concerning the variance of the noise component of the public
signal, every agent believes in the correct variance 2 . In the model, it is essential
that some noisy public information is released after the private signal.
The following proposition, the proof of which is given in Exercise 9.13, collects
some fundamental properties of the Daniel et al. [516] model described above.
Proposition 9.12 In the context of the above model (see Daniel et al. [516]), the
equilibrium price of the risky security is given by
d2
p1 D .dQ C "Q/;
d2 C c2
(9.55)
d2 .c2 C 2 / d2 2 2 2
p2 D dQ C "Q C d c ;Q
D D D
Q where D WD d2 .c2 C 2 / C c2 2 . As a consequence, it holds that
and p3 D d,
Cov. p3 p2 ; p2 p1 / > 0;

Cov. p2 p1 ; p1 p0 / < 0;
(9.56)
Cov. p3 p1 ; p1 p0 / < 0;
Cov. p3 p2 ; p1 p0 / < 0:
As shown in Exercise 9.13, due to the assumption of risk neutral informed agents,
the equilibrium price p1 is given by the expectation of the dividend dQ conditionally
on the information received by the informed agents at date t D 1, whereas the
equilibrium price p2 is given by the expectation of the dividend dQ conditionally on
the information received by the informed agents at date t D 1 and the public signal
released at t D 2. Moreover, formula (9.55) shows the impact of overconfidence
on the dynamics of the equilibrium price. The overconfidence of informed agents
in the precision of their private signal makes the equilibrium price at date t D 1
overreact to the information conveyed by the observation of dQ C "Q. This overreaction
is partially corrected at the following date t D 2 when the public signal Q is
released and fully corrected at the terminal date t D 3, when the fundamental value
is revealed. The effect of the correction to the initial overreaction is captured by
the covariances given in (9.56). Indeed, the overconfidence of informed investors
induces a positive correlation between the price change in response to the arrival of
public information and later price changes, while the price change resulting from
the arrival of private information is on average partially reversed in the long run.
This latter property can be seen by noting that the price changes in the first two
periods are negatively correlated, while the price changes in the last two periods
are positively correlated (see Proposition 9.12). Arguing as in Daniel et al. [516],
if the econometrician is equally likely to pick either pair of consecutive dates, then
the overall autocorrelation is negative. Moreover, in the presence of overconfidence,
the volatility induced by private information is higher than the volatility that would
be observed in a fully rational economy, whereas the volatility due to the arrival of
public information can be higher or lower than the corresponding volatility in a fully
rational economy (see Daniel et al. [516, Proposition 3]).
In the setting described above, overconfidence accounts for long run reversals in
asset prices (negative long lag serial correlation) but not for short run momentum
effects (positive short lag serial correlation). In order to overcome this limitation,
Daniel et al. [516] propose an extended version of the model where the degree of
overconfidence is allowed to change in response to previous actions and outcomes
(biased self-attribution). In this version of the model, the public signal released at
date t D 2 is assumed to be of the form
(
C1; with probability p I
sQ2 D
1; with probability 1 p:
It is allowed (but not necessarily required) that informed investors are initially
overconfident, meaning that c2 "2 . From the perspective of an informed trader,
the public signal observed at date t D 2 confirms his beliefs if
sign.d C "/ D sign.s2 /;
with d C " and s2 denoting the realizations of the random variables dQ C "Q and sQ2 ,
respectively. In this case, the public signal makes the confidence of an informed
investor increase, leading to an estimation of the variance of the noise component
equal to c2 k, for some k 2 .0; c2 /. On the contrary, if
sign.d C "/ ¤ sign.s2 /;
then the public signal does not confirm the beliefs of the informed agent and his
confidence remains constant at the level c2 . In other words, the confidence of the
informed trader on the precision of his private information depends on whether the
public signal released at date t D 2 confirms or not his beliefs.
Assuming an outcome dependent overconfidence, we can establish the following
version of Proposition 9.12 (the proof is given in Exercise 9.14).
Proposition 9.13 In the context of the above model with outcome dependent
overconfidence (see Daniel et al. [516, Section III]), the equilibrium price of the
risky security is given by
d2
p1 D .dQ C "/;
Q
d2 C c2
8 (9.57)
 0;

Cov. p3 p1 ; p1 p0 / < 0; (9.58)
Cov. p3 p2 ; p2 p1 / < 0:
The fact that Cov. p2 p1 ; p1 p0 / > 0 shows that the overreaction phase, and
not only the correction phase, can contribute positively to short term momentum,
yielding positively serially correlated returns in the short run. In particular, positive
autocorrelation is due to continuing (delayed) overreaction and this allows to explain
the profitability of momentum strategies. The empirical analysis in Jegadeesh &
Titman [1028] provides support to this interpretation (however, difficulties arise in
producing a mean reversion effect in the long run, see Fama [664]). The Daniel
et al. [516] model also allows for an explanation of post-event continuation (e.g.,
post-earnings announcement drift, see Chap. 7).
A model similar to the one presented above with overconfident informed
investors allows to address several cross-sectional anomalies discovered when
testing the CAPM, as shown in Daniel et al. [517]. In the latter paper, the authors
consider a model with static overconfidence in a single period. Risk averse investors
use their information incorrectly when choosing their investment strategies and,
as a consequence, in equilibrium securities are mispriced. In turn, this implies
that variables that proxy for the mispricing are informative about future returns.
According to this perspective, it is shown that overconfidence allows to rationalize
the ability of the dividend yield, size, earnings=price and book to market value
ratio to predict cross-sectional differences in returns. Moreover, these variables
can dominate the ˇ coefficient in explaining the cross section of expected returns.
The rationale is that overconfidence leads to a misvaluation which is reflected in
market prices and, hence, in the above variables. Misvaluation due to overreaction
to favorable news leads to a high price, low ratios and low future returns, whereas
underreaction implies the opposite relationship. It is important to remark that this
perspective does not provide a risk based explanation of CAPM anomalies, but
rather it represents a departure from classical asset pricing theory. Note also that
rational investors exploit the pricing errors induced by overconfident agents, but
they cannot fully eliminate the mispricing because of risk aversion.
A variant of the above model has been considered in Daniel & Titman [521]: in
this model agents are overconfident about intangible information (e.g., management
decisions’ information) and not about tangible information (e.g., firm performance
information). Therefore, agents tend to overreact to intangible private information
and to underreact to tangible public information. The implications of these models
are confirmed empirically and the book to market value ratio is shown to be a good
proxy of the intangible return and, therefore, it allows to forecast future returns.
Odean [1379] analyses some models with private information, assuming that
agents are rational in all respects except how they process information. Odean
[1379] considers both a perfectly competitive market with noisy supply and dissem-
inated information (as considered for instance in Diamond & Verrecchia [569] and
Hellwig [931]) as well as a market with strategic insider trading and concentrated
information (as considered for instance in Kyle [1147]). Coherently with the
Daniel et al. [516] model presented above, agents overweight their private signals
and underweight market price signals. In a perfectly competitive market, trading
volume and volatility are increasing with respect to the agents’ overconfidence.
Furthermore, overconfidence worsens the informational quality of market prices
and the expected utility of overconfident agents is lower than that of rational
agents (and, typically, overconfident agents exhibit less diversified portfolios). If
overconfident agents underestimate (overestimate, resp.) new information, then
price changes exhibit positive (negative, resp.) serial correlation and underreaction
(overreaction, resp.). The degree of under/overreaction depends on the fraction of
traders who under/over weight information. In a strategic market model with an
insider, trading volume, market depth and volatility increase with respect to the
insider’s overconfidence (see also Benos [193] and Deaves [531] for experimental
evidence on the market effects of overconfidence). In particular, the effect on the
trading volume is the most robust effect of overconfidence. Overconfident insiders
improve price quality, but overconfident price takers worsen it. Insider expected
profits decrease in his overconfidence. However, in the context of a similar model,
Kyle & Wang [1150] show that an overconfident trader may get higher expected
profits-utility than a rational trader and, as a consequence, overconfident agents may
persist and survive in the long run in a market with a market maker.
An analogous result is obtained in Hirshleifer & Luo [946] in the context of
a perfectly competitive market: overconfident traders taking more risks perform
better than rational traders at exploiting the mispricing induced by liquidity and
noise traders. Assuming an imitation based population dynamics of the form
(9.53), overconfident agents are shown to represent a substantial proportion of
the population in correspondence of the long run steady-state equilibrium (similar
results are also obtained in Fischer & Verrecchia [705]). In Hirshleifer & Luo [946,
Proposition 1], it is shown that the higher the volatility of the underlying security
payoff is, the higher is the proportion of overconfident traders in equilibrium and,
on the contrary, the greater the confidence of the overconfident traders is, the lower
is the proportion of them surviving in equilibrium. Moreover, the more volatile
is liquidity/noise trading, the higher the proportion of overconfident traders is. In
a multi-period economy, overconfidence can generate disagreement among agents
(see Scheinkman & Xiong [1502]) and disagreement can lead to a bubble (asset
prices are above the corresponding fundamentals, see Sect. 6.5), trading volume,
excess volatility and return predictability (see also Dumas et al. [601]).
In several models, it has been shown that overconfidence leads to a significant
trading volume and to suboptimal portfolio choices by overconfident investors and,
therefore, to lower levels of expected utility (see Biais et al. [238] for experimental
evidence). These implications have been empirically tested with positive evidence,
see Chuang & Lee [447], Glaser & Weber [786], Grinblatt & Keloharju [831]. In
Barber & Odean [143, 145] it is shown that households who trade frequently earn
an annualized geometric return smaller than that of similar households trading less
frequently. In this context, overconfidence can explain the high trading levels and
the resulting poor performance of individual investors. Psychological research has
shown that women are typically less overconfident than men: supporting this fact,
Barber & Odean [144] and Graham et al. [819] show that men trade more frequently
and that their average return is smaller than the average return obtained by women.
In Odean [1381], it is shown that securities purchased by small traders underperform
securities sold by the same agents, meaning that small investors trade too much and
in the wrong direction. Note also that the tendency of typical traders to hold losers
and to sell winners can be explained in terms of overconfidence. With biased self-
attribution, the level of investor overconfidence and, hence, trading volume varies
with past returns. Statman et al. [1565] test the trading volume predictions of formal
overconfidence models and find that the share turnover is positively related to lagged
returns.
We have pointed out that overconfident agents may survive in the long run,
since under suitable assumptions they can earn a higher utility than rational
agents. Gervais & Odean [773] analyse patterns in trading volume, volatility and
expected prices as well as the profits resulting from endogenous overconfidence.
In this setting, agents may learn to become overconfident by overestimating the
degree to which they are responsible for their own success (self-serving attribution
bias). There is an endogenous tendency to overconfidence: overconfidence does
not make traders wealthy, but the process of becoming wealthy can make traders
overconfident (in other words, traders take too much credit for their own success).
Wang [1641] examines the survival of non-rational investors in an evolutionary
game: underconfidence or pessimism cannot survive, but moderate overconfidence
or optimism can survive and even dominate the economy, especially when the
fundamental risk is significant.
Feedback Traders
Price dynamics in the presence of feedback traders have been analysed in Cutler
et al. [509] and in De Long et al. [548]. The main feature of economies with
feedback traders is that such agents buy or sell an asset on the basis of its past
returns, rather than on the basis of the expectation of future fundamentals. More
specifically, positive (negative, respectively) feedback traders buy (sell, respec-
tively) an asset after a run up of its price. Positive feedback trading could result from
many different phenomena, including stop loss orders, portfolio insurance strategies,
technical analysis rules, risk aversion decreasing in wealth, positive wealth elasticity
of the demand of risky assets, contagion among the agents in the market, asset price
extrapolation, trend chasing. On the other hand, negative feedback trading can be the
outcome of profit taking strategies or strategies targeting a constant share of wealth
in different assets.
In Cutler et al. [509] it has been shown that the introduction of feedback traders
allows to generate positive serial correlation in asset returns over short horizons
as well as negative serial correlation over longer horizons. The authors consider a
market for a future contract, with a well defined fundamental value which is equal to
the terminal value of the contract and without dividend payments. It is assumed that
the asset is in zero net supply. The economy is populated by three types of agents:
rational traders, fundamental traders and feedback traders. The model is linear and
it is assumed that the proportion of the three different types of agents in the economy
is the same and does not change over time.
Rational traders invest in the asset on the basis of rational forecasts of its future
returns, holding a higher proportion of the asset when expected returns are high.
Denoting by tC1 r
the quantity of the asset held by the rational agents in the period
Œt; t C 1, it is assumed that

tC1
r
D EŒrtC1 jFt ; (9.59)
for some constant > 0, with rtC1 and denoting respectively the return of the asset
on the period Œt; t C 1 and the required return for holding the asset, respectively. In
particular, if the expected return equals the required rate of return, then rational
agents do not trade. The rational agents’ demand is linear in the expected excess
return of the asset.
Fundamental traders trade on the basis of the expected returns with respect to
some perceived fundamentals: when prices are high relative to perceived funda-
mentals, then demand is low. This approach is represented by the investment rule

tC1
f
D ˇ pt ˛.L/ft ; (9.60)
where tC1
f
denotes the quantity of the asset held by fundamentalist traders in the
period Œt; t C 1, ˇ < 0 and where ˛.L/ represents a delay polynomial operator
defining the weights associated to the previous fundamental values, with ˛.1/ D 1.
The process . ft /t2N represents the fundamental value of the asset and is assumed to
follow a random walk, i.e.,
ftC1 D ft C "QtC1 ; for all t 2 N;
where .Q"t /t2N is a sequence of i.i.d. random variables with zero mean and constant
variance. In (9.60), we allow ˛.L/ to be different from one, thus allowing for the
possibility that perceived fundamentals reflect true fundamentals with a time lag.
Finally, there are feedback traders, who base their demand on the past returns
of the asset. More specifically, denoting by tC1 fb
the quantity of the asset held by
feedback traders in the period Œt; t C 1, it is assumed that
tC1
fb
D ı.L/.rt %/; (9.61)
where ı.L/ is a delay polynomial operator, where the sign of the weights defining it
characterizes the feedback trader as a positive or a negative feedback trader.
Imposing the equilibrium condition
tr C tf C tfb D 0; for all t 2 N;
assuming a constant required return % equal to zero and making use of (9.59)–(9.61),
we obtain the following rational expectations difference equation for the equilibrium
price process . pt /t2N of the asset:
ˇ 1
EŒ ptC1 jFt pt D pt ˛.L/ft ı.L/. pt pt1 /; for all t 2 N:

As explained in Cutler et al. [509], solving this equation under the rational
expectations hypothesis (and considering the fundamental solution, compare with
Sect. 6.5), the equilibrium price of the asset can be expressed as a function of
past prices, expected future fundamentals and past fundamentals. Furthermore, the
fundamental innovations .Q"t /t2N are fully reflected in equilibrium prices.
The model illustrated above can generate positive serial correlation in asset
returns in three different ways:
• first, if fundamental traders learn about the true fundamental value of the asset
with a delay (i.e., if ˛.L/ ¤ 1), then, in the absence of noise traders (ı.L/ D 0),
asset returns will be positively serially correlated for as many periods as it takes
for the information on the asset fundamentals to be incorporated in the demand
of fundamental traders;
• second, negative feedback traders (i.e., ı.L/ < 0) are a source of positive serial
correlation. Indeed, assuming that there are no fundamental traders (ˇ D 0), if
the asset price increases (positive return), then negative feedback traders reduce
their demand of the asset. This implies that in equilibrium rational agents have to
hold a larger share of the asset and this can happen only if the expected return is
higher than in the past. Therefore, on average successive returns will be positively
serially correlated;
• finally, positive serial correlation can be due to the presence of feedback traders,
whose demand depends on the asset returns on several previous periods. Hence,
if the asset returns in one period affect the demand of feedback traders for
many subsequent periods, feedback traders will persistently exhibit long or short
positions. In turn, this implies that required returns for rational investors will be
above or below the average for several periods, respectively, and this induces a
positive serial correlation in asset returns. Note that this effect can occur also in
the absence of fundamental traders.
Observe also that, in the first two cases, the initial reaction of the asset price to
the news is incomplete. Positive feedback traders with long memory induce price
overreaction to fundamental news and in the long run returns will be negatively
serially correlated. A model of this type with fundamental and feedback traders
provides a rationale for the presence of serial correlation in asset returns at different
horizons by acting on two key factors: delay in the incorporation of fundamental
news in the demand of fundamental traders and feedback trading effects (see also
De Long et al. [548] for a related model).
A similar setting has been analysed in Hong & Stein [961] by considering two
classes of agents with limited rationality: “newswatchers” and “momentum traders”.
Traders of the first type make forecasts based on private information without taking
into account market prices (i.e., newswatchers fail to extract other newswatchers’
information from market prices), while momentum traders use only past prices
in order to form their strategies. Neither type of agent is fully rational and infor-
mation gradually diffuses among agents. The first class induces an underreaction
phenomenon, in the sense that firm-specific information is incorporated into prices
only gradually. On the other hand, momentum traders induce an overreaction effect,
since their trading strategy conditioned on past prices is similar to that of positive
feedback traders. Momentum traders try to exploit the underreaction generated
by newswatchers but are only able to partially eliminate the mispricing and in
doing so they create an excessive momentum in prices that will ultimately lead
to overreaction. In line with the implications of the model presented above, price
changes are positively correlated over short horizons and negatively correlated over
longer horizons. This model has been tested with positive evidence in Hong et al.
[958]. Sentana & Wadhwani [1515] find that when volatility is low returns are
positively autocorrelated and, on the contrary, when volatility is high returns exhibit
negative autocorrelation. This suggests that positive feedback traders have a greater
influence on the price when the volatility is high, while negative feedback traders
dominate when the volatility is low.
Heterogeneous Agents, Market Selection and Speculation
Concerning the assumption of rationality, two conjectures have been made in the
literature: i) rational agents having more sophisticated information perform better
than less informed agents; ii) the speculation activity of rational agents helps to
stabilize the market. The first conjecture can also be extended to an economy
populated by both rational and non-rational agents, since it can be conjectured
that non-rational agents earn less profits than rational agents and, hence, they are
more likely to default. As a consequence, in the long run only rational and well
informed agents should survive. These conjectures have been formulated by many
authors, notably in Friedman [743] (Friedman’s conjecture). The first conjecture
is based on an evolutionary argument, in the sense that the market selects the
best agents (the better informed agents and/or the rational agents). Of course,
the validity of these two conjectures depends on the characteristics and on the
population of the economy. If every agent is rational, then agents having access
to better information should perform better than less informed agents (compare
with Sect. 8.1). Furthermore, rational agents exploit mispricings and, therefore,
their trading activity stabilizes market prices around the corresponding fundamental
values. However, in an economy populated by heterogeneous agents, the two
conjectures are not necessarily valid. For instance, we have already seen in this
section that noise traders are not necessarily eliminated from the market in the
long run. The evolution of the population of an economy has been analysed by
considering the wealth dynamics of different types of agents or the evolution of the
agents’ trading behavior via methods developed in the evolutionary game literature.
For instance, as considered in (9.53), agents tend to imitate the trading strategy
which has better performed in the past.
In Figlewski [701], the validity of the first conjecture has been discussed and
related to the efficiency of the market (in a strong form). The author considers
an economy with two types of agents, which can differ in price expectations,
risk aversion, predictive ability and wealth. Each of the two groups of agents
is composed of identical individuals. The two types of agents receive different
information and their prediction error (with respect to the fundamental value of the
asset) is normally distributed, with zero mean and a variance which depends on the
type of agent. It is assumed that every agent determines his demand by maximizing
a mean-variance utility function. Figlewski [701] shows that in the short run the
market tends towards efficiency, but perfect efficiency is not reached, not even in
the long run. Indeed, in the long run the economy exhibits deviations from the
distribution of wealth that would prevail in a fully efficient market. In the long run,
better informed agents hold more wealth than less informed agents (see Figlewski
[701, Table 1]). A model exhibiting convergence towards market efficiency has been
proposed in Luo [1254].
As already mentioned, non-rational (noise or overconfident) agents are not
necessarily eliminated from the market but can survive in the long run and may
earn higher returns than rational agents. In some cases, non-rational agents can
even dominate the market. The results obtained in Biais & Shadur [241] are along
these lines, thus providing a counter-argument to Friedman’s conjecture. Biais &
Shadur [241] consider a market with rational agents and non-rational overoptimistic
sellers (or pessimistic buyers), with the dynamics of the population depending on the
realized utility. It is shown that, in the long run, non-rational agents are concentrated
on one side of the market generating under or overpricing. Moreover, the survival of
non-rational traders reduces the opportunities for profitable risk sharing and induces
a welfare loss for the economy. This model also offers an alternative interpretation
of bubbles observed in speculative markets suggesting that, even though bubbles
can be generated by investors’ irrationality, they fail to be corrected by the forces
of market selection. In a similar direction, it has been shown in Blume & Easley
[255] and Shefrin & Statman [1532] that non-rational agents (i.e., agents who do
not update their beliefs according to Bayes’ rule) may survive in the market.
On the other hand, Blume & Easley [256] have provided a formal proof of
the market selection hypothesis, see also Sandroni [1495]. The authors consider a
general equilibrium model in infinite horizon and show that, if markets are complete,
assuming that the aggregate endowment is bounded from above and away from zero
and controlling for the discount factor, then only the traders with correct beliefs can
survive in the long run. In other words, the market selects those agents who update
their beliefs according to Bayes’ rule and for whom the true distribution belongs to
the support of the prior distributions. Moreover, agents’ preferences (risk aversion)
do not affect the long run fate of traders, only the discount factor matters: a low
discount rate leads to a high saving rate and, hence, a large consumption in the
future. Therefore, irrational agents with a discount rate lower than that of rational
agents may survive in the long run: due to incorrect beliefs, irrational agents take
bad investment decisions but nevertheless they save more, thus consume more in the
future and in the long run they may even survive. However, this result can fail when
markets are incomplete, when consumption grows too quickly, or when discount
factors and beliefs are correlated. In particular, the boundedness of the aggregate
endowment represents a strong assumption. Indeed, assuming an economy with an
unbounded growth rate, the market selection hypothesis does not hold in general.
Kogan et al. [1116] have shown that the market selection hypothesis does not
hold under the assumption that the aggregate endowment evolves as a geometric
Brownian motion and that irrational traders have incorrect opinions on the drift of
the endowment process. In this case, the expected aggregate endowment grows as
an exponentially linear function and agents disagree on the coefficient appearing in
the exponential. Kogan et al. [1116, Proposition 4] shows that moderately optimistic
irrational agents (who overestimate the rate of growth of the aggregate endowment)
may dominate the market in the long run and rational traders may disappear. In
a related context, Yan [1663] has shown that irrational agents with a sufficiently
low discount rate (or a sufficiently high elasticity of intertemporal substitution) save
more and are more likely to survive in the future. For an analysis in a mean reverting
environment we refer to Barucci & Casna [170]. The survivorship of irrational
agents has been confirmed in a bounded rationality framework in Berrada [207].
In the absence of full rationality, agents typically adopt a “rule of thumb” when
forming forecasts. This behavior can be justified by an evolutionary argument when
agents have to choose how to form their forecasts. In the context of a simple
asset pricing problem, Brock & Hommes [306] have analysed an economy with
heterogeneous agents: fundamental traders and agents who forecast the future price
by looking at past deviations of the price from the fundamental value (such agents
can represent trend followers, contrarians or technical analysts). Similarly as in the
imitation-based population dynamics (9.53), agents switch to the more successful
forecasting rule on the basis of its past performance. In Brock & Hommes [306] it
is shown that this behavior yields a highly irregular evolution of the economy, with
chaotic asset price fluctuations when the switching intensity from one forecasting
rule to another is high. This result finds confirmation in LeBaron et al. [1172],
where the agents’ evolution is modeled via a genetic algorithm. A model of
this type allows to generate serially correlated returns and patterns of volatility
and volume which are consistent with those empirically observed (see Brock &
LeBaron [309], Gaunersdorfer [762], Boswijk et al. [272], Chiarella & He [429]).
In Routledge [1474] a process of adaptive (or evolutionary) learning is analysed in
the context of the model proposed in Grossman & Stiglitz [849] (see Sect. 8.3).
Agents must choose whether or not they wish to become informed on the dividend
of a risky asset and the proportion of informed agents in the economy evolves
by means of experimentation and imitation, in the sense that successful behavior
is imitated. Therefore, if being informed allows to reach a higher level of utility,
then more agents will buy information in the next period. Unlike in Grossman &
Stiglitz [849], both the choice of whether to buy information and the inference
are determined by adaptive learning rather than optimization, thus capturing non-
rationality. Routledge [1474] shows that the equilibrium defined in Grossman &
Stiglitz [849] is asymptotically stable with respect to the learning process.
The second conjecture reported above (namely, that the speculation activity of
rational traders helps to stabilize the market) also does not hold in general. Indeed,
as shown in Hart & Kreps [907], even speculation with storage in a commodity
market may lead to a destabilization of the market. Stein [1567] shows that the
introduction of a new group of speculators may destabilize prices and cause welfare
losses to the economy.
The stabilizing effect of rational trading (speculation) has been analysed in a
large literature (see, e.g., De Long et al. [548, 549]). The analysis developed in
De Long et al. [549] suggests that, in the presence of a mean reverting process for
the noise traders’ misperception, the optimal investment strategy is not necessarily
the strategy of investing on the basis of the expected return of the asset (rational
speculation). Indeed, the optimal strategy for sophisticated investors is a market
timing strategy, with an increased exposure to stocks after their prices have fallen
and a decreased exposure after they have risen. In other words, the optimal trading
strategy for sophisticated investors seems to be in line with the Keynes’ quotation
reported at the beginning of this chapter: buy when noise traders want to sell and
sell when noise traders want to buy (a contrarian strategy, see De Long et al. [549,
Section IV]). Note also the similarity of this strategy to the trading behavior of
feedback traders. In this setting, in an economy with non-rational agents, rational
agents (sophisticated investors) must consider the beliefs and the behavior of non-
rational agents. In this setting, rational speculation is always stabilizing, but the
average returns earned by rational speculators is not necessarily as high as those
earned by noise traders, thus showing that “destabilizing speculation” (i.e., the
trading activity of noise traders) is not necessarily unprofitable.
De Long et al. [548] show that speculation can destabilize the market in the
presence of positive feedback traders. Moreover, feedback traders and rational
agents together induce an increase in asset price volatility. The rationale of this
effect can be traced back to Keynes’ quotation: suppose that rational agents receive
positive news about the asset and buy it, making the price climb up, then positive
feedback traders react to this price increase by buying the asset and, therefore,
feedback traders make the price increase even further. Rational agents anticipate
this behavior of feedback traders and, therefore, they act to increase the price well
above the fundamental value (by buying more of the asset today) in order to sell the
asset at a higher price at a later date. In other words, the fact that feedback traders
will buy the asset at a later date induces rational traders to make the price of the asset
overreact to the information arrived in the market. The combined effect of these two
classes of traders generates positive serial correlation of returns. Note that in this
9.4 Bounded Rationality and Distorted Beliefs 541
setting rational agents buy an asset when its price is likely to run up and this may
also occur when the price is already above the corresponding fundamental value. In
this way, speculation can destabilize the market. Similar results have been obtained
in Hirshleifer et al. [947] in the case where agents acquire the same information
sequentially.
Summing up, the analysis developed in the present section shows that rational
agents have two motives to trade in the market: buy an asset in a buy and hold
perspective in order to receive its future dividends and buy an asset in order to
sell it at a later date at a more favorable price. If there are rational agents with the
same time horizon of the economy and liquidity traders (price inelastic demand),
then the first motive for trade is more relevant and, therefore, speculation stabilizes
the market. On the other hand, in a market where non-rational agents are present
the second motive plays an important role, especially in a short time horizon, and
speculation may destabilize the market.
9.4 Bounded Rationality and Distorted Beliefs
In the analysis developed in the previous chapters, one of the main hypotheses is that
of rational expectations. The rational expectations hypothesis consists of two main
assumptions: i) agents perfectly know the economic model; ii) agents are able to
efficiently elaborate all the available information and rationally update their beliefs
according to Bayes’ rule.
Clearly, the rational expectations hypothesis represents a strong assumption on
the behavior of the agents. In this regard, Radner assessed as follows the role of
the rational expectations and perfect foresight hypotheses (see Radner [1439]): “the
perfect foresight approach is contrary to the spirit of much of competitive market
theory in that it postulates that individual traders must be able to forecast, in some
sense, the equilibrium prices that will prevail in the future under alternative states of
the environment. [: : :] This approach still seems to require of the traders a capacity
for imagination and computation far beyond what is realistic. An equilibrium of
plans and price expectations might be appropriate as a conceptualization of the ideal
goal of indicative planning, or of long run steady state toward which the economy
might tend in a stationary stochastic environment”.
In this section, we consider two different ways of relaxing the rational expecta-
tions (perfect foresight) hypothesis (see Thaler [1587] for a survey on the topic):
• bounded rationality and learning, meaning that agents are assumed to believe in
a simplified model (different from the true underlying model) and update their
beliefs via some learning mechanism;
• distorted beliefs, meaning that agents’ beliefs are modeled by taking into account
psychological biases;
In the notes at the end of the chapter, we shall also briefly address models with
incomplete information, where the economy is affected by some unobservable
factors which have to be estimated by the agents.
Bounded Rationality and Learning
In a nutshell, bounded rationality is based on two main ideas: i) agents do not have
a complete knowledge of the true underlying economic model but rather believe in
some simplified model; ii) agents use the simplified model when formulating their
expectations and update their beliefs via some recursive learning mechanism (see
Evans & Honkapohja [657] for an overview of this strand of literature).
In order to illustrate an economy where agents exhibit bounded rationality, let us
consider the fundamental equation (6.85). For simplicity, let us assume the existence
of a risk neutral representative agent and suppose furthermore that the dividend
process .dt /t2N is predictable, meaning that the dividend at date dtC1 is known at
date t, for each t 2 N. In this case, equation (6.85) can be rewritten as
1
st D .EŒstC1 jFt C dtC1 / ; for all t 2 N; (9.62)
rf
where 1=rf is the discount factor of the representative agent, .st /t2N is the price
process of the security paying the dividend stream .dt /t2N and the filtration .Ft /t2N
represents the information flow of the economy. Of course, EŒstC1 jFt is the
expectation of the price prevailing at date t C 1 on the basis of all available
information up to date t. Let us introduce the further assumption that the dividend
process .dt /t2N follows a first order autoregressive process of the form
dtC1 D ˛ C dt C "Qt ; for all t 2 N; (9.63)
where .Q"t /t2N is a sequence of i.i.d. random variables with zero mean and where
˛ 2 R and 2 .1; 1/ are some parameters which are assumed to be unknown to
the agents.
As shown in Exercise 9.15, under the above assumptions, the rational expecta-
tions price process .st /t2N of the security paying the dividend stream .dt /t2N is given
by
˛ 1
st D C dtC1 C ˇt ; for all t 2 N; (9.64)
.rf 1/.rf / rf
where .ˇt /t2N represents the bubble component (see Sect. 6.5). In particular, the
fundamental (bubble free) solution under rational expectations is obtained by
taking ˇ D 0. Formula (9.64) shows that the fundamental value of a security
under rational expectations is linear with respect to the dividend released at date
t C 1. According to the rational expectations hypothesis, the representative agent

formulates his expectation of the future price of a security by exploiting all the
available information.
On the other hand, in a bounded rationality setting, the agent formulates his
expectations on the basis of a simplified model which can be taken of the following
form:
st D t1 C t1 dtC1 C ˇt ; for all t 2 N; (9.65)
where .t /t2N and .t /t2N are two sequences of parameters which are estimated on
the basis of past prices and dividends according to some learning rule (for instance,
ordinary least squares). The simplified model (9.65) is misspecified and coincides
with the rational expectations solution only when it is correctly specified, i.e., when
t and t coincide with the true values of the coefficients given in (9.64). The
asymptotic behavior of t and t as a result of a learning procedure can be described
by a system of differential equations having as equilibrium the rational expectations
solution given in (9.64). In this sense, the bounded rationality hypothesis has
interesting implications. Indeed, even in the most simple asset pricing model (as
considered in Timmermann [1590]), the rational expectations solution admits an
evolutionary justification by considering the limit behavior of the agents’ learning
process. In other words, if agents do not know the true underlying economic
model, but they use the public information to make their forecasts and they update
their beliefs by means of a learning algorithm, then, under suitable conditions, the
economy will converge to the rational expectations equilibrium price.
In Evans [654, 655] and Adam & Marcet [14] it is shown that the fundamental
solution is stable, without restrictions on the parameters of the model. Depending
on the model specification, it is also possible that the learning process converges
to a bubble solution, as considered for instance in Branch & Evans [278] by
assuming a learning procedure based on ordinary least squares. The convergence
of the bounded rationality equilibrium price to the rational expectations price can
also be established without introducing the assumption that the dividend process
is predictable, as considered above. In this case, an agent must estimate both the
expected dividend and the expected price, typically by relying on an autoregressive
(misspecified) model for the dividend and on a model of the type described above
for the asset price (see Timmermann [1590, 1593] and Sogner & Mitlohner [1554]).
On the other hand, the convergence of the learning procedure is not ensured in the
presence of a feedback effect of the price on the dividend in the case of multiple
rational expectations equilibria, see Timmermann [1592].
As explained above, an important characteristic of a bounded rationality econ-
omy is that, under suitable conditions, the learning procedure converges to the
rational expectations equilibrium in the long run. However, in a finite time horizon,
a bounded rationality economy may significantly depart from a rational expectations
economy. In turn, this fact opens the door to the introduction of bounded rationality
as a potential explanation for many of the asset pricing anomalies encountered when
testing the implications of classical asset pricing theory (see Chap. 7).
In particular, Buckley & Tonks [321] suggest that bounded rationality may
explain the excess volatility phenomenon. The authors remark that most of the
criticisms on excess volatility results have focused on the statistical properties of
the tests, without paying sufficient attention to the fact that most of the volatility
tests assume a strong form of the rational expectations hypothesis, namely that
the agents know and employ the true underlying economic model when forming
expectations (for instance, the detrending procedure adopted in the seminal work
of Shiller [1533] is based on the full observation sample). Hence, the empirical
evidence reported in the literature on excess volatility can also point at violations
of the rational expectations hypothesis. Starting from this observation, Buckley &
Tonks [321] suggest to test excess volatility by assuming a weak form of rationality
(bounded rationality), assuming that agents form expectations by relying on a model
estimated from historical data using unbiased estimation techniques, where at each
date t the expectations are formed on the basis of the data available only up to date t.
Let us present in more detail the volatility test proposed by Buckley & Tonks
[321]. Following Shiller [1533], the perfect foresight rational price (or ex-post
rational
P1 price) of a security paying the dividend stream .dt /t2N is given by set WD
sD1 dtCs =rf , for t 2 N, as in formula (7.2). Buckley & Tonks [321] consider the
s
following weak form rational expectation of the perfect foresight price, given the
information set Ft available at date t:
X1
1
EOt Œset jFt D s EOt ŒdtCs jFt ; for all t 2 N; (9.66)
r
sD1 f
where is a vector of parameters describing the model generating the dividends

and Ot denotes an estimation of given the information Ft available up to date
t. Observe the double role of the information in relation (9.66): on the one hand,
the available information appears directly as the conditioning information Ft ,
representing the current realization of the explanatory variables of the model. On the
other hand, the information is used for computing the estimates Ot of the parameters
, which are then used for computing the conditional expectation appearing in
(9.66). In order to derive the volatility test of Buckley & Tonks [321], we start by
expressing the perfect foresight price set in the following form:
set D E Œset jFt C vQt ; (9.67)
for all t 2 N, where E ŒjFt denotes the Ft -conditional expectation assuming that
the true underlying model is known and where vQt is a random variable uncorrelated
with E Œset jFt and with mean zero. On the other hand, under weak form rational
expectations, in an efficient market the actual price st is given by
st D EOt Œset jFt : (9.68)
Assuming that agents form their forecasts by relying on unbiased estimation

techniques, we may write
EOt Œset jFt D E Œset jFt C uQ t ; (9.69)
where uQ t represents the forecast error due to the use of an estimated model. From
(9.67) and (9.69) we get
set D EOt Œset jFt uQ t C vQt ;
for all t 2 N. Assuming that the random variables uQ t and vQ t are uncorrelated, for
every t 2 N, the last relation yields the following inequality:
h 2 i h 2 i
E set EOt Œset jFt E st EOt Œset jFt ; (9.70)
for all t 2 N. This inequality captures the fact that the actual price st , being an
expected value (see (9.68)), should be closer than the realization set to the predicted
value of set . In particular, inequality (9.70) should be compared to the analogous
inequality (7.6) adopted in the context of the Shiller [1533] volatility test.
Buckley & Tonks [321] apply this methodology to UK data, providing empirical
evidence that the volatility bound (9.70) is not violated (however, the inequality
is rejected on U.S. data, see Shiller [1538]). Part of the excess volatility can be
attributed to revisions of the estimated parameters Ot of the dividend model by agents
with bounded rationality, since the true parameters are not known to the agents.
In a related context, Timmermann [1592] shows that, before reaching convergence
to the rational expectations equilibrium, asset prices exhibit a high volatility, thus
suggesting that the existence of a learning process may explain excess volatility.
In Timmermann [1590] it is shown that learning is a natural candidate in order
to explain the predictability of asset returns and the excess volatility of asset prices.
The author considers a geometric random walk for the evolution of dividends:
log dtC1 D C log dt C "Qt ; for all t 2 N; (9.71)
where .Q"/t2N is a sequence of i.i.d. random variables with mean zero and constant
variance "2 and where C "2 =2 < log rf . Similarly as above, by adopting the
rational expectations equation (6.85) (under the assumption of a risk neutral repre-
sentative agent), it follows that the fundamental (bubble free) rational expectations
solution is given by
g
st D dt ; for all t 2 N;
rf g
where g WD exp.C"2 =2/, thus showing that the fundamental rational expectations
price is proportional to the dividend. However, in the presence of bounded rational-
ity, the agents do not know the true values of the parameters and "2 . Agents need
to estimate the values of those parameters on the basis of past dividend observations.
In the long run, agents learn the true values of the parameters and "2 . However,
in the transient period the learning process induces excess price volatility and
correlation between returns and the lagged dividend yield. The rationale behind this
result is simple, noting that agents need to estimate the dividend growth rate. Indeed,
suppose that an agent has underestimated the dividend growth rate at some date t
(i.e., the estimated growth rate is lower than the true unknown growth rate). In this
case, the agent’s forecasts of future dividends will be smaller than their “true” values
and the stock price, computed as the present value of expected future dividends,
will be lower than the price computed on the basis of the true growth rate of the
dividends. With significant probability, the dividend realized at the following date
t C 1 will be higher than the forecasted value and this will induce the agent to revise
upwards the estimated dividend growth rate, thus increasing the estimated price
of the asset. This learning effect produces a positive correlation between dividend
yields and future returns, whereas returns display only weak serial correlation.
The impact of learning on volatility is more difficult to assess. As shown above,
in correspondence of the rational expectations fundamental solution, the asset price
is proportional to the dividend, with a proportionality factor that does not depend
on time. This implies that dividend shocks will be reflected in proportional shocks
in the stock price. Under bounded rationality, learning implies an additional effect
on stock prices since the estimated dividend growth rate is also influenced by the
dividend shock. A priori, it is not clear how the two effects relate, i.e., whether they
offset each other or whether they act in the same direction. In Timmermann [1590],
by relying on a simulation analysis, it is shown that learning leads to an increased
volatility, in the presence of a sufficiently small sample. These results have been
confirmed assuming different dynamical models for the dividends (see Timmermann
[1593], Guidolin & Timmermann [855], Branch & Evans [278, 277], Hong &
Stein [964]). Similar results have been also obtained in Lewellen & Shanken
[1214] and Ozguz [1390] with Bayesian learning in a constant or time-varying
expected returns setting: they show that learning contributes to explain CAPM
anomalies and the predictability of the market return through the aggregate dividend
yield. See however Carceles-Poveda & Giannitsarou [366] for a skeptical view
on the capability of bounded rationality to explain excess volatility and return
predictability.
In Barsky & De Long [168], bounded rationality is invoked in order to explain
the overreaction of prices to dividend shocks, i.e., the fact that a long run 1%
increase in the level of dividends is associated with an approximately 1.5% increase
in equity values (in other words, the price elasticity of the market index with respect
to dividends is greater than one). As explained above, if the logarithmic dividends
follow a random walk, then the dividend yield is constant and this overreaction
phenomenon cannot occur. For this reason, Barsky & De Long [168] assume that
the long run dividend growth rate is uncertain and time-varying and propose the
following dynamics:
X
t1
log dt D "Qt C .1 /Q"tk C g0 DW "Qt C gt ; for all t 2 N; (9.72)
kD1
where g0 is the permanent dividend growth rate as of date t D 0 and .gt /t2N is
defined recursively by
gtC1 D gt C .1 /Q"t ; for all t 2 N:
The random variables ."Qt /t2N represent shocks to the dividend growth rates that not
only affect log dt (through the first term on the right-hand side of (9.72)) but also
have an attenuated but permanent effect on future dividend growth rates (through
the second term on the right-hand side of (9.72)). By relying on (9.72), the expected
dividend growth rate, conditionally on the information available at date t, is given
by
X
t1
EŒ log dtCj jFt D gt D .1 /Q"tk C g0 ;
kD1
for all t; j 2 N. As can be easily verified, the permanent dividend growth rate can
therefore be written as a geometric average of past dividend shocks:
X
t1
gt D .1 / k log dt1k C t g0 :
kD0
This formula shows that agents extrapolate past dividend growth into the future and
compute prices by taking the expectation of future dividends estimated according to
the above growth rate. This model allows to generate overreaction effects in financial
time series (long swings in stock prices after a shock in the dividends).
In Barucci et al. [173], assuming an autoregressive stationary process for the div-
idends, the authors analyse a bounded rationality economy when agents update their
price cum dividend expectation according to an adaptive scheme. They show that the
learning procedure introduces serial correlation in returns and correlation between
lagged dividend yields and returns. In particular, serial correlation is positive when
returns are computed over a small time window and negative when the window
is long. The bounded rationality learning mechanism generates positive/negative
serial correlation in returns, as documented in the asset pricing empirical literature.
Overreaction and delayed overreaction of prices to dividend news induce the above
serial correlation effects. The memory of the learning mechanism plays a crucial
role: a longer memory induces a smaller degree of dependence when the horizon of
the return is long (the mean reversion effect is weaker), while over a short horizon
the serial correlation of returns and their correlation with the dividend yield is
always significant and positive. The interesting point is that long run dependencies
in financial time series can be explained by short memory in the learning mechanism
of the agents. When the dividend at each time can be zero or constant with uncertain
probability, Bayesian learning induces mean reversion in asset returns (see Cassano
[377]).
Veronesi [1616] analyses an economy where stock dividends are generated
by a Gaussian diffusion process, with the dividend growth rate (the drift of the
diffusion) following a non observable two-state Markov process, with the two states
representing respectively a high and a low growth rate. The crucial point is that
agents cannot observe the dividend growth rate but rather have to estimate it on the
basis of realized dividends. It is shown that the equilibrium price of the asset is an
increasing and convex function of the investor’s posterior probability of the high
state of the unobservable Markov process. This effect is due to the fact that, when
the posterior probability is around 1=2, the agent requires a positive risk premium
for the high uncertainty which is absent when the model becomes close to full
observability (i.e., a posterior probability close to zero or one). As a consequence,
prices show overreaction to (bad) news in good times and underreaction to (good)
news in bad times, an asymmetric behavior widely observed empirically. Moreover,
this non-linearity increases with the investors’ degree of risk aversion. The model
also generates persistence in volatility changes, excess volatility (with a maximum
when there is a high degree of uncertainty and in recessions) and time-varying
expected returns. In a similar model, assuming a very small probability for the
economy to enter into a very long recession state (a small probability of switching
to a low dividend growth rate regime), Veronesi [1617] shows that, if the deep
recession does not occur, then the model may generate ex-post a high equity
premium, as suggested in Rietz [1451].
Finally, it is worth to mention that bounded rationality may also help to explain
the equity premium puzzle, see Brandt et al. [281], Guidolin [854], Cogley &
Sargent [470], Weitzman [1651] and also Liu et al. [1227] for a model with rare
disasters. Assuming heterogeneous agents endowed with diverse correlated beliefs
and bounded rationality (agents do not know the true probability distribution of the
dividend process), it is possible to address the equity premium, the risk free rate
puzzle and excess volatility, see Kurz & Motolese [1145]. In the latter work, the
authors argue that most of the observed volatility in financial markets is generated
by the agents’ beliefs and the asset pricing puzzles are all driven by the structure of
market expectations. In this context, there is a premium for heterogeneous beliefs.
Distorted Beliefs
A model based on bounded rationality and generating underreaction or overreaction

phenomena (as depicted in Bernard & Thomas [202]) and, therefore, positive and
negative serial correlation in asset returns has been proposed in Barberis et al. [153].
The model is based on two well grounded behavioral patterns: representativeness

heuristic and conservatism. Representativeness heuristic means that agents view
events as typical or representative of some specific class and ignore the true
probability law, whereas conservatism consists in the slow updating in face of new
evidence.
In the Barberis et al. [153] model, the earnings process .xt /t2N is supposed to
follow a random walk of the form
xtC1 D xt C yQ t ; for all t 2 N;
where yQ t represents the shock to earnings at date t and is supposed to take one of two
possible values fy; Cyg. All the earnings are paid out as dividends. The agents do
not realize that earnings follow a random walk. Instead, they believe that the value
of yQ t is determined by one of two models, depending on the state of the economy.
According to the first model, it holds that
P.QytC1 D Cy j yQ t D Cy/ D L and P.QytC1 D Cy j yQ t D y/ D 1 L :
According to the second model, it holds that
P.QytC1 D Cy j yQ t D Cy/ D H and P.QytC1 D Cy j yQ t D y/ D 1 H :
The difference between the two models is represented by the fact that
0 < L < 1=2 < H < 1:
This hypothesis captures the idea that, according to the first model, a positive shock
is likely to be reversed (mean reversion), while, according to the second model,
a positive shock is likely to be followed by another positive shock (trend). The
agents are convinced that they know the probabilities L and H . Moreover, they
are convinced that the process .t /t2N governing the transition from the first model
to the second model is also a Markov process (so that the state of the economy
tomorrow only depends on the state of the economy today), taking values in f1; 2g
and with transition probabilities
P.tC1 D 2jt D 1/ D 1 and P.tC1 D 1jt D 2/ D 2 :
If t D 1, then the earnings shocks .Qy/t2N evolve according to the first model (mean
reversion), while, if t D 2, then they evolve according to the second model (trend).
Barberis et al. [153] focus on small values for the parameters 1 and 2 , meaning
that transitions from one regime to the other occur rarely. In particular, it is assumed
that 1 C 2 < 1 and 1 < 2 . Since P.t D 1/ D 2 =.1 C 2 /, agents think that
the mean reverting model is more likely than the trend model.
In order to value the security paying the earnings .xt /t2N as dividends, agents
have to forecast future earnings. According to the above description, the evolution
of earnings is driven by one of two possible Markov models and the investor tries to
understand, on the basis of the earnings’ observations, in which of the two regimes
is the state of the economy. At each date t 2 N, the conditional probability of the
economy being in state 1 (where the earnings are governed by the mean reverting
model) is given by
qt D P.t D 1jyt ; yt1 ; qt1 /;
where qt is computed by relying on the Bayes rule (see Barberis et al. [153,
Section 4.2] for the explicit expression of qt ). Assuming the existence of a risk
neutral representative agent with discount factor ı, the security price is given by the
conditional expectation of discounted future earnings. If the representative agent
was able to realize that the earning process is a random walk, then the asset price
would be equal to xt =ı, at every date t 2 N. However, the representative agent does
not realize that earnings are simply driven by a random walk process, but rather
believes in some combination of the two models described above, neither of which
is a random walk. In this context, the following proposition holds (we refer to the
Appendix of Barberis et al. [153] for the proof).
Proposition 9.14 In the context of the above model (see Barberis et al. [153,
Section 4.2]), the equilibrium price process .st /t2N of the security paying the
earnings .xt /t2N is given by
xt
st D C yt . p1 p2 qt /; for all t 2 N; (9.73)
ı
where p1 and p2 are two constants depending on L , H , 1 and 2 . Moreover, if
k p 2 < p 1 < k p 2 and p2 0;
where k and k are two constants depending on L , H , 1 and 2 , then the

price process .st /t2N given in (9.73) exhibits both underreaction and overreaction
to earnings.
By means of numerical simulations, Barberis et al. [153] show that, according
to the above proposition, underreaction and overreaction are generated for a wide
range of the values of the model’s parameters. Moreover, the earnings to price ratio
has predictive power.
Uncertainty on the dividend growth rate under incomplete information (see also
below in this section) and different learning mechanisms have been analysed in
Brandt et al. [281]. The authors study the effects on the risk premium and volatility
of Bayesian and suboptimal learning rules, optimism, pessimism, conservatism and
limited memory. Bayesian learning leads to realistic variations in the conditional
equity risk premium, return volatility and Sharpe ratio, while significantly different
results are obtained under alternative learning rules. Moreover, it is shown that,
when agents are conscious of their limits and take them into account, then the
9.5 Incomplete and Imperfect Markets 551
economy with rational (Bayesian) learning and that with suboptimal learning are
almost indistinguishable. When investors have to estimate an unknown parameter
which is relevant for the valuation, conservatism (prior beliefs receive excessive
weights) and representativeness heuristic (agents overweight recent observations)
have been also analysed in Brav & Heaton [284], showing that overreaction and
underreaction arise in both cases.
In Cecchetti et al. [381], distorted beliefs without learning have been considered
in the context of a model which is classical in every respect, apart from the fact
that the beliefs on the endowment growth are distorted. Agents observe the true
state of the economy, but they do not know the true transition probabilities (as
estimated from U.S. data by means of maximum likelihood techniques) between
the high growth regime and the low growth regime as well as the true growth rates.
Assuming pessimism about the persistence of the expansion state and optimism
about the persistence of the contraction state (in other words, beliefs are distorted
towards a lower growth rate for the economy), the risk free rate and the risk premium
observed historically are matched with a time separable utility function and a rather
low coefficient of risk aversion (below 10). If beliefs about transition probabilities
fluctuate randomly about their subjective (distorted) mean values, then the model
also generates excess volatility and predictability in asset returns. Similar results
are obtained in Abel [7] assuming pessimism and doubt on the distribution of
the growth rates of aggregate consumption. Pessimism corresponds to a leftward
translation of the objective distribution of the logarithmic growth rate (the subjective
distribution is dominated by the objective distribution according to first order
stochastic dominance, see Sect. 2.3). Doubt means that the subjective distribution
is a mean preserving spread of the objective distribution. As shown in Cabrales
& Hoshi [326], a model with optimist and pessimist agents may generate time-
varying conditional variance of asset returns and explain the relationship between
volume and volatility. On the relevance of pessimism as an explanation of the equity
premium puzzle see also Cogley & Sargent [470] and Guidolin [854] and Jouini &
Napp [1051, 1052] and Bhamra & Uppal [220] in the case of heterogeneous beliefs.
9.5 Incomplete and Imperfect Markets
In most of the analysis developed in the previous chapters, we have assumed that
financial markets were frictionless, in the sense that there are no transaction costs,
taxes, trading constraints or liquidity constraints. In the present section, we discuss
the relevance of these hypotheses and present a broad overview of some of the main
results obtained in the financial economics literature by relaxing those assumptions.
In particular, we have seen that many crucial results rely on the assumption of
market completeness and we start this section by analysing the role of the market
completeness hypothesis in financial economics.
Market Incompleteness
As we have seen in Sect. 4.2, the assumption of market completeness has important
implications from the point of view of equilibrium theory, especially for the Pareto
optimality of equilibrium allocations. Indeed, market completeness implies that
agents are unrestricted in their possibilities to share risks and, therefore, every
equilibrium allocation will be Pareto optimal and all the idiosyncratic risk will
be diversified away. In correspondence of an equilibrium, the optimal consump-
tion of every agent depends only on aggregate risk and the marginal rates of
substitution are identical across all agents. Moreover, the analysis of an economy
with heterogeneous agents is simplified if markets are complete. Indeed, under
suitable conditions (see Chap. 4), the original economy with multiple agents is
observationally equivalent to a representative agent economy.
In an incomplete market setting, things are not so simple and, in correspondence
of an equilibrium, the variability of individual consumption may exceed that
of aggregate consumption and asset prices may differ from those obtained in
a representative agent economy. As a consequence, market incompleteness and
heterogeneous agents can offer an interesting perspective in order to reconcile
classical asset pricing theory with the asset pricing anomalies described above. For
instance, Mehra & Prescott [1319] suggest that market incompleteness may help to
explain the equity premium puzzle.
In this context, let us briefly present and discuss the model proposed in Weil
[1650], which illustrates the asset pricing implications of market incompleteness
due to the presence of uninsurable labor income risk. Consider a two-period
economy (i.e., t 2 f0; 1g) populated by a continuum of identical agents i 2 I. Each
agent i 2 I is supposed to maximize over all feasible consumption plans a time
additive expected utility of the form
u0 .ci0 / C EŒu1 .ci1 /;
where the utility functions u0 and u1 are the same for all the agents and are assumed
to be strictly increasing, strictly concave and twice differentiable. Every agent
receives a known endowment e0 at date t D 0 (identical across all the agents) and a
random endowment eQ i1 at date t D 1. It is assumed that the probability distribution
of eQ i1 is the same for all the agents (so that the endowments of all the agents are
ex-ante identical), but the realizations of eQ i1 are idiosyncratic to each agent (so that,
the endowments are ex-post different). Since the number of agents is infinite, the
law of large numbers implies that eN 1 WD EŒQei1 , meaning that the realized per capita
aggregate endowment is identical to the ex-ante expected endowment of every agent.
Therefore, there is no aggregate risk in the endowment at t D 1. It is crucial
to assume that the realization of eQ i1 is a private information of agent i, for each
i 2 I. As a consequence, the agents are restricted in sharing the endowment risk,
since they cannot insure against unobservable realizations of the individual random
endowment of the other agents. According to Weil [1650], this feature should model
the unavailability of insurance against fluctuations in labor income (labor risk).
Besides the endowments e0 and eQ i1 , i 2 I, agents are also endowed with one unit of
a security paying a dividend d0 at the initial date t D 0 and the non-negative random
dividend dQ 1 at date t D 1. We denote by p the price of the security at t D 0. In
the model, the dividend risk represents the aggregate risk and it is assumed that the
dividend risk and the idiosyncratic risks (i.e., the endowment risks) are independent.
In this setting, we can write as follows the budget constraint of each agent i 2 I:
ci0 D d0 .wi 1/p C e0 ; (9.74)

ci1 D wi dQ 1 C eQ i1 ; (9.75)
where wi represents the demand of the security by agent i. Due to the budget
constraints (9.74)–(9.75), the optimal consumption problem faced by agent i 2 I
can be formulated as follows:

max u0 .d0 .wi 1/p C e0 / C EŒu1 .wi dQ 1 C eQ i1 / :
wi 2R
The first order optimality condition leads to
p u00 .d0 .wi 1/p C e0 / D EŒu01 .wi dQ 1 C eQ i1 /dQ 1 ;
for every i 2 I. Since all the agents are ex-ante identical (because they have the
same preferences, the same endowments at t D 0 and random endowments at t D 1
with identical distributions) and in equilibrium the market must clear, it holds that
wi D 1, for all i 2 I. Hence, the equilibrium price p of the security is given by
EŒdQ 1 u01 .dQ 1 C eQ i1 /

p D : (9.76)
u00 .d0 C e0 /
The equilibrium expected rate of return r is given by
EŒdQ 1 EŒdQ 1 u00 .d0 C e0 /

r D D (9.77)
p EŒdQ 1 u01 .dQ 1 C eQ i1 /
and the equilibrium risk free rate rf is given by
u00 .d0 C e0 /
rf D : (9.78)
EŒu0 .dQ 1 C eQ i /
1 1
In correspondence of this equilibrium allocation, each agent does not trade

and simply consumes his endowment and the dividends delivered by the unit of
the security in his endowment. This equilibrium is Pareto dominated by the “risk
pooling” allocation where each agent consumes d0 C e0 at date t D 0 and dQ 1 C eN 1 at

date t D 1. Indeed, due to the concavity of the utility function u1 , Jensen’s inequality
implies that

EŒu1 .dQ 1 C eQ i1 / D E EŒu1 .dQ 1 C eQ i1 /jdQ 1

E u1 .dQ 1 C EŒQei1 jdQ 1 /
D EŒu1 .dQ 1 C eN 1 /;
where in the last equality we have used the independence of the random variables
dQ 1 and eQ 1 .
Let us now consider a fictitious representative agent economy, assuming that
each agent has received the average endowment eN at date t D 1, neglecting the fact
that individuals have their idiosyncratic and uninsurable labor risks. In other words,
this representative agent economy corresponds to the allocation where each agent is
supposed to receive the Pareto optimal consumption plan .d0 Ce0 ; dQ 1 C eN 1 / described
above. In this case, by following the above line of reasoning, the equilibrium price
pO of the security in the representative agent economy would be given by
EŒdQ 1 u01 .dQ 1 C eN 1 /

pO D
u00 .d0 C e0 /
and the equilibrium expected return rO and risk free rate rOf by
EŒdQ 1 EŒdQ 1 u00 .d0 C e0 /

rO D D ; (9.79)
p
EŒdQ 1 u01 .dQ 1 C eN 1 /
u00 .d0 C e0 /
rOf D : (9.80)
EŒu01 .dQ 1 C eN 1 /
The following proposition compares the pricing implications of the competitive

equilibrium allocation with that of the fictitious representative agent equilibrium
(see Exercise 9.16 for a proof).
Proposition 9.15 In the context of the above model (see Weil [1650]), it holds that
rOf rf and rO r
if and only if u000

1 > 0.
This proposition provides an example of the implications of market complete-

ness, since in the setting considered above the presence of uninsurable labor risk
makes the market incomplete. When u000 > 0 (the utility function u exhibits
prudence, see Sect. 3.3), there is a precautionary saving motive: the presence of
uninsurable labor risk increases the demand for both the risk free and the risky
asset. In equilibrium, since the risk free asset is in zero net supply and the risky
asset is in fixed unit supply (per capita), the increased demand must be compensated
by a higher price of the two securities or, equivalently, by a lower return. On the
other hand, the fictitious representative agent neglects this precautionary saving
motive, thus generating an overestimation of the expected rates of return of the
two securities. This result allows to explain the risk free rate puzzle (see also
Kocherlakota [1112]). Concerning the equity risk premium, as shown in Weil
[1650, Proposition3], if the utility function u1 exhibits decreasing absolute risk
aversion and exhibits decreasing absolute prudence, then the equity premium will
be underestimated in the fictitious representative agent economy in comparison to
the original competitive economy with uninsurable labor income. See also Mankiw
[1293] and Gollier & Schlesinger [804] for related models.
In the financial economics literature, even in a multi-period setting allowing
for dynamic trading, incomplete markets can be generated by an exogenous labor
income process affecting the wealth dynamics. In the presence of labor income, the
classical results obtained under the assumption of market completeness may change
substantially. As a first effect, the wealth process associated to a self-financing
strategy does not follow anymore the dynamics (6.4), since on the right hand side
an additional random variable yt appears, representing the labor income received at
date t. In this setting, as illustrated in the simple context of the two-period model of
Weil [1650], a precautionary saving motive emerges for a prudent agent: the agent
saves more in order to face the uncertainty of future labor income, reducing current
consumption.
The presence of a precautionary saving motive has been invoked in the macroe-
conomics literature in order to reconcile the properties of observed consumption
time series with a general equilibrium model of the type described in Chap. 6. As a
matter of fact, several anomalies appear when one tries to interpret consumption data
in the setting of the classical life cycle permanent income hypothesis (in this regard,
see Deaton [530] and Chap. 6). However, the relevance of a precautionary motive
as a possible explanation of consumption data is unclear. Indeed, some authors
emphasize the relevance of precautionary saving quantitatively and in explaining
some puzzles such as the excess sensitivity of consumption to anticipated and
lagged income changes, the excess smoothness of consumption with respect to
unanticipated income changes and the high growth of consumption with a low
risk free rate (this is related to the risk free rate puzzle, see Sect. 7.3), see Zeldes
[1676], Caballero [325], Banks et al. [132], Skinner [1550], Carroll [369], Carroll
& Samwick [374], Carroll [368], Dynan et al. [614], Gourinchas & Parker [815].
Carroll [368] and Carroll & Samwick [373] show theoretically and empirically
that agents with a greater predictable income uncertainty exhibit lower current
consumption, hold more wealth and the wealth held in order to face future income
fluctuations represents a relevant part of the total wealth. This component helps to
explain a high consumption growth rate in the presence of a low risk free rate.
Impatient and prudent agents with a borrowing constraint adopt a “buffer stock”
behavior (see Deaton [529], Carroll [369]). According to this approach, impatient
agents have a wealth target: below that level prudence and precautionary saving
dominate, so that the agent saves, while above that level the agent consumes and
does not save. The buffer stock model is shown to provide an adequate description
of the consumers’ behavior up to the age of fifty, while afterwards the standard life
cycle hypothesis gives an appropriate representation of the agent’s behavior (see
Gourinchas & Parker [815]). On the other hand, Aiyagari [33] and Guiso et al.
[859] show that the contribution of uninsured idiosyncratic risk to aggregate saving
is rather modest and Dynan [612] estimates a small coefficient of relative prudence
and, consequently, a limited precautionary motive.
Portfolio choices in the presence of uninsurable labor income have been analysed
in several papers. In particular, uninsurable labor income has three different effects
on portfolio choices: a mean effect, a variance effect, and a correlation effect.
As a first observation, note that an exogenous riskless labor income is equivalent
to a position in the risk free asset. As a consequence, the presence of a riskless
labor income increases the demand of risky assets for agents with DARA utility
functions (mean effect, see also Jagannathan & Kocherlakota [1009]). This happens
in particular when the agent is young, thus generating a horizon effect. On the other
hand, the presence of a labor income that is perfectly correlated to stock returns
should induce the agent to invest more in the risk free asset. Assuming the presence
of a non-perfect correlation between labor income shocks and stock market returns,
the hedging demand of risky assets has a sign which is the opposite to that of the
correlation (correlation effect). Moreover, ex-post labor supply flexibility induces an
agent to take more risks ex-ante, thus providing an explanation to the asset allocation
puzzle (young agents have more labor supply flexibility, such as the retirement
date choice, and, therefore, they invest in risky assets more than old agents, see
Bodie et al. [260], Dybvig & Liu [608], Farhi & Panageas [680], Barucci &
Marazzina [172]). As we have already remarked in a two-period setting in Sect. 3.3,
when facing a saving-portfolio problem, a prudent agent with decreasing absolute
prudence and decreasing absolute risk aversion who is exposed to an uninsurable
risk in the future (background risk) will reduce his demand of the risky asset
(variance effect), see also Elmendorf & Kimball [637]. In Elmendorf & Kimball
[637] it is also shown that in this case a reduction in labor income risk (which is
independent of asset returns and does not make the agent worse off) leads the agent
to invest more in risky assets. This result is confirmed in a multi-period setting, with
an increase in labor income risk inducing an agent to save more and to reduce his
positions in the risky assets (see Koo [1119], Viceira [1622], Campbell & Viceira
[356], Henderson [935], Benzoni et al. [195], Gomes et al. [806], Lynch & Tan
[1260]). Summing up, the presence of non-financial income has a (positive) mean
effect and a (negative) variance effect on the risky asset demand. Some empirical
evidence on these effects has been reported in Guiso et al. [860], Haliassos & Bertaut
[876], Carroll & Samwick [373], Vissing-Jorgensen [1625], while small effects of
labor income risk on portfolio choices have been detected in Letendre & Smith
[1192].
Multi-period portfolio choices in the presence of an idiosyncratic labor income
component which is uncorrelated with stock returns show that young (employed)
agents hold more risky assets than old (retired) agents, see Viceira [1622]. Young
agents borrow to invest in risky assets. The stock demand of a young agent increases
if the human capital decreases over the life span and is increasing with respect to
the expected retirement date. Positive correlation between non-financial wealth and
asset returns induces a negative hedging demand and a young agent may hold less
risky assets than a retired agent (positive evidence on the correlation effect has been
detected in Heaton & Lucas [929], while weak evidence is reported in Vissing-
Jorgensen [1625]). Heaton & Lucas [929] observe that households with high and
variable proprietary business income, which is highly correlated with stock returns,
hold less risky assets than otherwise similar wealthy people, but they still hold a
significant amount of risky assets (thus contradicting the variance effect).
If markets are complete, then full mutual risk insurance implies that consumption
should not vary across agents in response to purely idiosyncratic shocks, since the
consumption plan of every agent depends only on the aggregate risk. In particular,
considering a power utility function, optimal consumption growth rates should
not respond to idiosyncratic risks, they should be highly correlated and more
correlated than income growth rates. In the literature, several empirical tests of
the implications of optimal risk sharing on consumption have been proposed. In
particular, Cochrane [459] has regressed consumption growth rates on a set of
idiosyncratic variables, showing that full insurance is rejected for long illness and
involuntary job loss, but not for unemployment and involuntary move. Negative
results for full insurance have been also obtained in Attanasio & Davis [83], Banks
et al. [132], Blundell et al. [259] by analysing the effects of changes in the relative
hourly wage across birth cohorts and education groups on the distribution of
household consumption. Perfect risk sharing among agents of different countries
is also rejected when analysing consumption growth rates of different countries
(risk sharing home bias), see Lewis [1216, 1215]. However, considering utility
functions with a high risk aversion near the subsistence level and a decreasing
risk aversion in its neighborhood, positive evidence for risk sharing is obtained
in Ogaki & Zhang [1383]. Assuming homothetic preferences, Mace [1265] tests
the full insurance hypothesis by regressing individual consumption on aggregate
consumption and some idiosyncratic risk variables. Any variable other than the
change in the aggregate consumption is predicted to be insignificant in explaining
changes in the agents’ consumption. The empirical results reported in Mace [1265]
agree with the full insurance hypothesis in the case of an exponential utility function
but not for a power utility function.
The market completeness hypothesis has also important implications for the
validity of the aggregation property (the existence of a representative agent) in
financial markets. Several empirical tests of the aggregation property have been
proposed in the literature. Considering a power utility function and a joint log-
normal distribution for the real interest rate and the consumption growth rate, the
relations derived from (6.78) can be tested for the representative agent (with respect
to aggregate consumption) as well as for cohort data. The empirical results reported
in Attanasio & Weber [87] show the existence of aggregation imperfections, namely
the estimates of the elasticity of intertemporal substitution of consumption obtained
from aggregate data are substantially lower than those obtained from average cohort
data. Moreover, Euler conditions are rejected on aggregate data, but they are not
rejected on average cohort data. Similar results are also obtained in Attanasio &
Weber [88] for non-durable expenditure in U.S. Non-linearities and demographic
factors are at the origin of these differences. Aggregation may also explain the
excess sensitivity of consumption to income shocks, while the effect disappears
when including demographic variables (see Attanasio & Browning [82]).
Jacobs [1005] has argued that market incompleteness is not a relevant phe-
nomenon, in the sense that many well documented asset pricing puzzles do not
result from aggregation problems. Indeed, the Euler conditions for a time separable
power utility function can be tested for individual consumers (households), even
in incomplete markets. If markets are complete, then the conclusions of tests
based on an individual consumer should agree with those of tests based on a
representative agent. The empirical evidence reported in Jacobs [1005] indicates
that the restrictions implied by the Euler conditions are rejected. Similar results have
been also obtained in Telmer [1583] assuming a non-diversifiable idiosyncratic com-
ponent in the agents’ endowments. A limited set of securities suffices to approach
almost perfect risk sharing and this result is even reinforced if idiosyncratic risk
is correlated with the market return. Market incompleteness introduces very little
variability in the intertemporal marginal rate of substitution (the test proposed
in Hansen & Jagannathan [891] is violated). Levine & Zame [1203] propose a
general equilibrium model with a single good showing little relevance of market
incompleteness on welfare, prices and consumption when agents are patient and
have a long horizon, a risk free bond is traded and shocks are transitory.
While there is empirical evidence of market incompleteness, its effects on asset
prices in a multi-period economy are limited because agents can trade repeatedly
over time (see Brav et al. [283], Kocherlakota [1112], Huggett [992], Lucas
[1250], Heaton & Lucas [926, 927], Cochrane [463], Lettau [1193], Marcet &
Singleton [1302], Krusell & Smith [1140]). It is difficult to explain asset pricing
anomalies (notably, the equity premium puzzle) by introducing uninsurable endow-
ment components. Typically, market incompleteness induces a risk free rate lower
than the risk free rate that would prevail under market completeness (due to the
presence of a precautionary saving motive, as discussed at the beginning of the
present section) but the reduction of the risk free rate depends on the set of available
assets as well as on the presence of other market frictions.
Assuming a power utility function, Krueger & Lustig [1138] show that the
absence of insurance markets for idiosyncratic labor risk has no impact on the
premium for aggregate risk if the distribution of idiosyncratic risk is independent
of aggregate shocks and the aggregate consumption growth rate is independent over
time. A common feature of many models is that the time series of the ratio of
each consumer’s labor income over the aggregate labor income is assumed to be
a stationary Markov process with low persistence. If this is the case, then agents
can effectively self-insure in the asset market against temporary income changes.
Constantinides & Duffie [496] have relaxed the above hypothesis and shown that
the joint hypothesis of incomplete insurance and agents’ heterogeneity enriches the
pricing implications of a representative agent model. They assume that individual
income processes are heteroskedastic with a high degree of persistence (e.g.,

idiosyncratic income shocks follow a random walk) and, moreover, that the variance
of idiosyncratic risk increases when the market declines (countercyclical conditional
variance). In an incomplete market, if labor income shocks are permanent, then
agents cannot easily self-insure against them. Under this assumption, agents’
heterogeneity and incomplete insurance help to address several asset pricing puzzles
without the need of introducing other market frictions. Assuming a time additive
utility function with constant relative risk aversion and given a set of processes of
arbitrage free prices, dividends and aggregate income satisfying some restrictions,
Constantinides & Duffie [496] show that there exists an equilibrium supporting the
price processes by a judicious choice of the agents’ labor income process. Therefore,
the Euler conditions hold.
The result of Constantinides & Duffie [496] has been extended by relaxing
the distributional restrictions in Krebs [1132] and providing an explanation of
some asset pricing anomalies. The pricing density is affected by the aggregate
consumption growth and by the cross-sectional variance of consumption growth.
If the variance of idiosyncratic shocks is negatively correlated with equity returns
and aggregate shocks, then the economy with incomplete markets generates a larger
equity premium and a lower risk free rate than the complete market economy
(compare also with the result Proposition 9.15 above). The key features to obtain this
result are the strong heterogeneity and the high persistence in shocks to consumers’
income with high volatility during recessions. The model also addresses the
countercyclical behavior of the equity premium. The equity premium is positively
affected by the covariance of security returns with the cross-sectional variance
of individual agents’ consumption growth rates. Moments of the cross-sectional
distribution of the household consumption growth rate affect equity returns. Jacobs
& Wang [1006] show that agents are not fully insured and that the cross-sectional
variance of consumption growth is also a priced factor. The resulting two factor
consumption-based asset pricing model significantly outperforms the CAPM.
Some empirical evidence in favor of persistence in idiosyncratic shocks and
higher volatility during recessions has been provided in Storesletten et al. [1572,
1573]. Introducing life cycle effects (young agents face more idiosyncratic risks than
old agents), the authors can explain the U.S. equity premium and Sharpe ratio with a
plausible level of risk aversion. However, estimating the processes of idiosyncratic
shocks on U.S. data, Heaton & Lucas [927] find a low degree of persistence and
obtain a risk free rate similar to that obtained under the hypothesis of complete
markets (in this regard, see also Kocherlakota [1112], Cochrane [463], Lettau
[1193]). Cogley [469] shows that the cross-sectional dispersion of logarithmic
consumption growth is only weakly correlated with stock returns and the correlation
is not strong enough to generate an equity premium comparable to that observed
in the U.S. financial market. Krusell & Smith [1140] show that with a plausible
degree of heterogeneity (i.e., with an equilibrium distribution of income and wealth
matched to the U.S. data) the model cannot address the equity premium puzzle.
Market Imperfections
Market imperfections can refer to many different aspects of financial markets:

borrowing constraints, short sale constraints, transaction costs, solvency constraints,
bid-ask spreads as well as other types of market frictions.
The fundamental theorem of asset pricing has been extended to markets with
transaction costs (bid-ask spread) in several papers, see Jouini & Kallal [1049], Naik
[1369], Schachermayer [1500], Kabanov & Safarian [1056] in discrete time models.
In a nutshell, it holds that a market with bid-ask spreads does not allow for
arbitrage opportunities if and only if there exists at least one equivalent probability
measure that transforms a process taking values between the bid and the ask prices
into a martingale. Similarly as in the context of Sect. 4.4, such a probability
measure can be interpreted as a stochastic discount factor or as the intertemporal
marginal rate of substitution of a maximizing agent. In turn, the latter allows to
define arbitrage bounds on a the price of a contingent claim. For a version of the
fundamental theorem of asset pricing under short sale constraints see Jouini &
Kallal [1050], under solvency constraints see Hindy [942], under credit constraints
see Loewenstein & Willard [1243]. Holding costs allow the price of an asset to
deviate from its fundamental value without yielding riskless arbitrage opportunities,
see Tuckman & Vila [1602]. See Basak & Croitoru [177] and Detemple & Murthy
[565] for general equilibrium models under portfolio constraints.
Constantinides [491] has shown that proportional transaction costs (a transaction
fee which is proportional to the prices) induce agents to reduce the frequency and
the volume of trading. In particular, there is a no-trade region which is increasing
with respect to the transaction costs. However, the effect on an agent’s utility is
limited, the (positive) liquidity premium and, therefore, the effect on asset prices
induced by transaction costs are small. The result has been confirmed in Vayanos
[1614] assuming an overlapping generations model and it is also shown that a stock
price may increase in transaction costs. Empirical evidence supporting these results
(namely that transaction costs induce a small positive risk premium and reduce
trading volume) has been provided in Barclay et al. [159]. A strong discount effect
on asset prices (or, equivalently, a large risk premium) has been derived in Lo
et al. [1239] assuming fixed transaction costs. When asset returns are predictable,
ignoring transaction costs induces a substantial utility cost, see Balduzzi & Lynch
[118] and Garleanu & Pedersen [761]: in this setting, there is a state dependent
no-trade region, which is greater than in the case without predictability and the
classical horizon effect is obtained (agents with a long horizon hold more stocks,
see Lynch & Balduzzi [1259]). The horizon effect can be also obtained in a model
with transaction costs and a finite horizon.
The bid-ask spread can be regarded as a proxy for the liquidity of an asset and is
positively related to its expected return, as documented in Amihud & Mendelson
[59], Chalmers & Kadlec [393], Brennan & Subrahmanyam [297], Eleswarapu
[635], Amihud [57], Easley et al. [615], Bekaert et al. [185]. The rationale is simple:
an agent anticipates the cost associated with trading in the future an illiquid asset
and, for this reason, requires a higher risk premium as a compensation. Equilibrium
models studying the relation between liquidity and risk premia have been considered
in Huang [975] and Acharya & Pedersen [11]. Transaction costs are shown to have
a relevant effect on asset returns if agents are constrained from borrowing against
future income (a similar effect is obtained in Longstaff [1244]). Market liquidity as
a predictor of future returns (in the sense that an increase of liquidity predicts lower
future returns) is explained in Baker & Stein [108] by relying on a model with
irrational traders and short sale constraints. Cho et al. [430] show that, allowing
agents to choose assets to trade with trading costs, then lead-lag effects of the type
described in Lo & MacKinlay [1237] arise between portfolios sorted by size.
Asset prices with transaction costs in an incomplete market have been analysed in
Aiyagari & Gertler [34]. In this model, agents are exposed to idiosyncratic shocks to
their income (with no aggregate shocks) and this risk component cannot be hedged
in financial markets, similarly as in the seminal model of Weil [1650] presented at
the beginning of this section. Two types of assets are traded in the market: stocks
and short term government bonds, with stock trading incurring in transaction costs.
Market incompleteness generates a low risk free rate and the presence of transaction
costs increases the equity premium. This type of model produces a smooth aggregate
consumption and a highly variable individual consumption. However, very large
transaction costs are needed in order to match the historically observed values, as
confirmed in Jang et al. [1014].
Transaction costs in consumption have been analysed in Cuoco & Liu [508],
Grossman & Laroque [844], Liu & Loewenstein [1224], showing that transaction
costs have an effect on consumption smoothing. Fixed adjustment costs in consump-
tion can account for much of the discrepancy between the observed low variance
of aggregate consumption growth and predictions of the CCAPM. Adjustment
consumption costs make agents respond to changes in asset prices with a lag and,
therefore, consumption growth covaries weakly with equity returns. Adjustment
costs help to address the equity premium puzzle (see Lynch [1257] and Marshall
& Parekh [1309]) and yield a strong upward bias in the estimation of the coefficient
of relative risk aversion by means of an Euler equation. Taking into account this
effect, Euler equation tests do not reject the classical CCAPM.
Borrowing, credit and liquidity constraints are relevant for the agents’ decisions:
many empirical studies have shown that these constraints are binding for some
groups of agents (for instance, low income and young agents in the market for
consumer loans, see Attanasio et al. [84]). Similarly, Jappelli [1015] estimates that
20% of the families are rationed in the credit market and also Hayashi [917] and
Gross & Souleles [838] report evidence supporting the fact that liquidity constraints
are binding. In Zeldes [1675] and Jappelli et al. [1017], by relying on tests on the
Euler conditions, it is shown that liquidity-borrowing constraints are binding. On the
other hand, constraints are shown to be non-relevant in the tests proposed in Runkle
[1488] and Meghir & Weber [1318] (except for young households).
The buffer stock model with precautionary saving motives, impatience and
restrictions on borrowing allows to explain both excess sensitivity, life cycle and
business cycle patterns as well as excess smoothness of consumption (see Carroll
[368], Deaton [529], Carroll [369], Ludvigson & Michaelides [1253], Gross &
Souleles [838], Browning & Crossley [317], Deaton [530], Carroll [370]). When
borrowing constraints are binding (i.e., wealth is below a given target), agents do not
save and consume the wealth generated by the assets. In this sense, assets act like a
buffer stock protecting the level of consumption against low income. In this case, the
marginal propensity to consume is higher than in the standard model. Gourinchas &
Parker [815] show that agents behave like buffer stock consumers early in their life,
while afterwards they behave according to the life-cycle hypothesis. Huggett [993]
show that idiosyncratic income shocks and liquidity constraints induce positive
precautionary saving, independently of the third derivative of u (compare with
Proposition 9.15 above). A generalization of the model with a hyperbolic discount
factor (see the last part of Sect. 9.2) has been proposed in Harris & Laibson [897].
Guiso et al. [860] show that in a setting with labor income risk, the presence of
liquidity constraints reduces the investment in risky assets and increases the saving
rate: liquidity constrained agents hold less risky assets than agents who are not. The
rationale of this phenomenon is that liquidity constraints reduce the effect of labor
income and agents invest more in the risk free asset in order to prevent future income
shocks (see also Jappelli & Pagano [1016] and Koo [1119]). Portfolio choices in an
incomplete market (due to labor income) under borrowing and short sale constraints
have been analysed in Heaton & Lucas [928], Cocco et al. [456], Haliassos &
Michaelides [877]. Unless returns and labor income shocks are (significantly)
positively correlated, agents hold only stocks. If agents are not constrained, then
they short sell the risk free asset. The empirical evidence on a significantly positive
correlation is weak. With low positive correlation, labor income is a close substitute
for the risk free asset and, therefore, it induces the classical horizon effect (a young
agent invests a large fraction of wealth in risky assets and afterwards reduces his
stock holdings when he becomes older, see Cocco et al. [456]). Including return
predictability (mean reversion), it is shown that there are substantial effects on the
optimal consumption but only limited effects on portfolio choices (in particular, a
result similar to the one in Heaton & Lucas [928] is obtained), see Campbell et al.
[342]. Michaelides [1335] analysed the relevance of the hedging demand in the
presence of returns predictability, incomplete markets and borrowing constraints.
Asset prices with short sale and borrowing constraints in an incomplete market
have been analysed in Lucas [1250]. The model takes into account the presence
of non-diversifiable labor income idiosyncratic and serially uncorrelated shocks.
The incompleteness of the market is shown to reduce the risk free rate, but does
not suffice to solve the equity premium puzzle. The model reproduces both the
observed risk free rate and the equity premium only when asset markets are closed
or the borrowing constraint becomes severe. With limited access to capital markets,
asset prices will be similar to those obtained in a representative agent economy.
Individual consumption is highly correlated with aggregate consumption and the
volatility of pricing kernels is increased. Similar results are also obtained in Marcet
& Singleton [1302], Telmer [1583], Krusell & Smith [1140], Detemple & Serrat
[566], while more positive results have been obtained in Alvarez & Jermann [54]
with endogenous solvency constraints. Moreover, as shown in Zhang [1679], strong
borrowing constraints allow to match the mean and the variance of discount bond
returns but not higher moments (like the skewness and the kurtosis). Assuming
heterogeneous preferences (namely, different degrees of risk aversion among the
agents) with a time additive utility function, heterogeneous borrowing constraints,
incomplete markets and a stochastic investment opportunity set, a low risk free
rate and a high risk premium are obtained in equilibrium with a tightening of the
borrowing constraint, while volatility increases with the cross-sectional dispersion
of risk aversion (see Kogan et al. [1115] and Gomes & Michaelides [809]). Capital
market restrictions help to explain the lack of international consumption risk sharing
(see Lewis [1215]).
Heaton & Lucas [925, 927] analyse an incomplete market model assuming
systematic and idiosyncratic labor income risk. There are borrowing constraints,
short sale constraints and transaction costs. If there are (large) transaction costs both
in the stock and in the bond market and there is a binding borrowing constraint, then
the risk free rate is low and a large equity premium can be reproduced (compare
also with Huggett [992]). Constantinides et al. [495] show that the presence
of borrowing constrains in an overlapping generations economy (where young
agents face borrowing constraints) generates a risk premium higher than that of
an unconstrained economy. Once more, market incompleteness alone does not help
to solve the puzzles. In general, the main drawback of these models is represented
by the high volatility of the risk free rate. Luttmer [1255] and He & Modest [918]
show that a combination of proportional transaction costs, short-sale constraints,
borrowing constraints and solvency constraints helps to explain consumption data
in the context of a standard intertemporal equilibrium asset pricing model: the
presence of these market frictions generates low comovements in consumption and
returns and reduces the required volatility in the intertemporal marginal rate of
substitution needed to satisfy the bound of Hansen & Jagannathan [891]. A power
utility function with a low risk aversion cannot be rejected by the data. Bansal &
Coleman [133] propose a monetary model with transaction costs which can explain
the equity premium, the risk free rate and the term premium puzzle.
If transaction costs and frictions affect asset prices, then they should primarily
affect high frequency returns and have a weak effect on long horizon returns. This
claim is confirmed empirically in Daniel & Marshall [519], where it is shown
that asset pricing models with keeping up with the Joneses or habit formation
preferences are rejected on quarterly data but perform well at longer horizons.
Correlation between consumption growth and market return increases with respect
to the time horizon and both the equity premium and the risk free rate are reproduced
at longer horizons. However, in many models transaction costs must be very high
in order to be able to reproduce the observed equity premium and the risk free
rate. Under some conditions, Luttmer [1256] estimates a fixed costs of at least
3% of monthly per capita consumption for a logarithmic utility function. The
estimated fixed cost decreases if risk aversion is increased, if preferences exhibit
habit persistence or if short sale constraints or bid-ask spreads are introduced. In
this perspective, the decline of the U.S. risk premium in the last three decades can
be attributed to a reduction of market imperfections.
Market Participation
So far, we have always assumed that all the agents of the economy participate to the
market. However, this hypothesis is not confirmed in reality, as already mentioned
in Sect. 6.6.
There is evidence of a high degree of heterogeneity in portfolio choices, agents
change their portfolio infrequently, their entry in the financial market does not
occur early in the life span and there is a hump shaped (or increasing) age profile
in the share of stocks in household portfolios. Portfolio diversification is mostly
observed in wealthy people. These results are puzzling, since the classical theory
(also considering precautionary saving) predicts significant positions in stocks, early
entrance in the market and a large participation in the market. The low participation
rate also contrasts with the observed high equity premium, i.e., given a high risk
premium agents should invest a large fraction of their wealth in stocks. Fixed
(information) participation costs, preference structures different from expected
utility, borrowing constraints and uninsurable labor income may contribute to
explain the heterogeneity in portfolio choices, the low participation rate and the late
entry in the stock market (see Haliassos & Bertaut [876], Haliassos & Michaelides
[877], Gomes & Michaelides [807, 808], Vissing-Jorgensen [1625], Bertaut [209],
Constantinides et al. [495], Storesletten et al. [1572], Davis et al. [528], Guiso &
Sodini [862]). Abel [6] analyses a model with a fixed cost to participate in the
stock market and borrowing constraints for young agents: in equilibrium, only high
income consumers trade in financial markets, while young agents are inclined to
remain outside of the market. More recent studies have shown that several non-
financial factors affect the decision to invest in stocks: cognitive abilities, trust
in the stock market, awareness of financial assets, financial literacy, familiarity
(geographical or professional proximity) and social interaction (see Christelis et al.
[443], Guiso et al. [861], Guiso & Jappelli [858], Grinblatt et al. [832], Massa &
Simonov [1311], Hong et al. [957], Huberman [981], van Rooij et al. [1609], Brown
et al. [313]).
The observations reported above suggest to explicitly consider limited partic-
ipation as an important factor in the analysis. The main source of the equity
premium puzzle is that consumption growth covaries too little with asset returns
(or the intertemporal rate of substitution is not variable enough). Mankiw & Zeldes
[1298], Brav et al. [283], Attanasio et al. [81] have shown that stockholders’
consumption is more volatile and more correlated with the stock market than with
the economy as a whole and this fact reduces the risk aversion level needed to
match the data (approximately one third of the level of risk aversion based on
aggregate consumption), thus providing a partial resolution of the equity premium
puzzle. Brav et al. [283] show that the unexplained equity premium decreases as
limited participation and incomplete insurance are taken into account. Considering
only asset holders with total assets exceeding a certain threshold, the coefficient
of relative risk aversion needed to explain the equity premium decreases with the
threshold. Euler conditions on asset holders consumption are not rejected with a
coefficient of risk aversion between two and four. In Attanasio et al. [81] it is shown
that the CCAPM is not rejected when it is tested with respect to stockholders’
consumption. A plausible degree of risk aversion to replicate the observed equity
premium is obtained in Ait-Sahalia et al. [32] considering luxury and basic goods
with non-homotetic preferences (a certain amount of basic goods is always required)
and consumption only of wealthy investors. Vissing-Jorgensen [1626] shows that
the elasticity of intertemporal substitution estimated considering only asset holders
is larger than the value obtained considering also non-asset holders. The value for
asset holders is around 0:4, while for non-asset holders it is insignificantly different
from zero. These results help to explain the equity risk premium: as a matter of
fact, considering also non-stockholders, the coefficient of relative risk aversion is
five times the value obtained by considering only stockholders (see also Paiella
[1396], Attanasio & Vissing-Jørgensen [85], Guvenen [867, 868]).
Merton [1332] has shown that limited participation in the market allows explain-
ing some of the CAPM anomalies. Limited participation and agents’ heterogeneity
together with market incompleteness are analysed in Balduzzi & Yao [120]: on
an empirical basis, it is shown that these two features reduce the coefficient
of risk aversion needed to explain the historical equity premium. Moreover, the
model satisfies Euler condition tests with plausible levels of risk aversion. Similar
conclusions are reached in a similar setting (limited participation and incomplete
markets) in Danthine & Donaldson [524] introducing wage negotiation between
workers and firms. Basak & Cuoco [178] have proposed an asset pricing model
confirming the above results. In this model, a risky asset and a risk free asset are
traded in the market and some agents with a logarithmic utility function cannot
invest in the risky asset due to the presence of frictions-information costs. In
equilibrium, the historical risk premium and the risk free rate can be reproduced
with a coefficient of relative risk aversion of the representative agent equal to 1:3 and
the risk free rate is smaller than in the standard model. Furthermore, the volatility
of consumption of unrestricted agents is higher than that of per capita consumption.
As suggested by the above analysis, stock market participation can be included
as an endogenous variable in asset pricing models. In Allen & Gale [45] and
Pagano [1393], the participation of the agents in the financial market is determined
endogenously assuming a fixed entry cost. It is shown that the volatility and the
size of the financial market are strictly related. By increasing market depth, each
additional trader in the market generates a positive externality for the other traders
by reducing asset volatility. This effect leads other traders to participate to the
market, lowers risk, increases the asset price and, therefore, induces the firm to
issue more stocks. In this context, multiple equilibria arise: low volatility-small
markets and high volatility-large markets. An equilibrium with a large market Pareto
dominates an equilibrium with a small market. When the fixed cost is small, there
is full participation. Orosel [1387] analyses an overlapping generations model with
fixed costs of stock market participation and dividends following a Markov process.
In equilibrium, it is shown that market participation is positively correlated with past
dividends and current prices (positive feedback trading). Endogenous participation
in the market contributes to increase the asset price volatility and generates a mean
reversion effect. Hong et al. [957] propose a model of stock market participation
with social interaction and multiple equilibria and show that social households are
more likely to invest in stocks.
Incomplete risk sharing has been modeled by considering a participation con-
straint such that agents are never willing to revert permanently to autarchy (i.e.,
only consuming their private endowment, see Kehoe & Levine [1078], Zhang
[1678], Alvarez & Jermann [53, 54]). An intertemporal asset pricing equilibrium
model including this endogenous participation constraint has been analysed. An
agent exits the financial market, defaults on his debt and reverts to autarchy if
the expected utility derived from consuming the future private endowment without
trading in the financial market is larger than the expected utility derived from
participating in the market. Imposing the participation constraint that autarchy is
never preferred, interest rates are shown to be smaller than in an standard economy
without constraints. Under some conditions, incomplete risk sharing is obtained in
equilibrium. Under plausible conditions on the endowment process (relative shocks
are correlated with aggregate shocks) and on risk aversion, the model satisfies the
volatility constraint on the intertemporal rate of substitution established in Hansen
& Jagannathan [891] and generates a large equity premium (see Alvarez & Jermann
[53, 54]).
Considering the role of uncertainty aversion, in Dow & Werlang [589] a high price
volatility is obtained under information ambiguity if the preferences of the agents
are represented by an expected utility with a non-additive probability measure. In a
related direction, Kelsey & Milne [1086] show that the equilibrium version of the
APT obtained in Connor [475] can be extended to an economy with uncertainty
averse agents whose preferences are represented by an expected utility with a
non-additive probability measure. We refer to Chateauneuf [411] for a simplified
presentation of the framework of Schmeidler [1505]. In a purely subjective setting,
Alon & Schmeidler [52] introduces some axioms for the derivation of the maxmin
expected utility rule. For a survey see Etner et al. [653] and Guidolin & Rinaldi
[856].
The impact of non-classical preference functionals on portfolio choices and
asset pricing has been studied in several papers. For instance, Chew et al. [428]
show that higher aversion with a rank dependent utility function induces a smaller
risky asset demand in the presence of a single asset. In Dekel [543], considering
a utility function à la Machina, it is shown that risk aversion does not imply risk
diversification. Kelsey & Milne [1086] shows that the APT results of Connor [475]
can be extended to preferences satisfying the smoothness condition imposed in
Machina [1266] or represented by rank dependent utility functions.
Considering loss aversion, a model for portfolio choice and stock trading volume
has been proposed in Gomes [805], showing that the optimal demand of risky assets
is discontinuous and non-monotonic. As wealth reaches a threshold level, investors

follow a generalized portfolio insurance strategy (see Gomes [805], Berkelaar et al.
[201]). Loss-averse investors will not hold stocks unless the equity premium is
sufficiently high. The model generates positive correlation between trading volume
and stock return volatility. We refer to Bowman et al. [274] for an analysis of
consumption with loss averse preferences. Note also that, as we have already shown
in the context of the Barberis et al. [150] model, loss aversion can also generate a
horizon effect on portfolio choices when returns are predictable, in the sense that a
long horizon induces agents to hold more stocks (see Ait-Sahalia & Brandt [31]).
A non-parametric version of recursive utility has been proposed in Epstein &
Melino [642]. Assuming positive serial correlation in consumption growth, Epstein
& Melino [642] show that their model can explain the equity premium puzzle and
the risk free rate puzzle but is not able to match second order moments of returns.
Increasing the elasticity of intertemporal substitution, the risk free rate and the
equity volatility decrease. Confirming Kandel & Stambaugh [1069], it is shown that
it is not possible to match simultaneously the risk free rate and the volatility of equity
returns. The empirical evidence reported above shows that recursive preferences do
not allow to simultaneously explain three empirically observed phenomena: a high
risk premium, a low risk free rate and a high volatility of asset returns.
In Campbell [333], the asset pricing methodology of Campbell [332] has been
implemented by including human capital as a factor. The author proposes a multi-
factor model including the market return and variables that help to predict future
stock returns and future labor income growth (which proxies for human capital).
It is shown that the aggregate stock market represents the main risk factor and
the model is able to explain cross-sectional returns and to capture mean reversion
effects. Positive evidence on this model is also reported in Hodrick & Zhang [952].
In an analysis of the role of habit formation and its relation to behavioral patterns,
Antonelli et al. [74] propose a preference functional which takes into account
disappointment and anticipation, by assuming that the habit process is affected by
past utility. When the utility functional is increasing (decreasing, respectively) in the
habit, then the anticipation (disappointment, respectively) effect is captured. In the
first case, when savoring in the past a high level of expected utility, the agent gets
a high level of utility from the current consumption rate. In the second case, a high
level of expected utility in the past induces the agent to require a higher consumption
rate today. Asset pricing results show that an anticipation effect generates a risk
premium smaller than the one obtained with an additive expected utility, whereas
the disappointment effect leads to a higher risk premium. Therefore, disappointment
yields an interesting perspective to address the equity premium puzzle. Mankiw
et al. [1296] and Eichenbaum et al. [634] have introduced leisure in the utility
function. Empirical tests have shown that leisure does not help to improve the fit
of the intertemporal equilibrium model.
Recursive preferences have been formulated in continuous time in Duffie &
Epstein [594, 595]. In Bergman [196], an ICAPM is obtained with a recursive utility
function characterized by a discount factor which is a function of consumption
(Uzawa utility). Epstein & Zin [652] extend the recursive utility approach by
relaxing the independence axiom. However, such a generalization does not seem
to enhance the performance of an asset pricing model.
A different formulation of the keeping up with the Joneses phenomenon has been
proposed in Bakshi & Chen [113], assuming that the wealth level of the agent is
compared to a social wealth index. A volatility bound test of the type described in
Hansen & Jagannathan [891] and a generalized method of moments test provide
positive evidence for this model (see also Smith [1551]). DeMarzo et al. [554]
show that this type of preferences may induce a bubble in the market. As shown
in DeMarzo et al. [553], if agents care about consumption relative to per capita
consumption in their community, then they tend to hold undiversified portfolios
similar to those of the other members of the community. Linking the keeping up with
the Joneses phenomenon to market participation, Hong et al. [957] show that social
investors (investors with social activities) find the market more attractive when more
of their peers participate.
An axiomatization of intrinsic and internal habit has been proposed in Rozen
[1484]. In Jermann [1032] and Boldrin et al. [263], habit formation has been
included in a real business cycle model (endogenous endowment process). Assum-
ing capital adjustment costs or limited inter-sectoral factor mobility, the model can
explain the historical equity premium, the average risk free rate and the asset Sharpe
ratio. However, the model incurs in some problems when trying to reproduce the
second moments of asset returns, in particular of the risk free rate. Note that some
form of input market inflexibility is necessary in order to obtain the above results,
otherwise consumption will be smoother than in a pure exchange economy (see
Boldrin et al. [263] and Lettau & Uhlig [1198]). Standard business cycle properties
are replicated by the models. Carroll et al. [372] show that habit formation may
reproduce the phenomenon that an increase in growth leads to an increase in saving,
as observed empirically.
External habit preferences also allow to reproduce some stylized facts on return
predictability: Menzly et al. [1325] show that time-varying risk preferences induce
a positive relation between dividend yields and expected returns. Countercyclical
variations in the Sharpe ratio of stock returns and in the risk aversion of the economy
are obtained in Chan & Kogan [409] by considering a general equilibrium model
with external habit, heterogeneous preferences and constant risk aversion. The
agents’ risk aversion is constant over time but varies across the population. External
habit preferences imply that risk aversion and, hence, portfolio choices are time-
varying. However, according to Brunnermeier & Nagel [320], this is not consistent
with the empirical data, which typically exhibit inertia in portfolio decisions with
respect to changes in wealth.
For a survey on psychology, decision making and finance see Camerer [328],
Barberis & Thaler [154], Daniel et al. [518], Hirshleifer [945]. In an imperfectly
competitive market, Palomino [1398] shows that noise traders may earn higher
expected returns and higher expected utility than rational traders. Barberis et al.
[147] show that a CCAPM with rational traders and agents extrapolating future
returns from past returns provides a behavioral explanation of return predictability.
In Barucci & Landi [171], the convergence of an agent’s learning process is analysed
in the model proposed in Cutler et al. [509] (a model with rational, fundamentalist
and feedback traders). Agents are able to learn the fundamental solution without
restrictions on the parameters of the model.
Concerning the role of heterogeneity, Gollier [801] investigates the effect of
wealth inequality on asset prices in the case of identical agents differing with
respect to their initial endowments. In the context of a two-period model, the author
shows that wealth inequality affects asset prices in comparison to an egalitarian
economy if and only if the absolute risk tolerance is non-linear (non-hyperbolic
utility functions). Wealth inequality increases the equity premium when absolute
risk tolerance is concave and, introducing a precautionary saving component, wealth
inequality also decreases the risk free rate if the inverse of absolute prudence is
concave.
Considering irrational noise traders with erroneous beliefs, a model of the type of
De Long et al. [549] has been proposed in order to explain the difference between
the closed-end fund price and the net value of the assets held in the fund. In Lee
et al. [1174] the authors report evidence that the small investor sentiment affects the
risk of common stocks, i.e., it is priced by the market (see however Elton et al. [639]
for a different interpretation).
The presence of noise traders as introduced in Shiller [1535] (price inelastic
random demand) has been analysed in Campbell & Kyle [347] in the context of
a continuous time model. In this model, there are two classes of agents: infinitely
lived rational agents maximizing an exponential utility function with a constant
coefficient of absolute risk aversion and noise traders. The demand of noise traders
is described by an Ornstein-Uhlenbeck stochastic process and changes in the levels
of (de-trended) dividends and prices are represented by normal random variables
with constant variance. It is shown that the presence of noise traders affects the
behavior of rational agents by introducing an additional risk component. The
analysis of Campbell & Kyle [347] allows the identification of three components in
the asset price volatility: a first component associated with the information on future
dividends contained in the dividend time series, a second component associated with
the information on future dividends not contained in the dividend time series and a
third component associated with noise trading. Campbell & Kyle [347] calibrate
their model on the Standard and Poor’s time series and show that the observed stock
price movements can be explained by assuming a low discount rate (lower than
4%) and a high risk aversion (or, in the case of a discount factor larger than 5%, by
assuming the presence of a relevant noise traders’ component highly correlated with
fundamental values). In particular, it is argued that the presence of noise traders can
explain several asset pricing anomalies, including excess volatility, overreaction and
mean reversion.
The presence of noise traders has been empirically confirmed in Kelly [1085]
by looking at the rate of participation to the financial market. The author tests with
positive evidence that the rate of noise traders’ participation is a negative predictor
of future stock returns. In Lee et al. [1176] (using the Investors’ Intelligence of New
Rochelle as an investors’ sentiment index), it is shown that investors’ sentiment is a
systematic risk that is priced in the market. Excess returns are contemporaneously
positively correlated with shifts in sentiment and, moreover, the magnitude of

bullish (bearish) changes in sentiment leads to downward (upward) revisions in
volatility and higher (lower) future excess returns.
The effect of the speculation activity of convergence traders has been analysed
in Xiong [1670] and Kyle & Xiong [1151] in an economy populated by noise and
fundamental traders. Convergence traders bet that the price difference between two
assets with similar (but not identical) characteristics will narrow in the future. In
particular, convergence traders exploit short term profitable investment opportuni-
ties generated by noise traders. Under some conditions, it is shown that convergence
traders destabilize the market, i.e., they unwind their positions by buying (selling)
when prices are high (low) after a capital loss. This type of trading strategy acts
as an amplification and contagion mechanism. In Abreu & Brunnermeier [9] it is
shown that rational traders “time the market”, rather than correcting the mispricing
right away ( delayed arbitrage) and delay trading when they are uncertain about the
time their peers will be informed and will exploit the mispricing.
Infrequent feedback trading has been analysed in Balduzzi et al. [115]. The
authors consider an economy where bonds and stocks are traded and where there are
two classes of agents: “speculators” and “feedback traders”. Speculators maximize
their utility from consumption and have rational expectations, while feedback
traders submit infrequent and discrete market orders, mechanically responding to
price changes. It is shown that positive feedback traders increase the volatility
of stock returns and the response of prices to news on dividends, while negative
feedback traders induce the opposite effect. In both cases, feedback trading gener-
ates predictability and heteroskedasticity in returns, even though dividend growth
rates are assumed to be i.i.d. In a related setting, Balduzzi et al. [116] consider
an economy with agents trading continuously and maximizing an expected utility
and agents buying and selling assets only when the price reaches certain triggering
thresholds. The second type of behavior derives from the adoption of optimal
strategies with transaction costs or portfolio insurance strategies and generates
resistance-support levels, jumps in asset prices and volatility when the price barrier
is reached.
Barberis & Shleifer [151] have analysed an economy populated by fundamen-
talist (rational) agents and a class of agents adopting a style investing strategy,
meaning that they classify assets in some groups (according to size, industry
or other characteristics) and allocate wealth among the different groups (styles)
rather than among the individual securities, switching between different groups
according to their past performance. In other words, investors get into styles that
have performed well in the past relative to other styles, financing the reallocation of
wealth by withdrawing money from styles that have performed poorly. It is shown
that style investing leads to comovements between stocks belonging to the same
style (see also Barberis et al. [152]). Furthermore, asset returns are positively serially
correlated at short horizons and mean reverting in the long run. Prices deviate from
fundamental values but there is a long run pressure towards fundamental values.
The volatility turns out to be higher in comparison to an economy populated only
by fundamentalist agents.
Assuming that agents have confidence on information that depends on its

reliability, Bloomfield et al. [250] show by means of experiments that agents
tend to moderate their confidence on their prior expectation of reliability and, as
a consequence, overreact to unreliable information (overestimating its reliability)
and underreact to reliable information (underestimating its reliability). Evidence of
underreaction to cash flow news by individuals, and not by institutions, is provided
in Cohen et al. [471].
In an intertemporal setting, the optimal investment-consumption problem has
been analysed in Chap. 6 under the assumption that agents perfectly know the
economic model and fully observe the realization of the random variables describing
asset returns and the state of the economy. However, these two assumptions can be
significantly relaxed, allowing in particular for the presence of non-observable or
hidden economic factors (incomplete information).
Assuming incomplete information or partial observability of the economic
variables, the solution to the optimal investment-consumption problem requires
to estimate the unobserved random variables. Moreover, several implications of
the classical models change in the presence of unobservable factors. For instance,
assuming parameter uncertainty, the results obtained on return predictability can
admit different interpretations, as shown for instance in Kandel & Stambaugh [1071]
in the context of a two-period economy. In this setting, there is an estimation risk
as well as a learning risk. Estimation risk refers to the fact that an agent does
not know with certainty the parameter values, while learning risk refers to the
fact that an agent will learn more in the future, so that there is a risk that he will
discover bad news (low mean returns). Avramov [91] introduces model uncertainty,
namely uncertainty about which economic variables should appear in the return
forecasting model, and shows that this component significantly affects portfolio
choices. Barberis [146] examines how the evidence of return predictability affects
the optimal portfolio choice for investors with long horizons. The author shows that,
even after incorporating parameter uncertainty, return predictability makes investors
allocate more wealth to stocks, the longer their horizon. Moreover, there is evidence
of estimation risk.
In a continuous time setting, Gennotte [767] (see also Dothan & Feldman
[582] and Detemple [561]) shows that a certainty equivalence result is valid:
in the presence of unobservable parameters the optimal investment-consumption
problem can be solved as in the classical full information setting by replacing
the unknown parameters with their estimates. In Brennan [290] a continuous time
optimal portfolio problem is analysed assuming that the investor knows that the
investment opportunity set is constant but he is uncertain about the mean return
of the risky asset (the volatility instead is known), learning about the mean rate
of return from the observation of historical returns. This learning process affects
the portfolio problem, inducing a learning risk and a hedging demand because
the investor attempts to protect himself against learning bad news. This hedging
component is null when the utility function is logarithmic: in this case the investor
holds the same position in the risky asset as in the full information economy where
the mean return of the risky asset is equal to the investor’s current assessment.
Otherwise, the investor takes a larger or a smaller position in the risky asset in
comparison to the full information case, depending on whether he is more or less
risk tolerant than a logarithmic investor. This effect increases with the investor’s time
horizon and decreases with the volatility of the market return. The learning effect
joined by possibly predictable returns makes the relationship between investment
horizon and risky allocation quite complex (see Xia [1669]): in particular, it is no
longer true that young agents should hold more risky assets than old agents.
Incomplete information and partial observation help to address some asset
pricing puzzles. Brennan & Xia [302] analyse a representative agent economy with
dividends described by a stochastic log-normal process with a mean reverting drift
that is not directly observable but must be estimated from the realized growth rates
of dividends and aggregate consumption. The learning procedure increases stock
price volatility and allows to match the interest rate level and the equity premium
observed in U.S. capital markets with a coefficient of relative risk aversion of
15. Moreover, learning can help to resolve the apparent discrepancy between high
volatility of stock prices and low volatility of dividends and consumption. Learning
by rational agents also provides an explanation to value and size anomalies of asset
prices, see Pástor [1405] and Brennan & Xia [303].
The APT in a setting characterized by incomplete information on the parameters
of the model generating asset returns has been analysed in Handa & Linn [886].
Under some conditions, a linear relation for expected returns still holds. However,
beta coefficients and prices differ from those computed under complete information.
Agents attribute more systemic risk to an asset with low information than to an asset
with high information, leading to relatively lower betas for high information assets
in comparison to low information assets. In an equilibrium model, Coles et al. [473]
show that for low information securities the beta coefficients and expected returns
tend to be higher than without estimation risk, while the opposite is verified in the
case of high information assets.
In Kahn [1059], it is shown that assuming imperfect information and moral
hazard leads to incomplete insurance. To illustrate the phenomenon, consider a risk
averse agent who supplies unobservable labor in order to produce some output. If
labor is observable, then complete risk sharing is obtained, while, if it is not observ-
able, then there is imperfect insurance (market incompleteness). In equilibrium
agents bear idiosyncratic risk in consumption and the variance of a representative
agent’s consumption is larger than the variance of per capita consumption. This
fact contributes to a partial resolution of the equity premium puzzle. Kocherlakota
[1113] analyses a model with moral hazard under the hypothesis of complete
markets and that all the trades are public information. In this setting, it is shown
that the equity premium is lower than the one obtained in a representative agent
economy.
The asset pricing implications of market imperfections have been studied in
Heaton & Lucas [926], surveying the related literature. The main message of the
authors is that, despite the different forms of market frictions and the differences
in the models considered, market frictions allow for the resolution of some of the
anomalies encountered in the empirical tests of classical asset pricing theory.
In a continuous time setting, the mathematical theory of transaction costs has

been the subject of a considerable amount of research in mathematical finance
in the last years (see Muhle-Karbe & Guasoni [1360] for a survey on portfolio
optimization in the presence of transaction costs). Longstaff [1244] analyse the
intertemporal optimal consumption problem in continuous time under the constraint
that trading strategies have bounded variation and show that an investor facing
this type of constraint behaves very differently from an unconstrained investor.
It is shown that the optimal trading strategy consists in trading as much as
possible, whenever possible, and the investor acts as if facing borrowing and short
selling constraints. Mean-variance portfolio strategies under solvency constraints
are analysed in Nguyen & Portait [1374], while portfolio choices with illiquid assets
have been studied in Garleanu [760] and Longstaff [1245].
Liquidity constraints, margin requirements, and, in general, financial constraints
may weaken classical no-arbitrage arguments used to exclude asset pricing anoma-
lies and, therefore, may allow for an explanation of return predictability, overreac-
tion, excess volatility, inefficient allocations and also a financial crisis with diver-
gence of prices from their fundamentals, see Aiyagari & Gertler [35], Chowdhry &
Nanda [442], Shleifer & Vishny [1541], Liu & Longstaff [1230], Gromb & Vayanos
[837], Kupiec [1143], Hsieh & Miller [970], Kupiec & Sharpe [1144], Chabakauri
[383]. When short sale constraints are binding, assets exhibit high valuations and
low subsequent returns (see Jones & Lamont [1042]). As shown in Gromb &
Vayanos [837] and Liu & Longstaff [1230], arbitrageurs may even underinvest in
an arbitrage opportunity if margin requirements are present. In a related direction,
pointing out the limits of arbitrage, Shleifer & Vishny [1541] show that if arbi-
trageurs invest wealth of outsider investors, who direct their funds according to past
performance of arbitrageurs, then their activity may not be effective in exploiting
arbitrage opportunities in the market and, therefore, in reducing the gap between
asset fundamentals and prices. The reason is that performance-based funds inflow
induces the arbitrageur not to take risky investments (arbitrage opportunities). The
arbitrage argument is particularly ineffective in extreme circumstances.
A form of market friction is represented by the presence of taxation. No-arbitrage
analysis has been extended to an economy with taxes in Dybvig & Ross [610]
and Ross [1469]. On the optimal consumption-portfolio problem with capital gains
taxation see Constantinides [489]. A general equilibrium model with taxes has been
proposed in Basak & Croitoru [176]. Michaely et al. [1338], Michaely & Vila
[1337] show that taxation generates trading volume around ex-dividend days, while
transaction costs reduce trading volume. Moreover, trading volume is positively
related to the degree of tax heterogeneity. Supporting empirical evidence is also
provided.
Considering the model proposed in Hong & Stein [962] with differences of
opinions as well as short sale constraints, Chen et al. [418] show that the number of
short sale constrained agents can be regarded as a proxy of pessimism in the market
and, therefore, predicts low future returns. Moreover, this proxy and, therefore, short
sale constraints themselves are strictly related to the momentum effect. Lamont et al.
[1163] shows the relevance of financing frictions-constraints of a firm in the cross-

sectional analysis of asset returns (see however Gomes et al. [811] for a different
conclusion).
9.7 Exercises
Exercise 9.1 By relying on arguments similar to those used in the proof of

Proposition 2.10, complete the proof of Proposition 9.3.
Exercise 9.2 Consider a preference functional of the cumulative prospect theory
form (9.4). Let xQ 1 and xQ 2 be two gambles on the same finite probability space:
xQ 1 D fx1 ; : : : ; xS I 1 ; : : : ; S g and xQ 1 D fx1 ; : : : ; xS I 10 ; : : : ; S0 g;

P PS 0
with x1 < : : : < xS and SsD1 s D sD1 s D 1. Suppose that xQ 1 dominates
xQ 2 in the sense of first order stochastic dominance (see Sect. 2.3). Show that, for
any choice of the value and of the weighting functions u./ and w./, the preference
functional (9.4) satisfies U.Qx1 / U.Qx2 / (compare also with Hens & Rieger [938,
Proposition 2.39]).
Exercise 9.3 Consider the Barberis et al. [150] model described by equations (9.5)–
(9.9). In this exercise, following the Appendix of Barberis et al. [150], we provide
the proof of Proposition 9.5. Denote by .w; c/ WD ..wt ; ct //t2N a self-financing
trading-consumption strategy, where the investment strategy wtC1 is parameterized
in terms of the amount of wealth invested in the single risky asset between date t
and date t C 1. In the context of the present exercise, we assume that a trading-
consumption strategy .w; c/ is admissible if wt is bounded, for all t 2 N. Similarly
as in relations (6.4) and (6.19), it is easy to see that a trading-consumption strategy
.; c/ satisfies the following self-financing condition:
WtC1 D .Wt ct /rf C wtC1 .rtC1 rf / C ytC1 ; for all t 2 N;
where .Wt /t2N denotes the self-financing wealth process associated with the strategy
.w; c/ and .yt /t2N is the exogenous income stream.
(i) Let .w ; c / D ..wt ; ct //t2N be a trading-consumption strategy satisfying the
above self-financing condition. Let .w; c/ D ..wt ; ct //t2N be an alternative self-
financing strategy and define the strategy

.w C ˛w; c C ˛c/ D .wt C ˛wt ; ct C ˛ct / t2N ;
for ˛ 2 RC . Show that, according to the preference functional (9.7), the

difference in the expected utilities associated to .c C˛c; w C˛w/ and .c ; w /
9.7 Exercises 575
satisfies
" 1
#
X
E ı t
u.ct C ˛ct / C bt ı tC1
v.XtC1 ; wtC1 C ˛wtC1 ; zt /
tD0
" 1
#
X
E ı t
u.ct / C bt ı tC1
v.XtC1 ; wtC1 ; zt /
tD0
"1 #
X
t 0
.˛c; ˛w/ WD E ı u .ct /˛ct C bt ı ˛wtC1 v.r
tC1
O tC1 ; zt / ;
tD0
where the function vO is defined as in the statement of Proposition 9.5.

(ii) Suppose that, for all t 2 N, it holds that
u0 .ct / D rf ıEŒu0 .ctC1 /jFt ; (9.81)

0
u .ct / D ıEŒu 0
.ctC1 /rtC1 jFt C ıbt EŒv.r
O tC1 ; zt /jFt ; (9.82)
with u.x/ D x1 =.1 / and where the function v./ O is defined as in

Proposition 9.5 and the process .bt /t2N is specified by bt D b0 cN t , for all
t 2 N. Show that, for all t 2 N, it holds that
h ˇ i
ˇ
E u0 .c O tC1 ; zt /jFt D u0 .c
t /˛ct C ˛wtC1 ıbt v.r
0
t /˛Wt ıE u .ctC1 /˛WtC1 ˇFt ;
(9.83)
where .Wt /t2N is the wealth process associated to the self-financing strategy
.w; c/.
(iii) Deduce that
.˛c; ˛w/ D u0 .c0 /˛W0 lim ı T EŒu0 .cT /˛WT :

T!C1
As explained in Barberis et al. [150, Appendix], the right-hand side of the last
equality vanishes if W0 D 0 and the following condition holds:
2 c2 2c d C d2

log ı gc C gd C < 0:
2
This shows that the Euler conditions (9.81)–(9.82) are necessary and sufficient
for the optimality of the strategy ..t ; ct //t2N .
(iv) Show that the risk free return rf given in (9.10) satisfies the Euler condition
(9.81) when the consumption process .ct /t2N is given by the aggregate per
capita consumption process .Nct /t2N .
(v) Suppose that the equilibrium price-dividend ratio of the risky asset satisfies
pt
D f .zt /; for all t 2 N;
dt
for a suitable function f ./. Show that

ˇ
ˇ
cN tC1 ˇ 2 2 .1 2 / 1 C f .ztC1 / .d c /"tC1 ˇˇ
E rtC1 ˇFt D egd gc C c 2 E
cN t ˇ f .z /
e ˇFt :
t
Deduce that condition (9.11) of Proposition 9.5 implies that the Euler
optimality condition (9.82) holds.
(vi) Deduce that the strategy of consuming ct D cN t D dt C yt , for all t 2 N,
and holding the market supply of the risky asset satisfies the Euler conditions
(9.81)–(9.82) and the claim of Proposition 9.5 follows.
Exercise 9.4
(i) Consider the general form of recursive preferences as defined in (9.16). Suppose
that the functions v and uQ are of the following form:
v.c; y/ D u.c/ C ıy and uQ .x/ D x:
Show that in this case the preference functional (9.16) reduces to the classical
time additive utility.
(ii) Consider the recursive utility functional of the Epstein & Zin [651] form given
in (9.19). Show that, if ˛ D %, then the recursive preference functional reduces
to the classical time additive expected utility with a power utility function.
Exercise 9.5 In this exercise, we prove Proposition 9.8, considering a recursive
preference functional of the form (9.19), with UT .c; / D u.c/ D .1 ı/1=.1/ c.
(i) By following the arguments given in the proof of Proposition 6.4, show that
the Euler conditions (9.23)–(9.24) hold.
(ii) Denoting by .Wt /tD0;1;:::;T and .ct /tD0;1;:::;T the optimal wealth and consump-
tion processes, respectively, in the maximization of the recursive preference
functional U0 .c; / defined in (9.19), define the adapted stochastic process
.
t /tD0;1;:::;T as the solution to the recursive relation
!%˛ %

V.WtC1 ; t C 1/ ctC1

tC1 WD ı

t and
0 D 1:
Consider the backward stochastic difference equation

ˇ

tC1 ˇ V.WT ; T/
Xt D c
t CE XtC1 ˇˇFt ; for t D 0; 1; : : : ; T1; with XT D ;

t u0 .c
T/
9.7 Exercises 577
the solution of which is an adapted stochastic process .Xt /tD0;1;:::;T . Show that
this backward stochastic difference equation admits a unique solution which
is given by Xt D Wt , for all t D 0; 1; : : : ; T. Note also that the process
.
t /tD0;1;:::;T corresponds to the stochastic discount factor introduced in (9.29).
(iii) Deduce that the optimal wealth process .Wt /tD0;1;:::;T satisfies relation (9.25)
and that the return process .rt /tD1;:::;T associated to the optimal wealth process
can be represented as in (9.26).
(iv) Deduce the validity of the Euler condition (9.27).
Exercise 9.6 In this exercise, we prove Proposition 9.9.
(i) As a first step, show that relation (9.23), suitably rewritten in terms of the
optimal return .rt /tD1;:::;T process, implies that
r2 %2 c2
r D log ı C %c C %cr :
2 2
(ii) Using part (i) of the exercise together with relation (9.23) with respect to the
risk free asset with return rf , deduce that the equilibrium risk free rate rf is
given by (9.33).
(iii) Suppose furthermore that .log.ctC1 =ct /; log rtC1
1
; : : : ; rtC1
N
/ are jointly nor-
mally distributed conditionally on Ft , for all t D 0; 1; : : : ; T 1. Deduce
that asset risk premia satisfy relation (9.34).
Exercise 9.7 (An approximate formula for the optimal demand of the risky asset
with recursive preferences, see Campbell & Viceira [354]). Consider the recursive
preference functional defined in (9.19), with parameters ˛ 2 .0; 1/, % 2 .0; 1/ and
ı 2 .0; 1/, in an economy with a single risky asset with return process .rt /tD1;:::;T
and a risk free asset delivering the constant return rf . Denote by .t /tD0;1;:::;T a
trading strategy, measured in terms of the number of units of the risky asset held in
the portfolio, and let the couple . ; c / be the trading-consumption self-financing
strategy associated with the maximization of the recursive preference functional
(9.19). Letting .Wt /tD0;1;:::;T be the associated wealth process, the corresponding
optimal return process .rt /tD1;:::;T satisfies

WtC1 D .Wt ct /rtC1

; for all t D 0; 1; : : : ; T 1:
As mentioned after Proposition 9.9, relation (9.34) holds as an approximation for

the risk premium of the risky asset, i.e.,
Var.log rtC1 jFt /

EŒlog rtC1 jFt log rf C .1 / Cov.log rtC1 ; log rtC1 jFt /
2
C % Cov. log ctC1 ; log rtC1 /;
(9.84)
for all t D 0; 1; : : : ; T 1, where log ctC1 WD log ctC1 log ct denotes the
logarithmic growth rate of the optimal consumption process.

(i) Starting from the relation WtC1 D .Wt ct /rtC1

, prove the following log-
linear approximation of the self-financing condition:

1
log WtC1 log rtC1 C 1 .log ct log Wt / C K;

for all t D 0; 1; : : : ; T 1, where
log.1 /
K WD log C .1 / and WD 1 eEŒlog ct log Wt :

(ii) Suppose that the return process .rt /tD1;:::;T associated to the strategy . ; c /
satisfies the following approximate relation:

tC1
log rtC1 tC1 .log rtC1 log rf / C log rf C .1 tC1 / Var.log rtC1 jFt /;
2
(9.85)
for all t D 0; 1; : : : ; T 1. By relying on part (i) of the exercise, show that the
asset pricing relation (9.84) can be rewritten in the following form:
Var.log rtC1 jFt /

EŒlog rtC1 jFt log rf C .1 /tC1 Var.log rtC1 jFt /
2

C% Cov.log rtC1 ; log ctC1 log WtC1
jFt / C tC1 Var.log rtC1 jFt / :
(9.86)

(iii) Deduce that the optimal demand tC1 of the risky asset satisfies the following
approximate relation:
Var.log rtC1 jFt /

1 EŒlog rtC1 jFt log rf C 2
tC1
˛ Var.log rtC1 jFt /
˛ 1 % Cov.log rtC1 ; log ctC1 log WtC1

jFt /
:
˛ %1 Var.log rtC1 jFt /
This approximate relation shows that the optimal demand of the risky asset can
be decomposed into two terms: a first term which depends exclusively on the
risk premium of the asset and a second term which reflects the intertemporal
hedging demand (see Campbell & Viceira [354, Section III] for more details).
Exercise 9.8 Consider the habit formation preferences introduced in (9.35)–(9.36),
with respect to a general utility function u.
9.7 Exercises 579
(i) Derive the Euler condition (9.39) (compare also with Munk [1363, Theo-
rem 6.9]).
(ii) Consider the habit formation preferences introduced in (9.40). Show that in this
case the general Euler condition (9.39) reduces to (9.41).
Exercise 9.9 Consider the habit preferences introduced in (9.43)–(9.44). In this
exercise, following the original paper Abel [4], we derive the Euler conditions and
the equilibrium price-dividend ratio of a risky asset.
(i) For a consumption process c D .ct /tD0;1;:::;T , let us denote
X
T
U.c/ D ı t u.ct ; zt /:
tD0
As a first step, show that, for all t D 0; 1; : : : ; T 1, it holds that

1˛ 1˛ ! 1˛
@ ctC1 zt ct 1
U.c/ D ı t 1 ıˇ : (9.87)
@ct ct ztC1 zt ct
(ii) Let the process .dt /tD0;1;:::;T represent the total amount of consumption good
per capita delivered by the capital stock (dividend). Assume that, in equilib-
rium, all dividend is consumed as soon as it is delivered. If we furthermore
assume that all the agents of the economy are identical, this implies that
ct D cN t D dt , for all t D 0; 1; : : : ; T, so that
ctC1 cN tC1 dtC1

D D DW xtC1 ;
ct cN t dt
where the process .xt /tD1;:::;T represents the growth rate of the dividend.
Rewrite (9.87) as follows:
@
.1˛/
U.c / D ı t 1 ıˇx1˛ x
tC1 t z˛1 .ct /˛ (9.88)
@ct t
and deduce that the following Euler condition holds:

" .1˛/ ˇ #
1˛
1 ıˇxtC2 xtC1 ˇ
xtC1 rtC1 ˇˇFt
.˛1/ ˛
ıE .1˛/
xt D 1;
1˛
1 EŒıˇxtC1 xt jFt
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N, where .rt /tD1;:::;T is the return

process of the stock.
(iii) Let the process . pt /tD0;1;:::;T denote the equilibrium price process of the stock.
Show that the equilibrium price dividend ratio process . pt =dt /tD0;1;:::;T satisfies
" ˇ #
1˛ .1˛/ ˇ
pt 1 ıˇxtC2 xtC1 ptC1
xtC1 ˇˇFt ;
.˛1/ ˛
D ıE .1˛/
xt xtC1 1 C
1˛
dt 1 EŒıˇxtC1 xt jFt dtC1
for all t D 0; 1; : : : ; T 1.
Exercise 9.10 Consider the Campbell & Cochrane [343] model with external habit
preferences, where the surplus process and the consumption growth process satisfy
the dynamics (9.46)–(9.47), respectively. Prove Proposition 9.10.
Exercise 9.11 Consider the De Long et al. [549] model described in Sect. 9.3 with
a proportion 2 .0; 1/ of noise traders. Prove Proposition 9.11.
Exercise 9.12 Consider the De Long et al. [549] model introduced in Sect. 9.3 and
extended as in De Long et al. [549, Section III.B] to the case where the risky asset
yields at each date t 2 N a dividend dt equal to
dt D rf C "Qt ; for all t 2 N;
where .Q"t /t2N is a sequence of i.i.d. normal random variables with zero mean and
constant variance "2 , independent of the noise traders’ misperceptions .%t /t2N . By
following the arguments given in Exercise 9.11, show that:
(i) the optimal demand of the risky asset by rational and noise traders is
respectively given by
rf C EŒ ptC1 jFt .1 C rf /pt

tC1
r;
D ;
2.Var. ptC1 jFt / C "2 /
rf C EŒ ptC1 jFt .1 C rf /pt %t
tC1
n;
D C I
2.Var. ptC1 jFt / C "2 / 2.Var. ptC1 jFt / C "2 /
(ii) the steady-state equilibrium price process . pt /t2N of the risky asset satisfies
!
t %N 2 2 %2 t .%t %/
N
pt D1C "2 C C I
rf rf .1 C rf /2 1 C rf
(iii) in correspondence of a steady-state equilibrium, the difference EŒ RtC1 ./

in the expected optimal return between the noise and the rational traders is
9.7 Exercises 581
given by
%N 2 C %2
EŒ RtC1 ./ D %N %2

"2
2 .1Crf /2
C
if > 0 and EŒ RtC1 ./ D %N if D 0, for all t 2 N.

Exercise 9.13 Consider the model proposed in Daniel et al. [516] and described
in Sect. 9.3, with uninformed agents and overconfident informed agents. Prove
Proposition 9.12.
Exercise 9.14 Consider the model proposed in Daniel et al. [516, Section III] and
described in Sect. 9.3, with uninformed agents and overconfident agents, assuming
that the overconfidence of informed traders depends on the outcome of the public
signal sQ2 released at date t D 2. Prove Proposition 9.13.
Exercise 9.15 Consider a risk neutral representative agent economy under the
assumption of rational expectations, with a risky security paying the dividend stream
.dt /t2N , which is supposed to follow a first order autoregressive process of the
form (9.63). Assume furthermore that the dividend process .dt /t2N is predictable,
meaning that the random variable dtC1 is Ft -measurable, for every t 2 N. Starting
from equation (9.62), show that the rational expectations solution for the price
process .st /t2N of the risky security is given by (9.64).
Exercise 9.16 Consider the Weil [1650] model introduced in Sect. 9.5. By applying
Jensen’s inequality, prove Proposition 9.15.
Chapter 10
Financial Markets Microstructure
Noise trading is essential to the existence of liquid markets [: : :]

Noise makes financial markets possible, but also makes them
imperfect [: : :] Noise creates the opportunity to trade profitably,
but at the same time makes it difficult to trade profitably.
Black (1986)
In the previous chapters, we have considered financial markets as abstract places

where trading occurs. In reality, financial markets are not idealized entities and
are characterized by a well defined institutional setting. In particular, taking
the institutional setting into account is fundamental in order to understand the
functioning of real financial markets. To this effect, it is important to analyse the
different roles of market makers, dealers and brokers. These types of investors trade
according to their own objectives and they also coordinate and supervise the market
by making sure that the demand meets the supply.
In this chapter, we focus our attention on financial markets microstructure by
considering the effects of different institutional settings on the functioning of
financial markets. To this purpose, we relax the Walrasian-perfectly competitive
market hypothesis and introduce market models with imperfect competition. We
shall be concerned with the following features of financial markets:
a) Order-driven and quote-driven (or price-driven) markets. In the case of an
order-driven market, buying and selling orders arrive in the market and prices
are then defined through some automatic mechanism matching the orders or
through the action of market makers who supervise the market and set prices
with the commitment to satisfy the net market order at that price. In the case
of a quote-driven market, some agents (dealers) provide quotes for bid prices
(corresponding to sell orders) and ask prices (corresponding to buy orders) and,
under some conditions, are committed to satisfy market orders at the quoted
prices.
b) Consolidated and fragmented markets. This feature reflects the fact that, in some
cases, trades occur in a single market (consolidated market), while, in other
cases, it is possible to trade the same asset in more than one market (fragmented
markets).

584 10 Financial Markets Microstructure
c) Broker and dealer markets. This distinction concerns the fact that brokers act
only for their own customers while dealers also trade on their own account.
d) Possible competition among market makers (dealers), meaning that several
market makers can manage the same security simultaneously.
Moreover, one can also consider discrete time or continuous time markets,
meaning that market orders can be assumed to arrive either in correspondence of
a finite set of predetermined dates or at any time during the trading period. If
an auction market functions in discrete time, then orders (in whatever form they
are specified) arrive in the market and are matched among themselves or satisfied
thanks to the intervention of market makers. On the other hand, if an auction market
functions in continuous time, so that orders can arrive at any point in time, it is
typically assumed that market makers determine the trading price and clear the
markets in correspondence of a set of predetermined dates (batch auction market).
We refer to Foucault et al. [735, Chapter 1] for a complete overview of different
market structures and trading mechanisms.
One of the most interesting perspectives to analyse the market is represented by
welfare analysis. Even though this is not always an easy task, welfare analysis can
be useful to evaluate some of the features of financial markets, for instance:
a) information revelation, namely the possibility that market prices aggregate
and transmit information in the context of non-perfectly competitive financial
markets;
b) price discovery, meaning that in some markets agents do not observe trading
prices before sending orders to the market;
c) execution risk, consisting in the risk due to the possible uncertainty of whether
and when an order will be executed;
d) market liquidity, where liquidity is generically understood as the cost that an
agent must support when buying and selling the same asset at the same time.
Typically, the bid-ask spread is taken as a measure of market liquidity, but
liquidity is not always totally reflected in the bid-ask spread and is also related
to the trading volume;
e) market depth, meaning the capacity of the market to absorb large market orders
without producing large price changes;
f) transparency, consisting in the capability of agents to recover information
concerning trades in the market (prices of executed trades, bid and ask prices,
trading volume, identity of the agents).
This chapter is structured as follows. In Sect. 10.1, by relying on an equilibrium
analysis, we discuss the interplay between private information and imperfect
competition. In Sect. 10.2, we present the seminal Kyle (1985) model and several
extensions of it. Section 10.3 deals with the modeling of quote-driven markets.
In Sect. 10.4, we provide a brief overview on the dynamic modeling of market
microstructure, also discussing the existence of intraday and intraweek regularities
in asset prices. Section 10.5 is devoted to a discussion of insider trading behavior,
while Sect. 10.6 contains a survey on the effects of the different institutional features
10.1 The Role of Information Under Imperfect Competition 585
of financial markets. At the end of the chapter, we provide a guide to further readings
as well as a series of exercises.
In this chapter, if xQ and yQ are two random variables, we shall denote (with some
abuse of notation) by EŒQxjQy the conditional expectation of xQ given yQ , viewed as a
random variable, and by EŒQxjy WD EŒQxjQy D y the conditional expectation of xQ given
the realization y of the random variable yQ .
10.1 The Role of Information Under Imperfect Competition
In Chap. 8, we have analysed the role of information in the context of perfectly

competitive markets, where agents are assumed to be price takers. In particular,
we have focused our attention on whether the prices associated to a Green-Lucas
equilibrium of the economy aggregate and transmit the agents’ private information.
In this section, we are going to address the same problem in the context of an
imperfectly competitive market, focusing on the model introduced in Kyle [1148]. In
this model, an imperfectly competitive market with private information is analysed,
assuming that agents can submit to a central auctioneer a demand function which
depends on the asset price and on the possible observation of a private signal. In
an imperfectly competitive market, asset prices are affected by the agents’ demand.
Agents are aware of this effect and, therefore, they do not behave as price takers but
act strategically, exploiting the dependence of market prices on their demand. This
argument induces a rational agent to maximize his expected utility considering the
informational content of prices as well as the effect of his demand on market prices.
The model proposed in Kyle [1148], which has some similarities to the model of
Grossman [839] discussed in Chap. 8, provides a partial equilibrium analysis in a
single-period setting (i.e., t 2 f0; 1g). It is assumed that there is a single risky asset,
with liquidation value (dividend) dQ at date t D 1, traded at date t D 0 at some market
clearing price p. The economy is populated by three homogeneous classes of agents:
noise traders, informed speculators and uninformed speculators. Noise traders trade
an aggregate quantity of the asset represented by the random variable zQ, capturing
the fact that the aggregate demand of noise traders is exogenous and does not result
from a maximizing behavior. There are N informed speculators n D 1; : : : ; N and
each informed speculator receives a private signal yQ n , supposed to be of the form
yQ n D dQ C "Qn ; for every n D 1; : : : ; N;
Q zQ, "Q1 ; : : : ; "QN are mutually independent and normally

where the random variables d,
distributed with variances
Q D d1 ;
Var.d/ Var.Qz/ D z2 and Var.Q"n / D "1 ;
for all n D 1; : : : ; N, with d and " representing the precision. The hypothesis
that the random variable zQ is independent from the other random variables means
that noise trading has no informational content, while the assumption that the
random variables "Q1 ; : : : ; "QN are independent simply means that different informed
speculators have access to independent private information. After observing a
realization yn of his private signal yQ n and conditionally on the observation of the
market price p, the demand of the n-th informed speculator is given by Xn .p; yn /, for
n D 1; : : : ; N.
Besides noise traders and informed speculators, there are M uninformed specula-
tors (m D 1; : : : ; M), who do not receive private signals. The demand of uninformed
speculators conditionally on a market clearing price p is given by the quantities
Ym .p/, for m D 1; : : : ; M. Similarly as in Chap. 8, uninformed agents have only
access to the public information represented by the observation of the market
price. All speculators have deterministic initial endowment, normalized at zero for
simplicity.
Every speculator is characterized by an exponential utility function, with a
coefficient of absolute risk aversion given by ˛inf or ˛un , for informed or uninformed
speculators, respectively. Hence, given the demand Xn .p; yn / and Ym .p/ of an
informed speculator and of an uninformed speculator, respectively, the realized
utilities at date t D 1 in correspondence of a market clearing price p and of the
observation of the signal yn are given by
˛inf .dp/Xn .p;yn /
n .d p/Xn .p; yn / WD e
uinf ; for all n D 1; : : : ; N;
and
˛un .dp/Ym .p/
m .d p/Ym .p/ WD e
uun ; for all m D 1; : : : ; M;
where d denotes a generic realization of the random variable d. Q In the setting

considered in Kyle [1148], the demand schedules Xn .; yn / and Ym ./ (i.e., the
demands of informed and uninformed speculators, respectively, viewed as functions
of the price and, in the case of informed agents, given the observation of the private
signal) are allowed to be arbitrary convex upper-semicontinuous correspondences,
taking values in Œ1; C1. In particular, such a general specification includes the
order trades typically taking place in organized exchanges, such as market orders,
limit orders and stop orders (together with all the linear combinations thereof).
In this setting, under the assumption of zero aggregate supply of the risky asset,
the market clearing condition in correspondence of a price p requires that
X
N X
M
Xn .p; yQ n / C Ym .p/ C zQ D 0: (10.1)
nD1 mD1
Similarly as in Sect. 8.1, a perfectly competitive Green-Lucas rational expec-

tations equilibrium consists of an equilibrium price functional W RNC1 ! R
and of a family of demand schedules X1 .; /; : : : ; XN .; /; Y1 ; ./; : : : ; YM ./ such that
the market clearing condition (10.1) holds almost surely in correspondence of the
equilibrium price pQ D .Qy1 ; : : : ; yQ N ; zQ/ and

ˇ
Q Q /Xn .Qp ; yQ n / ˇpQ D .y1 ; : : : ; yn ; z/; yQ n D yn
n .d p
E uinf
ˇ
E uinf Q Q /Xn0 .Qp ; yQ n / ˇpQ D .y1 ; : : : ; yn ; z/; yQ n D yn ;
n .d p
h ˇ i (10.2)
Q Q /Ym .Qp / ˇpQ D .y1 ; : : : ; yn ; z/
m .d p
E uun
h ˇ i
Q Q /Ym0 .Qp / ˇpQ D .y1 ; : : : ; yn ; z/ ;
m .d p
E uun
for all n D 1; : : : ; N and m D 1; : : : ; M, for every realization .y1 ; : : : ; yN ; z/

of the random variables .Qy1 ; : : : ; yQ N ; zQ/ and where X10 .; /; : : : ; XN0 .; /; Y10 ; ./;
0
: : : ; YM ./ is any alternative family of demand schedules. Condition (10.2) means
that both informed and uninformed speculators maximize their expected utilities
taking market prices as given and conditioning on the observation of the market
price. Note that both informed and uninformed agents use their knowledge of
the equilibrium price functional to extract all the available information from
the market price. Moreover, informed agents also condition on the realization
of their own private signal. This notion of equilibrium is in line with the perfect
rationality hypothesis, according to which every agent fully exploits all the available
information, under the perfect competition hypothesis. Observe also the similarity
between the present notion of equilibrium and the results presented in Sect. 8.3.
In the setting considered in Kyle [1148], relaxing the perfect competition
assumption, each trader submits to a central auctioneer his demand schedule.
The auctioneer aggregates the demand schedules submitted by all the traders in
the economy, sets a price which satisfies the market clearing condition (10.1)
and allocates quantities in order to satisfy the traders’ demands (see Kyle [1148,
Section 3] for a description of the way the auctioneer operates in the case of an
infinite demand by some traders or in the case of a multiplicity of clearing prices).
In this context, imperfect competition is present since every trader realizes that,
by submitting to the central auctioneer a demand schedule, he can influence the
resulting equilibrium price. Speculators are assumed to be rational and, therefore,
they are aware of the fact that their behavior can influence the market price, since
by changing his demand, an agent changes both the quantity he trades and the
market clearing price at which he trades that quantity. In order to emphasize the
dependence of the equilibrium price on the demand schedules submitted by the
informed and uninformed speculators we let pQ D .X; Y; z/, where X and Y
are the vectors of demand schedules defined by X WD .X1 .; /; : : : ; XN .; // and
Y WD .Y1 ./; : : : ; YM .//.
The functional admits a conjectural as well as an institutional interpretation.
The conjectural interpretation follows the reasoning presented in Chap. 8 when
discussing the Green-Lucas equilibrium. Speculators are aware of the fact that their
behavior affects market prices and they conjecture a functional relating their demand
to the market price. Having conjectured a price functional, speculators then take
their decisions according to such a conjecture and act strategically (knowing also
the other agents’ conjectures and that other agents do the same). In equilibrium, the
agents’ conjectures are confirmed. According to the institutional interpretation, the
functional derives from a specific price formation rule determined by the market
institutional setting. Agents know it and exploit it when defining their demand
schedules.
The above discussion leads us to introduce the notion of rational expectations
equilibrium under imperfect competition, defined as a couple .X; Y/ of vectors of
demand schedules X D .X1 .; /; : : : ; XN .; // and Y D .Y1 ./; : : : ; YM .// such that
the market clearing condition (10.1) is satisfied in correspondence of the equilibrium
price pQ D .X; Y; z/ and
ˇ
E uinf Q Q /Xn .Qp ; yQ n / ˇpQ D .X; Y/; yQ n D yn
n .d p
ˇ
E uinf Q Q /Xn.n/ .Qp ; yQ n / ˇpQ D .X .n/ ; Y/; yQ n D yn ;
n .d p
h ˇ i (10.3)
Q ˇ
E uun
m . d Q
p /Ym .Q
p / pQ D .X; Y/
h ˇ i
E uun Q Q /Ym.m/ .Qp / ˇpQ D .X; Y .m/ / ;
m .d p
for all n D 1; : : : ; N and m D 1; : : : ; M, for every realization .y1 ; : : : ; yN ; z/ of the

random variables .Qy1 ; : : : ; yQ N ; zQ/. In (10.3), X .n/ denotes an arbitrary demand vector
.n/ .n/ .n/
X .n/ D .X1 ; : : : ; XN / such that Xj ¤ Xj only for j D n. Similarly, Y .m/ denotes
.m/ .m/ .m/
an arbitrary demand vector Y .m/ D .Y1 ; : : : ; YM / such that Yj ¤ Yj only for
j D m. Condition (10.3) defines a Nash equilibrium in trading strategies: given his
private information, each (informed) speculator chooses a demand schedule in order
to maximize his expected utility, taking into account the effect of his demand on the
market price and taking as given the demand functions of the other agents.
In Kyle [1148], the analysis is focused on symmetric linear equilibria, namely
couples .X; Y/ of demand schedules defining a rational expectations equilibrium
under imperfect competition, so that conditions (10.1) and (10.3) are satisfied, and
such that
Xn .p; yn / D inf C ˇyn inf p;

(10.4)
Ym .p/ D un un p;
for some constants inf , ˇ, inf , un and un , for all n D 1; : : : ; N and m D 1; : : : ; M,

for all realizations p and yn of the market clearing price and of the random signal,
respectively. As shown in Exercise 10.1, the market clearing condition (10.1)
implies that, in correspondence of a symmetric linear equilibrium, the equilibrium
price pQ must satisfy
!
X
N
pQ D Nˇ dQ C ˇ

"Qn C zQ C Ninf C Mun ; (10.5)
nD1
where WD 1=.N inf C Mun / is a measure of market depth (i.e., the order flow
necessary to induce the equilibrium price to rise or fall by one unit).
In particular, the linear structure (10.4) allows for a clear analysis of the
informativeness of the equilibrium price (see Kyle [1148, Section 4]). To this effect,
let us define F WD 1= Var.djQ Q y1 ; : : : ; yQ N / and note that, due to the assumption
of multivariate normality, the quantity Var.djQ Q y1 ; : : : ; yQ N / is constant and does
not depend on the realization .y1 ; : : : ; yN / of the conditioning random variables
.Qy1 ; : : : ; yQ N /. The quantity F can be regarded as a measure of the precision of the
forecast of the liquidation value dQ given all private information in the economy. The
distributional assumptions on .Qz; d; Q yQ 1 ; : : : ; yQ N / imply that
F D d C N" :
Recalling that d measures the precision of the prior forecast of dQ (i.e., the reciprocal
of the unconditional variance of d), Q this shows that the full information precision F
is equal to the prior precision d plus " units of precision for each private signal. Of
course, informed and uninformed speculators do not observe all the realizations
.y1 ; : : : ; yN / of the private signals but only the market price and, in the case of
informed agents, their own private signals. We can define as follows the precisions
corresponding to an informed and to an uninformed agent, respectively:
1 1
inf WD and un WD ;
Q p ; yQ
Var.djQ n/
Q p /
Var.djQ
for all n D 1; : : : ; N. Again, due to the assumption of multivariate normality, the

quantities inf and un are constant and do not depend on the realizations. Moreover,
inf is the same for every informed agent (i.e., it does not depend on n). The following
proposition, corresponding to Kyle [1148, Theorem 4.1, Corollaries 4.1 and 4.2],
characterizes the informativeness of the equilibrium price (see Exercise 10.1 for the
proof).
Proposition 10.1 In the context of the above model (see Kyle [1148]), it holds that
inf D d C " C 'inf .N 1/" and un D d C 'un N" ; (10.6)
where 'inf and 'un are two constants belonging to the interval Œ0; 1. Moreover, in
correspondence of a symmetric linear equilibrium .X; Y/ of the form (10.4), it holds
that
.N 1/ˇ 2 Nˇ 2
'inf D and 'un D : (10.7)
.N 1/ˇ 2 C z2 " Nˇ 2 C z2 "
The expected dividend for informed and uninformed speculators is respectively

given by

.1 'inf /"
Q p ; yQ n D 'inf " pQ
EŒdjQ yQ n C Ninf Mun ;
inf ˇinf
(10.8)
Q 'un " pQ
EŒdjQp D Ninf Mun ;
ˇun
where WD 1=.N inf C Mun / is the slope of the aggregate excess demand schedule
and the constants inf , ˇ, inf , un and un are as in (10.4). Furthermore, if
inf D un D 0, then EŒdjQQ p D pQ holds if and only if 'un " D ˇun . Finally,
in correspondence of a symmetric linear equilibrium, the informational efficiency
parameters 'inf and 'un satisfy 0 'inf 'un 1, with
.N 1/'un N'inf
'inf D ; 'un D ;
N 'un N .1 'inf /
(10.9)
1
'un 'inf D 'un .1 'inf /
N
and inf un D .1 'un /.1 'inf /" .

In the above proposition, the parameters 'inf and 'un measure the informational
efficiency with which the equilibrium price aggregates the private information of
informed speculators. More precisely, 'un represents the fraction of the precision of
the N informed speculators revealed by the market price to uninformed speculators.
Analogously, 'inf represents the fraction of the precision of the signals of the other
N 1 informed agents transmitted by the equilibrium price to each informed agent.
Note also that the informational efficiency parameters 'inf and 'un depend only on
ˇ and not on the parameters appearing in (10.4). This can be easily understood by
recalling that ˇ represents the sensitivity of the informed speculators’ demand with
respect to their private information, as shown in (10.4). If 'inf D 0 or 'un D 0, then
the equilibrium price does not reveal any information. On the contrary, if 'inf D 1
or 'un D 1, then the price is fully revealing.
In Kyle [1148, Theorem 7.1] it is shown that the value of 'inf associated to
an equilibrium under imperfect competition is less than the value obtained in the
corresponding perfectly competitive equilibrium (i.e., when agents do not perceive
any influence of their actions on market prices). The intuition for this result is rather
simple: an informed speculator facing an upward-sloping residual supply curve will
restrict the quantity he trades, thus reducing his demand elasticity with respect to his
private information (i.e., ˇ decreases). In turn, this makes the price less informative
under imperfect competition in comparison to the case of perfect competition.
The following proposition concerns the existence and the uniqueness of a
symmetric linear equilibrium of the economy (see Kyle [1148, Theorem 5.1]).
Proposition 10.2 In the context of the above model (see Kyle [1148]), suppose that
z2 > 0 and " > 0. If N 2 and M 1, or if N 3 and M D 0, or if M 3
and N D 0, then there exists a unique symmetric linear equilibrium. If N D 1 and
M 2, then a unique symmetric linear equilibrium exists if M is sufficiently large
and does not exist if ˛un is sufficiently large. If N C M 2, then a symmetric linear
equilibrium does not exist.
Proof We shall only give an outline of the proof, referring to the original paper Kyle
[1148] for full details. Suppose first that there exists a symmetric linear equilibrium
of the form (10.4). As shown in Exercise 10.1, the market clearing condition
(10.1) implies that the equilibrium price pQ must be of the form (10.5). Hence,
a first necessary condition for the existence of an equilibrium is that the quantity
D 1=.N inf C Mun / must be well defined, i.e., N inf C Mun ¤ 0, because
otherwise condition (10.1) would imply an infinite equilibrium price, in which case
the demand schedules Xn .; / and Ym ./ would not be optimal and, therefore, could
not correspond to an equilibrium. Furthermore, again the market clearing condition
(10.1) allows to define a residual supply curve for each speculator, of the form
pQ D pQ inf;n C inf xQ inf;n and pQ D pQ un;m C un xQ un;m ;
where xQ inf;n and xQ un;m denote the demand of the n-th informed agent and of the m-
th uninformed agent, respectively, for each n D 1; : : : ; N and m D 1; : : : ; M, and
where
1 1
inf D and un D ;
.N 1/ inf C Mun N inf C .M 1/un
and
0 1
B X
N
C
pQ inf;n D inf B
@ˇ yQ k C zQ C .N 1/inf C Mun C
A;
kD1
k¤n
!
X
N
pQ un;m D un ˇ yQ n C zQ C Ninf C .M 1/un :
nD1
For an equilibrium to be well defined, it is also necessary that the quantities inf and
un introduced above are well defined (i.e., the residual supply curves must have
finite slopes). Considering then the optimization problem faced by an informed
speculator, the residual supply curve introduced above, together with the linear
demand schedule (10.4), implies that
.1 C inf inf /Qp D pQ inf;n C inf ˇQyn C inf inf :

As shown in Kyle [1148, Section 5], in this case the demand function Xn .; / of an
informed speculator can be written in the following form:
Q ; yn p
EŒdjp
Xn .p ; yn / D ; for every n D 1; : : : ; N;
inf C ˛inf =inf
for all realizations p and yn of the equilibrium price and of the random signal,
respectively. The second order condition of the associated optimization problem
is given by 2inf C ˛inf =inf > 0. In particular, this implies that, in line with
economic intuition, informed speculators increase their demand when they receive
bullish signals and, on the contrary, reduce their demand when they receive bearish
information. Similarly, considering the optimization problem of an uninformed
speculator, it holds that
Q p
EŒdjp
Ym .p / D ; for every m D 1; : : : ; M;
un C ˛un =un
for every realization p of the equilibrium price, with the second order condition
2un C ˛un =un > 0. We have thus derived a set of necessary conditions for
a symmetric linear equilibrium to exist. The rest of the proof then proceeds by
verifying that this set of conditions is actually sufficient for the existence of
a symmetric linear equilibrium, making use of the result of Proposition 10.1.
We refer to Kyle [1148, Appendix B] for the detailed calculations and for the
explicit expressions of the coefficients appearing in (10.4) in correspondence of the
equilibrium. t
u
According to Proposition 10.2, a unique symmetric linear equilibrium exists
provided that there are enough speculators to generate a sufficiently competitive
trading environment. Moreover, as pointed out in Kyle [1148, Section 5], the
linearity of the demand schedules is part of the result, in the sense that for each
speculator, the equilibrium linear strategy dominates all non-linear alternatives.
Clearly, this result strongly depends on the assumption of exponential utility
functions together with a multivariate normal distribution.
The incidence of the private information on the equilibrium price can be
measured by the quantity
inf
D ˇ :
"
The quantity measures the variation in the equilibrium price pQ associated with a
unitary change in the valuation of the asset by the informed agent n as a result of a
higher realization of his private signal yQ n , for every n D 1; : : : ; N. In correspondence
of an imperfectly competitive equilibrium, Kyle [1148, Theorem 7.2] shows that
1
0 'inf and 'inf < 'un :
2
Under imperfect competition, the equilibrium price does never reveal more than
half of the private precision of the informed agents. This result contrasts with the
price revelation obtained in a perfectly competitive setting (see Chap. 8), where
the equilibrium price reveals almost all the private information when there is very
little noise trading. Moreover, when the variance of the aggregate demand of noise
traders vanishes (i.e., z2 ! 0), or when the risk aversion parameter ˛inf converges
to zero, or when the precision " increases to infinity, or, finally, when the number
of informed agents converges to infinity, then the quantity converges to the upper
bound 1=2, while 'inf remains strictly below 1=2, so that perfect revelation is never
reached. On the contrary, under the same conditions but in the presence of perfect
competition, the quantity 'inf converges to one, meaning that the equilibrium price
becomes fully informative. At first sight, the fact that the equilibrium price does not
become fully revealing as noise trading disappears from the economy looks like a
paradox. However, in the presence of imperfect competition, when the noise traders
trade a small amount, then the market becomes so illiquid that informed traders
are induced to trade a proportionally small amount as well and this prevents the
equilibrium price from fully revealing their private information.
The quantities 'inf and 'un are increasing with respect to the number M of
uninformed speculators and decreasing with respect to the risk aversion coefficient
˛un of the uninformed speculators (see Kyle [1148, Theorem 7.3]). The rationale for
this result is that an increase in M or a decrease in ˛un flattens the residual supply
curve faced by informed agents. As a consequence, informed agents will exhibit
a more aggressive demand schedule and, therefore, prices will be more affected
by their private information. If M ! 1, then EŒdjQ Q p D pQ , meaning that the
equilibrium price becomes unbiased even in the presence of imperfect competition
(see Kyle [1148, Theorem 7.4]). On the other hand, when M is finite, then the
equilibrium price is always a biased forecast of the dividend, even when uninformed
speculators are risk neutral. Moreover, as shown in Kyle [1148, Theorem 7.5],
the following conditions are equivalent (under the assumption that N 2 and
N C M 3):
(i) " ! 1, i.e., the precision of the informed speculators converges to infinity;
(ii) ˇ ! 1, i.e., informed trading is infinitely sensitive to private information;
(iii) informed speculators dominate trading, i.e., inf ! 1=N and un ! 0;
(iv) un ! 1, i.e., the equilibrium price is infinitely accurate;
(v) EŒ.dQ pQ /2 ! 0, i.e., the equilibrium price converges in mean square to the
dividend d. Q
In the presence of any of these conditions, speculative profits vanish. Furthermore,

both under perfect and imperfect competition, an increase in the number N of
informed speculators leads to an increase of the informational efficiency parameter
'inf (see Kyle [1148, Theorem 8.2]). Under some conditions (see Kyle [1148,
Theorem 9.2]), if N ! 1 then the imperfectly competitive equilibrium converges
to the perfectly competitive equilibrium as far as information transmission is
concerned. Note that, in the model, agents trade for both speculative and hedging
reasons.
In Kyle [1148, Section 10], the model is extended by considering the decision
to become informed as an endogenous decision, in the spirit of the model of
Grossman & Stiglitz [849] presented in Sect. 8.3. More specifically, Kyle [1148]
assumes that each informed speculator has ex-ante the possibility of paying a
positive cost in order to become informed (if he does not pay this cost, then he
remains uninformed). Costs are allowed to be different across speculators. In Kyle
[1148, Theorem 10.2] it is shown that, both in the perfectly competitive and in the
imperfectly competitive market, there exists a unique symmetric linear equilibrium
with endogenous acquisition of private information, under the assumption of free
entry of uninformed speculators. However, the equilibrium can be such that no
agent chooses to become informed. In particular, it is shown that, in the presence of
imperfect competition, the Grossman-Stiglitz paradox does not appear, since agents
keep prices inefficient enough to create profitable trading opportunities which are
sufficient to encourage traders to purchase costly private information.
The above discussion shows that the Grossman-Stiglitz paradox is strongly
dependent on the assumption of perfect competition. In Jackson [1002], in an
imperfectly competitive market (agents are not price takers) and assuming that
the asset dividend is distributed as an exponential random variable and that agents
are risk neutral, it is shown that there exist fully revealing equilibria with costly
information acquisition. Since the price is fully revealing, buying information does
not give an advantage to an agent and, actually, agents who acquire information
are worse off than those who do not acquire information. However, they would be
even worse off if they did not buy information, given that the other agents expect
them to do so. Agents are willing to invest in costly information even if they know
that the information will be completely revealed to the other agents by equilibrium
prices. Complementarities emerge in the acquisition of information because there is
a negative externality effect associated with the decision of not buying information.
This induces agents to invest in costly information.
If the adverse selection effect is severe, meaning that the informational disad-
vantage of uninformed agents is severe (i.e., the variance of the random variable
which is observed by informed agents exceeds a critical value) and the agent with
private information acts as a monopolist, then uninformed agents can refuse to trade
because the adverse selection effect outweighs the hedging motive to trade. In this
case, as shown in Bhattacharya & Spiegel [227], a market breakdown can occur.
10.2 Order-Driven Markets
In an order-driven market, there is a market maker who handles a book (specialist

book). In a batch auction market, orders arrive to the market maker, the market
maker looks at the order book (order imbalance) and then sets a price at which the
market is cleared, matching the aggregate demand. In this section, we first present
the model proposed in the seminal work Kyle [1147] and then we discuss several
extensions of this model.
10.2 Order-Driven Markets 595
The Kyle (1985) Model
In Kyle [1147], a simple two-period economy is considered, where a single risky

asset is traded. The liquidation value (dividend) of the asset is represented by the
random variable d,Q observed at date t D 1. The economy is populated by noise
traders, an insider trader and a market maker. Noise traders trade an aggregate
quantity represented by the random variable zQ, assumed to be independent of d. Q
Moreover, it is assumed that
dQ N .d;
N d2 / and zQ N .0; z2 /:
The trading mechanism is assumed to be decomposed into two successive

steps:
1. the insider trader learns the realization of the dividend, i.e., he observes the
realization of the random variable d.Q On the basis of this information, the insider
trader determines his optimal demand by choosing to trade a quantity xQ WD X.d/ Q
of the asset, for some measurable function X W R ! R. Note that the insider
trader cannot observe the realization of the random variable zQ;
2. the market maker determines the price pQ at which the market clears, observing
the aggregate market demand xQ C zQ. The market maker is assumed to be able to
only observe the aggregate demand and does not observe the two components xQ
and zQ separately. The price set by the market maker can therefore be represented
as pQ D P.Qx C zQ/, for some measurable function P W R ! R of the aggregate
market demand.
The insider trader is assumed to be risk neutral and determines his demand by
maximizing the conditional expectation of his profit given the observation of the
dividend. The profit of the insider trader is given by
Q WD .dQ pQ /Qx D .dQ pQ /X.d/:

Q
Recalling that the price pQ is set by the market maker according to the measurable
function P, let us write Q D .X;
Q P/ in order to emphasize the dependence of the
profit of the insider trader on the functions X and P introduced above. In a similar
way, let us also write pQ D pQ .X; P/ for the price set by the market maker. In this
context, an equilibrium of the economy is defined as a couple .X; P/ of measurable
functions such that
(i) the insider trader maximizes his expected profit, given the observation d of the
Q i.e., the insider trader demand function X satisfies
dividend d,
EŒ.X;
Q P/jd EŒQ .X 0 ; P/jd; (10.10)
for any alternative measurable function X 0 W R ! R and for every realization d;

(ii) the market is efficient, in the sense that the price pQ set by the market maker
satisfies
Q C z;
pQ .X; P/ D EŒdjx (10.11)
for every realization x C z of the random variable xQ C zQ.

The price setting rule (10.11) implies that the expected profit of the market maker
are equal to zero. This can be justified by considering the presence of several market
makers competing among themselves, thus making their profits converge to zero and
leading to the market efficiency condition (10.11).
The insider trader exploits his private information and takes into account the fact
that the quantity he chooses to trade will influence the market clearing price set by
the market maker. In other words, the demand function X also depends on the price
setting rule P. The insider trader takes the price setting rule used by the market
maker as given and is not allowed to condition the quantity he decides to trade on
the market price (i.e., the insider trader can only submit market orders and not limit
orders, see also below in this section).
In this setting, the equilibrium admits an explicit characterization, as shown in the
following proposition (compare with Kyle [1147, Theorem 1]), the proof of which
is given in Exercise 10.2.
Proposition 10.3 In the model considered above (see Kyle [1147]), there exists a
unique linear equilibrium .X; P/, where X W R ! R and P W R ! R are two
measurable functions such that
Q D ˇ.dQ d/
X.d/ N and P.Qx C zQ/ D dN C .Qx C zQ/; (10.12)
where the constants ˇ and are given by

s s
z2 1 d2
ˇD and D :
d2 2 z2
Proposition 10.3 yields several interesting insights. First, the fact that the optimal
demand of the insider trader is linear is not an assumption but rather a derived result,
in the sense that linear strategies are optimal even when non-linear strategies would
be a priori allowed. The equilibrium couple .X; P/ is completely determined by
the exogenous parameters d, N d2 and z2 . In particular, observe that the larger the
variance of the demand of the noise traders, the higher the coefficient ˇ representing
the sensitivity of the demand of the insider trader to his private information. The
rationale of this result is simple: if the demand of the noise traders exhibits a
large variance, then the insider trader can actively trade in the market knowing
that his private information will be more easily hidden to the eyes of the market
maker and, therefore, less reflected in the market price. The quantity 1= can be
regarded as a measure of market depth (i.e., the order flow necessary to induce the
equilibrium price to rise or fall by a unitary amount). This quantity is increasing
with respect to the variance of the demand of the noise traders and decreasing with
respect to the variance of the dividend, which represents the private information
of the insider trader. The sensitivity of the demand of the insider trader, measured
by the coefficient ˇ, is proportional to the market depth. Moreover, as shown
in Exercise 10.3, the expected profits of the insider trader (conditionally on the
observation of the private information dQ D d) are equal to .d d/ N 2 =.4/ and,
therefore, are proportional to the market depth. Ex-ante (i.e., before the observation
of the private information), the expected profits of the insider trader are equal to
d z =2.
In this model, the insider trader perfectly knows the dividend before trading,
while from the point of view of noise traders the dividend is a random variable
with variance equal to d2 . As shown in Exercise 10.3, in equilibrium it holds that
Q
Var.djp/ D d2 =2. Intuitively, this means that half of the insider trader’s private
information is incorporated in the equilibrium price. Therefore, the equilibrium is
not fully revealing. Note also that, unlike in the previous section, the conditional
variance of the dividend given the observation of the equilibrium price does not
depend on the variance of the noise traders’ demand. Since an increase in noise
trading brings forth more informed trading, noise trading does not destabilize prices.
This result is due to the fact that the insider trader is risk neutral: as the variance
of the noise traders’ demand increases, the insider will behave more aggressively,
without changing the informational content of the equilibrium price. On average (by
definition of market efficiency), the profits of the market maker are zero. The market
maker loses money when trading with the insider trader but gains an offsetting
amount of money when trading with noise traders. Therefore, the profits of the
insider trader are also the noise traders’ costs.
In Kyle [1147, Section 3], the above two-period model is extended to a
dynamic setting in discrete time, assuming that a series of trading rounds take
place sequentially, as will be discussed in more detail in Sect. 10.4. The model
is structured in such a way that, at each trading round, the equilibrium price reflects
the information contained in the present and past order flows and the insider, at
each trading round, maximizes his expected profits, taking into account his effect
on prices in the current trading round as well as in the future trading rounds.
The informed agent trades in such a way that his private information becomes
incorporated in the equilibrium price gradually. In the limit, as the time step between
two consecutive trading rounds goes to zero, a model of continuous trading emerges
and the price follows a Brownian motion.
Extensions
Risk Averse Informed Traders
In Subrahmanyam [1575], the model of Kyle [1147] has been extended by removing
the assumption that the insider trader is risk neutral and allowing for the presence
of N 1 risk averse insider traders. It is assumed that each insider trader is
characterized by a negative exponential utility function with coefficient of absolute
risk aversion a (the same for all insider traders). Similarly as in Kyle [1147], the
economy also comprises a certain number of noise traders, whose aggregate demand
is described by the random variable zQ, and a risk neutral market maker who absorbs
the net demand of the other traders and sets the market price in such a way that he
earns zero expected profits, conditionally on observing the aggregate demand. The
Q
liquidation value of the asset at date t D 1 is represented by the random variable d,
decomposed as
dQ D dN C "Q;
where dN is a constant known to every agent in the economy and "Q is a random
variable. Informed agents observe a noisy signal yQ given by a realization of the
random variable yQ D "Q C . Q The three random variables zQ, "Q and Q are mutually
independent and normally distributed, with zero mean and variances Var.Qz/ D z2 ,
Q D 2 and Var.Q"/ D 1 (without loss of generality).
Var./
Similarly as above, we denote by xQ D X.Q" C /
Q the demand of an informed agent,
for some measurable function X W R ! R representing the demand of an informed
agent as a function of his private signal, and we denote by pQ D P.N xQ C zQ/ the price
set by the market maker, for some measurable function P W R ! R representing
the price as a function of the total market demand. Similarly as in Kyle [1147],
the attention is focused on proving the existence and the uniqueness of a linear
equilibrium, represented by a couple .X; P/, where each of the N informed traders
maximizes his expected utility, conditionally on the observation of the private signal,
and the market efficiency condition (10.11) holds (i.e., the risk neutral market maker
obtains zero expected profits). This is the content of the following proposition, the
proof of which is given in Exercise 10.5.
Proposition 10.4 In the context of the above model (see Subrahmanyam [1575]),
there exists a unique linear equilibrium .X; P/, where X W R ! R and P W R ! R
are two measurable functions such that
X.Q" C /
Q D ˇ.Q" C /
Q and P.N xQ C zQ/ D dN C .N xQ C zQ/; (10.13)
where the constant ˇ is given by
1
ˇD (10.14)
.1 C 2 /.N C 1/ C a.2 C 2 z2 .1 C 2 //
and satisfies the quintic equation

1 N
N 2 .1 C 2 / C 2 z2 D :
ˇ ˇ
In Subrahmanyam [1575, Proposition 1] it is shown that the above quintic

equation for the coefficient admits a unique solution. In equilibrium, the value of
is decreasing with respect to the variance of the aggregate demand of noise traders.
Moreover, in equilibrium is either a unimodal function (i.e., first increasing
and then decreasing) of the number N of informed traders or is a monotonically
decreasing function of N. Similarly as in the model of Kyle [1147] presented above,
the quantity 1= represents a measure of market depth. Hence, these results imply
that market depth is increasing with respect to the variance of the noise traders’
demand and, in some cases, also with respect to the number of informed agents in
the economy. Similarly, with respect to the parameters a and 2 , the equilibrium
value of is either a unimodal function or a monotonically decreasing function. As
shown in Exercise 10.6, when the informed traders are risk neutral (similarly as in
the model of Kyle [1147]), the equilibrium value of is decreasing in the number of
informed traders and in the variance of the noise traders’ demand and is increasing
in the precision of information, as represented by the quantity 1=.1 C 2 /.
Unlike in the Kyle [1147] model, one of the main conclusions of Subrahmanyam
[1575] is that the equilibrium price reveals less information if the variance of the
noise traders’ demand is large (and this result is confirmed if the market maker is risk
averse, see Subrahmanyam [1575, Proposition 2]). Moreover, if informed traders are
risk averse, increasing the level of informational asymmetry between the informed
and the uninformed traders can yield an improvement of the terms of trade for the
uninformed agents. As shown in Subrahmanyam [1575, Proposition 2], if informed
traders are risk averse, price efficiency is decreasing with respect to the risk aversion
coefficient of informed traders. The rationale for this result is that more risk averse
informed agents trade less aggressively than less risk averse informed agents, with
the consequence that less private information will be reflected in the equilibrium
price.
Risk Averse Hedgers
The model of Kyle [1147] has been generalized in Spiegel & Subrahmanyam [1558]
considering an economy comprising three classes of agents: K risk neutral insider
traders, N risk averse uninformed traders and a competitive risk neutral market
maker. In particular, in this model the presence of noise traders is replaced by the
presence of risk averse uninformed expected utility maximizers and it is shown that
this leads to a significant modification of several results obtained in the original
model of Kyle [1147].
In the model considered in Spiegel & Subrahmanyam [1558], there is a single
risky asset, which is traded at the initial date t D 0 and liquidated at t D 1. The
Q
liquidation value (dividend) of the asset is represented by the random variable d:
dQ D dN C "Q;
where the constant term dN is known to every agent at date t D 0. Each informed
trader k 2 f1; : : : ; Kg observes a private signal yQ k of the form
yQ k D "Q C Q k ; for every k D 1; : : : ; K;
where .Q k /kD1;:::;K is a family of mutually independent random variables with normal
distribution with zero mean and identical variance 2 . The random variable "Q is
also normally distributed with zero mean and variance "2 and is assumed to be
independent of .Q k /kD1;:::;K . Every risk averse hedger n 2 f1; : : : ; Ng has a random
endowment eQ n of the asset and is characterized by a negative exponential utility
function with risk aversion parameter a. The family .Qen /nD1;:::;N is assumed to be
composed of i.i.d. random variables with normal distribution with zero mean and
identical variance e2 . Furthermore, it is assumed that the family .Qen /nD1;:::;N is
independent of .Q"; Q 1 ; : : : ; Q K /.
The price formation mechanism is similar to that considered in the models
presented above in this section. Namely, informed and uninformed traders submit
their demands to the risk neutral market maker, who then sets a price which equals
the expected liquidation value of the security, given the observation of the aggregate
demand. The analysis is focused on proving the existence and the uniqueness of
a linear equilibrium of the economy. In this context, a linear equilibrium can be
represented by a triplet .X; W; P/ of measurable functions, where X represents
the demand of each informed trader, given as a function of the private signal,
W represents the demand of each uninformed trader, given as a function of his
endowment, and P represents the price set by the market maker, given as a function
of the total market flow. The following proposition, whose proof is given in
Exercise 10.7, provides an explicit characterization of the linear equilibrium of the
economy (see Spiegel & Subrahmanyam [1558, Proposition 1 and Appendix A]).
Proposition 10.5 In the context of the above model (see Spiegel & Subrahmanyam
[1558]), if
a2 Ne2 ."2 C 22 /2 > 4K."2 C 2 /; (10.15)
then there exists a unique linear equilibrium .X; W; P/, where X W R ! R, W W

R ! R and P W R ! R are measurable functions explicitly given by
X.Q" C Q k / D ˇ.Q" C Q k /; W.Qen / D eQ n ; (10.16)

for all k D 1; : : : ; K and n D 1; : : : ; N, and

! !
X
K X
N X
K X
N
P X.Q" C Q k / C W.Qen / D dN C X.Q" C Q k / C W.Qen / ;
kD1 nD1 kD1 nD1
(10.17)
where the constants ˇ, and are explicitly given by

q
a"2 K."2 C 2 / ."2 C 22 /2 C K"2 2 C .K.N 1/=N/"2 ."2 C 2 /
D 2 p q ;
.1 C K/"2 C 22 a Ne2 ."2 C 22 / 2 K."2 C 2 /
p q
.1 C K/"2 C 22 a Ne2 ."2 C 22 / 2 K."2 C 2 /
ˇD q ;
a K."2 C 2 / ."2 C 22 /2 C K"2 2 C .K.N 1/=N/"2 ."2 C 2 /
p q
2 2 a N 2 . 2 C 2 2 / 2 K. 2 C 2 /
.1 C K/" C 2 e " "
D p :
2
Ne a ."2 C 22 /2 C K"2 2 C .K.N 1/=N/"2 ."2 C 2 /
If condition (10.15) is not satisfied, then a linear equilibrium does not exist.
Condition (10.15) means that a linear equilibrium exists if and only if the risk
aversion of the uninformed traders, or the variance of their endowment, or their
number, is sufficiently high and the number of informed traders is sufficiently
low. When condition (10.15) fails to hold, a market breakdown occurs, since the
demand of the informed traders overwhelms that of uninformed traders, with the
consequence that the market maker incurs in systematic losses, thus preventing the
existence of an equilibrium. Note that this is in contrast with the model of Kyle
[1147] presented above, where a linear equilibrium always exists.
It can be shown that the equilibrium value of given in Proposition 10.5 is
decreasing with respect to a and e2 , meaning that market depth is increasing in the
coefficient of absolute risk aversion of uninformed agents and in the variance of their
endowments (see Spiegel & Subrahmanyam [1558, Proposition 2]). The behavior
of with respect to N, K, "2 and 2 is not necessarily monotone and depends on
the parameters of the model. Note that this implies the counterintuitive result that
having more uninformed traders is not necessarily beneficial from the point of view
of market liquidity. However, the market becomes infinitely liquid (i.e., ! 0) if
2 ! 1 or N ! 1 and infinitely illiquid (i.e., ! 1) if "2 ! 1 (see Spiegel &
Subrahmanyam [1558, Corollary to Proposition 1]). Moreover, there exists a finite
value K such that as K converges to K the market becomes infinitely illiquid and,
for K > K , a linear equilibrium does not exist. The interpretation of this last result
is that, if there are too many insider traders, then the adverse selection effect is too
severe and a (linear) equilibrium cannot exist.
By relying on the distributional assumptions of the model and on the properties

of the bivariate normal distribution, it can be easily shown that in correspondence
of the linear equilibrium it holds that
ˇ K !
ˇX XN
"2 .Kˇ 2 2 C N 2 e2 /
Var dQ ˇˇ X.Q" C Q k / C W.Qen / D
kD1 nD1
Kˇ 2 .K"2 C 2 / C N 2 e2
and, by replacing the equilibrium value of the coefficient ˇ as given in Proposi-

tion 10.5, we get
ˇX !
ˇ K X
N
"2 ."2 C 22 /
Q ˇ
Var d ˇ X.Q" C Q k / C W.Qen / D ;
kD1 nD1
.1 C K/"2 C 22
see Spiegel & Subrahmanyam [1558, Section 1.2]. In particular, this shows that in
equilibrium the posterior variance of the liquidation value of the asset (and, hence,
the informational efficiency of the equilibrium price) does not depend on any of the
parameters a, e2 and N. This result depends on the assumption of risk neutrality
of the informed traders. The expected profits of informed traders are increasing
with respect to a and e2 , while the welfare of uninformed agents is decreasing with
respect to K and "2 (see Spiegel & Subrahmanyam [1558, Propositions 4 and 5]).
Informed Traders Submitting Limit Orders
In the model of Kyle [1147], it is assumed that insider traders can only submit
market orders, i.e., orders which do not depend on the price. In Vives [1628], the
author develops a model with heterogeneous risk averse informed agents who are
also allowed to submit orders depending on the market price (limit orders).
The model developed in Vives [1628] is similar to the noisy rational expectations
equilibrium models considered in Admati [15], Diamond & Verrecchia [569], Hell-
wig [931], Grossman & Stiglitz [849] (see Sect. 8.3). There is a single risky asset,
Q and a risk free asset with unitary return. The economy is
with liquidation value d,
populated by three types of agents:
• a continuum of informed risk averse agents, indexed in the interval Œ0; 1,
characterized by a negative exponential utility function with common risk
aversion parameter a;
• noise traders, who trade an aggregate quantity zQ of the risky asset;
• a risk neutral competitive marker maker, similarly as in the models considered
above in the present section.
Each informed agent i observes a private signal yQ i of the form
yQ i D dQ C "Qi ; for every i 2 Œ0; 1:

The initial wealth of informed agents is normalized to zero. Similarly as in the model
of Kyle [1148], each informed agent i 2 Œ0; 1 submits to the market a demand
schedule Xi .; yQ i /, taking into account the equilibrium functional relationship of
prices with respect to the random variables describing the economy. We can think
of the demand schedule Xi .; yQ i / as a form of generalized limit order contingent on
private information. Similarly as in Sect. 10.1, if the market clearing price of the
risky asset is p, then trader i demands Xi .p; yi / units of the risky asset, conditionally
on the observation of a realization yi of his private signal yQ i .
The market maker is assumed to be competitive and risk neutral. Hence,
observing the aggregate demand schedule
Z 1
L./ WD Xi .; yQ i /di C zQ;
0
the market maker sets an efficient price pQ for the risky asset characterized by the
Q
condition pQ D EŒdjL./. It is assumed that all the random variables appearing in the
model are normally distributed:
dQ N .d;
N d2 /; "Qi N .0; "2 / and zQ N .0; z2 /:
Moreover, the random variables dQ and zQ are assumed to be independent and the
family of random variables .Q"i /i2Œ0;1 is assumed to be composed of independent
random variables, independent of dQ and zQ. We define as follows the precision of the
random variables appearing in the model:
d WD 1=d2 ; " WD 1="2 and z WD 1=z2 :
It is furthermore assumed that

Z 1
"Qi di D 0; almost surely:
0
All the distributional assumptions are common knowledge among all the agents in
the economy.
Similarly as in the models considered above, in Vives [1628] the attention is
focused on proving the existence and the uniqueness of a linear equilibrium of the
economy. This is the content of the following proposition, corresponding to Vives
[1628, Proposition 1.1] and proved in Exercise 10.8.
Proposition 10.6 In the context of the above model (see Vives [1628]), there exists
a unique symmetric linear equilibrium, where the demand schedules Xi .Qp; yQ i / of the
informed agents are given by
"
Xi .Qp; yQ i / D .Qyi pQ /; for every i 2 Œ0; 1; (10.18)
a
and the equilibrium price is given by

" N
"Q
pQ D 1 dC d C zQ ; (10.19)
a a
where
" z =a
D :
d C "2 z =a2
In particular, the above proposition shows that, in equilibrium, informed traders

trade more intensively if the precision of their private signal is higher or their risk
aversion is lower. Moreover, the trading intensity is independent of the amount
of noise trading. The depth of the market, as measured by the quantity 1=, is
increasing in noise trading and monotonic (increasing or decreasing) in risk aversion
and in the precision of the information of informed traders. Expected trading volume
is increasing in noise trading (see Vives [1628, Corollary 2.1]).
Monopolistic Market Maker
In the model of Kyle [1147] presented at the beginning of this section, the market
maker sets the equilibrium price in such a way that his expected profits are null. This
price setting rule can be motivated by the presence of perfect competition among
several market makers. However, in some markets, the market maker operates as a
monopolist and, therefore, he does not set the market price according to the zero
expected profits condition. On the contrary, a monopolist market maker maximizes
his expected profits.
In Glosten [787] it is shown that, in a model with asymmetric information among
traders, the presence of a monopolistic market maker allows to reach a better risk
sharing in comparison with a competitive market maker when the adverse selection
effect is significant. Recall that, in the presence of traders with heterogeneous and
private information, the market maker faces an adverse selection problem, due to
the presence of informed agents in the market. The presence of competition among
market makers entails two different welfare effects. On the one hand, competition
should offer better prices to liquidity traders. On the other hand, competition can
prevent the market makers from effectively screening the different types of traders,
thereby inducing an adverse selection effect which negatively affects the welfare.
If the adverse selection effect is extreme, then a market breakdown can occur (i.e.,
an equilibrium fails to exist), since the competitive market maker expects to lose
money. On the other hand, in such a situation, the presence of a monopolistic market
maker does not necessarily lead to a market breakdown, since the monopolist can
decide to keep the market open and learn some of the information of the informed
agents by trading, thus reducing the adverse selection problem and making his
subsequent trades more profitable, offsetting the previous losses. In other words,
the monopolist has the possibility of averaging his profits across different trades,
thus implying a better liquidity even in the presence of extensive trading based on
private information. When the adverse selection effect is significant, the welfare of
informed and liquidity traders is higher in a monopolistic market maker economy
than in a perfectly competitive market makers economy.
The model of Glosten [787] can be described as follows. In the market, there are
two traded securities: a risk free security with constant rate of return (normalized
to one for simplicity) and a risky security with liquidation value d, Q supposed to
N
be normally distributed with mean d and variance 1=d . There is an informed trader
characterized by a negative exponential utility function with risk aversion parameter
a and endowed at date t D 0 with w0 units of the risk free asset and with wQ
units of the risky asset. It is assumed that wQ is a random variable, independent of
Q with a normal distribution with zero mean and variance 1=w . The informed
d,
trader observes a private signal yQ of the form yQ D dQ C "Q, where "Q is a normally
distributed random variable, independent of dQ and w, Q with zero mean and variance
1=" . The market maker is assumed to be risk neutral and cannot observe neither the
realization of the private signal yQ of the informed trader nor the realization of the
random endowment w. Q
Conditionally on the observation of a realization y of the private signal yQ and
on a realization w of the random endowment w, Q the informed trader defines his
demand schedule X WD XP .y; w/ by maximizing the expected utility of his wealth
at date t D 1, given the price schedule P W x 7! P.x/ set by the market maker
(with the interpretation that a trade of x will lead to a transfer of money of xP.x/
from the trader to the market maker). The notation XP .y; w/ emphasizes that the
demand schedule of the informed trader depends on the pricing rule P as well as
on the signal and on the endowment. Analogously to Kyle [1147], a competitive
market maker sets the market price according to the zero expected profits rule, i.e.,
he sets a market price defined by the condition Pc .X/ D EŒdjX. Q On the contrary,
a monopolistic market maker defines the price schedule Pm .X/ by maximizing his
expected profits, given the demand schedule X of the informed trader.
In Glosten [787, Proposition 1], it is shown that an equilibrium with a competitive
market maker exists if and only if
a 2 d 1
WD > :
a2 d C .
" w d C " / 2
If this condition is not satisfied, then an equilibrium cannot exist and the market
breaks down. The quantity can be regarded as a measure of adverse selection. If
there is no private information (i.e., " D 0), then D 1, while if " is large enough
(meaning that the adverse selection effect is significant) then the above inequality
is not satisfied and a competitive equilibrium does not exist. As shown in Glosten
[787, Section 2], there exists a value O such that a monopolistic market maker is
preferred by traders when < O (strong adverse selection), while the competitive
market makers setting is preferred when > O (weak adverse selection).
In a dynamic setting, Leach & Madhavan [1170] show that a monopolistic market
maker can also help the diffusion of information by experimenting with prices, i.e.,
by setting prices different from the conditional expectation of the liquidation value
in order to learn the agents’ information by observing the resulting market order
flow. This activity is costly, but a monopolistic market maker can recover the costs in
an intertemporal setting. However, this is not possible in the presence of competing
market makers.
Markets with a Limit Order Book
In several financial markets, a limit order book has been adopted in order to increase
the liquidity of the market. As mentioned above, one can distinguish between two
main types of orders: market orders (orders executed immediately at the best price
available in the market) and limit orders (orders which will be executed only at some
specified price). A market order typically leads to an immediate execution of the
order, but there is no certainty about the price at which the order will be executed.
For this reason, market orders are well suited for liquidity trades. On the contrary,
limit orders are characterized by an uncertain execution time (and, in some cases,
may also remain unexecuted) but will be executed at a known price and, hence,
are well suited for patient traders. With market orders, the cost of the immediate
execution of the order is represented by the fact that the price is determined by the
market, while with limit orders the cost of setting the execution price of the order is
represented by the fact that the execution of the order can be delayed. Limit orders
provide liquidity to future trades but are exposed to adverse informational price
changes, in the sense that a limit order might turn out to be mispriced ex-post. The
limit order book registers all the limit orders previously posted in the market (i.e.,
bids and offers specifying quantities and prices).
An analysis of a limit order book market system has been proposed in Glosten
[788], assuming that (risk neutral) liquidity suppliers behave competitively (i.e.,
they earn zero expected profits), facing informed traders in the market. The limit
order book is modeled as a publicly visible screen providing bids and offers together
with the corresponding prices and quantities. Transactions against the book pick off
the limit orders at their limit prices. The source of bids and offers is a large pool of
risk neutral liquidity suppliers. It is shown that the system works well in the presence
of strong adverse selection, providing as much liquidity as possible compared to
other market settings, and it allows to average profits and gains across trades as in
Glosten [787]. Limit order traders gain from small trades and liquidity-driven price
changes but lose from information-driven price changes. There is a positive small-
trade bid-ask spread and the limit order book is immune to competition from third
market dealers.
Biais et al. [240] provide an analysis with strategic liquidity suppliers (imperfect
competition) who earn positive expected profits. The trading volume is shown to
be lower than in the case of perfect competition. Under some conditions, in the
10.3 Quote-Driven Markets 607
presence of imperfect competition and no asymmetric information, risk averse

agents prefer a hybrid market (dealer market together with a limit order book) to
a dealer market or to a limit order book (see Viswanathan & Wang [1627]): small
trades are directed towards the limit order book, while large orders are handled by
dealers. A limit order book is preferred by risk neutral traders and the dealer market
is preferred when there are many market makers. Biais et al. [234] show that a limit
order market provides better risk sharing and narrower spreads than a dealer market
(see also Parlour & Seppi [1403] on the efficiency of a limit order book setting).
10.3 Quote-Driven Markets
In many financial markets, there exist market dealers who quote bid and ask prices
for a set of assets. A bid price corresponds to the price that the dealer accepts to pay
in order to execute a sell order in the market, while an ask price corresponds to the
price that the dealer accepts to receive in order to execute a buy order in the market.
The dealer is committed to satisfy the orders by applying the quoted prices, under
some conditions. In many cases, the commitment of the dealer only holds for orders
smaller than a certain amount and the dealer is allowed to revise the bid and ask
prices in the case of large orders. Typically, in a dealer market orders can arrive at
any point in time (hence, a dealer market usually operates in continuous time). The
main features of a dealer market are immediacy and quick price discovery.
By quoting bid-ask prices and accepting to execute orders, the dealer provides
a service to the market. This service exposes the dealer to at least two different
types of costs: first, the dealer is forced to accept a non-optimal position in the
traded assets since he has to keep an inventory of the assets; second, the dealer
faces an adverse selection problem due to possibility of trading with better informed
investors. Because of the presence of these costs, the dealer asks for a remuneration
for his service and this remuneration is reflected in the existence of a bid-ask spread.
The literature on dealer markets and, in particular, on the determination of the
bid-ask spread can be divided into two main groups of contributions, depending
on the relative relevance of the two types of costs described above (see O’Hara
[1384] and De Jong & Rindi [540] for a detailed account on these topics). A first
series of contributions concentrates on the first cost, emphasizing the fact that the
dealer is forced to hold a non-optimal portfolio and assuming that the dealer is risk
averse. The dealer manages an inventory of the asset, quotes bid and ask prices
and faces a stochastic order flow (often described by a Poisson process). The dealer
considers the risk of the inventory position with respect to some target portfolio,
revising the quoted bid and ask prices in order to manage the risk. Optimal inventory
management and bid-ask strategies have been determined in Amihud & Mendelson
[58], Ho & Stoll [949], O’Hara & Oldfield [1385], Biais [231], showing that the
bid-ask spread is increasing in the dealer’s risk aversion (but is independent of the
inventory).
Let us consider the portfolio management problem of a dealer in the context of a

simple two-period model first proposed by Stoll [1570], following the presentation
in O’Hara [1384, Section 2.2] and De Jong & Rindi [540, Section 5.1]. In this
model, the dealer is risk averse and demands a compensation when acting as a
liquidity provider in the market. In Stoll [1570], a dealer simply represents a market
participant who is willing to alter his own portfolio away from optimality in order
to match the demands of other agents.
Consider an economy with N risky assets, where each asset n 2 f1; : : : ; Ng is
traded at some price pn at the initial date t D 0 and delivers the random dividend dQ n
at the following date t D 1, where dQ n N .dN n ; n2 /. Consider a dealer endowed with
initial wealth w0 and characterized by a negative exponential utility function with
risk aversion parameter a. At the initial date t D 0, the dealer chooses a portfolio
of the N risky assets by maximizing the expected utility function of the random
wealth realized at date t D 1. We denote by D .1 ; : : : ; N /> 2 RN the optimal
portfolio of the dealer, with n representing the number of units of the n-th asset
held in the portfolio, for n D 1; : : : ; N, and we denote by W e the corresponding
optimal wealth realized at date t D 1 when no trade occurs in the market between
date t D 0 and date t D 1, i.e.,
X
N X
N
e D
W n .dQ n pn / D n dQ n w0 :
nD1 nD1
Recall that, as explained in Sect. 3.2, under the assumptions of an exponential utility
function and of a normal distribution, the optimal portfolio belongs to the mean-
variance efficient portfolio frontier.
Let us suppose that, at t D 0 and after the dealer has formed his optimal portfolio,
a trade of xn units of the n-th asset arrives to the dealer, for some n 2 f1; : : : ; Ng.
The dealer has to execute this order and, of course, after the execution of the order,
the initial optimal portfolio of the dealer will no longer be optimal. To compensate
for this risk, the dealer sets the bid and the ask prices for the n-th asset in such a
way that the modified portfolio, after the execution of the order, has the same mean-
variance profile of his original optimal portfolio . Let us denote by paskn and pn
bid
the ask and bid prices, respectively, set by the dealer for the n-th asset at date t D 0.
The dealer’s wealth at date t D 1, after the execution of the order xn , is given by
X
N
e0 D
W k .dQ k pk / xn .dQ n pN n / D W
e xn .dQ n pN n /;
kD1
where
pN n D pask
n 1fxn >0g C pn 1fxn <0g ;
bid
meaning that pN n is the ask price if xn is a buy order and the bid price if xn is a sell
order. As a compensation for the commitment of executing the order, the dealer
has the possibility of setting the bid and ask prices. In the model of Stoll [1570],
the dealer chooses pask bid
n and pn in such a way that he is indifferent (in the mean-
variance sense) between the original optimal portfolio and the modified portfolio
after the execution of the order, i.e.,
e a e 0 a Var.W
e / D EŒW e 0 /:
EŒW Var.W (10.20)
2 2
In the present setting, the mean-variance problem (10.20) can be explicitly solved
and the optimal bid and ask prices can be determined, as shown in the following
proposition, whose proof is given in Exercise 10.9.
Proposition 10.7 In the context of the above model (see Stoll [1570]), for each
n D 1; : : : ; N, the optimal bid price pbid ask
n and the optimal ask price pn , determined
as the solution to the mean-variance problem (10.20), are given by
N a 2 e Q
n D dn n jxn j a Cov.W ; dn /;
pbid
2
N a 2 e Q
n D dn C n jxn j a Cov.W ; dn /;
pask
2
where We denotes the original optimal wealth of the dealer, and the associated
bid-ask spread is given by
2
n pn D an jxn j:
pask bid
According to the above proposition, the bid-ask spread set by the dealer is
increasing in the dealer’s risk aversion, in the variance of the dividend of the traded
asset as well as in the size of the order to be executed.
The adverse selection problem faced by the dealer in the presence of informed
traders has been analysed in several papers and, in particular, we refer to the
contributions of Bagehot [105], Copeland & Galai [498], Glosten & Milgrom
[790]. In Copeland & Galai [498], a two-period dealer economy with liquidity and
informed traders has been analysed. In the market, there exists a risky asset, whose
current price at date t D 0 is given by p0 and whose underlying value is given by pQ ,
where pQ is a random variable with density function f ./. The price p0 represents the
“true” price of the asset as perceived by the dealer at date t D 0. It is assumed that in
the economy there are no taxes nor restrictions on short selling and that the density
function f ./ is common knowledge among market participants. The dealer makes
a commitment to buy a fixed quantity of the asset at the bid price pbid and to sell
a fixed quantity of the asset at the ask price pask . In the market, there are informed
traders and liquidity traders. Informed traders know ex-ante the fundamental value pQ
and trade on the basis of this information, while liquidity traders trade for exogenous
reasons and do not know the fundamental value. Both types of traders are assumed
to be risk neutral and decide whether to post a buy order, a sell order, or do nothing,
depending on the bid and ask prices quoted by the dealer. The dealer does not know
the fundamental value and also cannot distinguish between informed and liquidity
traders. The dealer is risk neutral and, hence, is only interested in maximizing his
expected profits (similarly as in the models presented before in this chapter, this can
be justified by the presence of competition among several dealers). In the model, it
is assumed that the probability that a trade is motivated by superior information
is exogenously given and equal to inf > 0, with liq WD 1 inf being the
probability that a trade is due to liquidity trading. The probability inf is known
to the dealer. Furthermore, it is assumed that a liquidity trader submits a buy order
with probability B;liq and a sell order with probability S;liq .
The dealer’s problem consists in setting the bid and the ask prices of the asset. If
the dealer sets a very large bid-ask spread, then he can reduce the potential losses
due to trading with better informed traders, but he risks to lose the profits generated
by trading with liquidity traders. On the other hand, if the dealer sets a very tight
bid-ask spread, then he can profit from trading with liquidity traders, but he is more
exposed to the risk of trading with informed traders. The optimal bid-ask spread
must be chosen as a compromise between these two situations. Due to the structure
of the model, the dealer’s expected profit from trading with liquidity traders is given
B;liq .pask p0 / C S;liq .p0 pbid /:
On the other hand, the expected loss of the dealer when trading with the better
informed traders are given by
Z 1 Z pbid
.p p /f .p/dp C
ask
.pbid p/f .p/dp:
pask 0
The integrals with respect to the density function f ./ represent the fact that, unlike
an informed trader, the dealer does not know the true underlying value of the asset.
The (risk neutral) dealer sets the bid and ask prices pbid and pask by maximizing the
following functional:

liq B;liq .pask p0 / C S;liq .p0 pbid /
Z 1 Z pbid !
inf .p p /f .p/dp C
ask
.p p/f .p/dp :
bid
pask 0
In the model, there exists a bid-ask spread as long as there is a positive probability
that some of the trades are motivated by superior information, even if the dealer is
risk neutral. In other words, risk aversion is not necessary in order to explain the
emergence of a bid-ask spread. The bid-ask spread decreases when we shift from
a monopolistic market dealer to an economy where several dealers compete among
themselves (see also below in this section for a discussion of this aspect). Moreover,
if the proportion of informed traders increases, the difference between the bid-ask
spreads in the case of monopoly and in the case of competition will decrease. If
the elasticity of demand for liquidity trading decreases then the ask price will also
decrease (see Copeland & Galai [498]).
In the presence of superiorly informed agents, a dealer faces an adverse selection
problem, as shown in Copeland & Galai [498]. However, this adverse selection
problem also has an intertemporal dimension which is not reflected in the above
model. The intertemporal dimension is related to the fact that, by trading with better
informed agents over time, the dealer can eventually learn their private information.
A dynamic version of the above model which allows to take into account this aspect
has been developed in the seminal paper Glosten & Milgrom [790]. In this work,
the authors consider a discrete time economy with a finite number of trading dates
and a pure dealership market, meaning that all orders are market orders (limit orders
are not allowed). In the economy, there is a single risky asset and market orders are
constrained to be for one unit of the asset (so that the dealer is not allowed to quote
different bid/ask prices for different quantities). The market mechanism considered
in Glosten & Milgrom [790] can be described by the following iterative sequence
of steps:
1. the dealer quotes ask and bid prices for buy and sell orders of one unit of the
asset;
2. a trader arrives at the dealer and, on the basis of the quoted bid/ask prices and his
information, decides whether to submit a market order to the dealer;
3. if the trader decides to post a market order, then the order is executed;
4. after the execution of the order and before the arrival of the next trader, the dealer
can revise his quotes for the bid and ask prices.
According to this mechanism, after executing a buy or a sell order, the dealer updates
his beliefs and is allowed to revise his quotes for the bid and ask prices of the asset.
The risky asset has a fundamental value (per share) represented by the non-
negative random variable vQ with finite variance. In the economy, there are informed
traders (insider traders) and pure liquidity traders. It is assumed that informed
traders know ex-ante the realization of the random variable v, Q while this is unknown
to liquidity traders as well as to the dealer. It is also assumed that there exists a finite
time T0 at which the fundamental value vQ is revealed to every market participant.
In this sense, T0 can represent a future date at which the information becomes
homogeneous and the informational advantage of informed traders disappears.
Every market participant (informed traders, liquidity traders and the dealer) is
risk neutral. In the market, exactly one investor arrives at the dealer at each date
t 2 f1; : : : ; T0 g and the dealer is not able to distinguish between informed investors
and liquidity traders (i.e., the market is anonymous). Since only one trader arrives at
the dealer at each date, we can identify investors by their time of arrival. Different
traders are characterized by different time preferences and these are identified by
a parameter ıt , which describes the preferences of the investor arriving at date t
between current consumption and future consumption derived from the ownership
of the asset. The utility of agent t can described by
ıt x vQ C c;
where x denotes the number of shares of the asset traded by investor t and c
denotes current consumption (for the dealer, it is conventionally assumed that
ı D 1). According to this preference structure, a high value of ıt represents
a desire to invest for the future, while a low value of ıt indicates a preference
for immediate consumption. Since the market is anonymous, the dealer cannot
discriminate between different investors. Accordingly, from the perspective of the
dealer, the collection .ıt /tD1;:::;T0 can be viewed as a stochastic process and it is
assumed to be independent of the fundamental value v. Q Note that, since every market
participant is risk neutral, it is necessary that different agents exhibit different values
of the time preference parameter ı, otherwise the no-trade result of Milgrom &
Stokey [1341] (see Sect. 8.1) would imply that the bid-ask spread is set large enough
to preclude any trade.
In this model, it is important to distinguish the different information flows
corresponding to the different types of traders. Similarly as in Chap. 6, we denote
by F D .Ft /tD1;:::;T0 the information flow of liquidity traders, consisting of all the
information generated by the observation of past transaction prices, the current bid
and ask prices as well as any publicly available information. We then denote by
G D .Gt /tD1;:::;T0 the information flow of informed traders, which consists of all
the information available to liquidity traders together with the knowledge of the
fundamental value v of the asset (the realization of the random variable v). Q In the
model, it is assumed that the knowledge of the process .ıt /tD1;:::;T0 does not bring
any useful information on the other random variables. In particular, it is assumed
that
EŒvjG
Q t _ .ıt / D EŒvjG
Q t and EŒvjF
Q t _ .ıt / D EŒvjF
Q t ; (10.21)
for all t D 1; : : : ; T0 , with .ıt / denoting the -algebra generated by the random
variable ıt . This means that the knowledge of the time preference parameter ıt does
not permit to make better forecasts about the fundamental value of the asset.
For each t D 1; : : : ; T0 , let pask t and pbid
t denote the ask and bid prices,
respectively, quoted by the dealer at date t. At each date t, an investor arrives at
the dealer, observes the ask and bid prices quoted by the dealer and decides whether
to trade and, in the case he decides to trade, whether to submit a buy or a sell order.
The investor t 2 f1; : : : ; T0 g decides according to the following rule:
buy if Zt > pask

t ;
(10.22)
sell if Zt < pbid
t ;
where the process .Zt /tD1;:::;T0 is defined by
Zt WD ıt .1 Ut /EŒvjG
Q t C ıt Ut EŒvjF
Q t ; (10.23)
where Ut is an indicator variable which is equal to one or zero depending on whether

the investor arriving at date t is a liquidity trader or an informed trader, respectively.
Arguing as in Glosten & Milgrom [790], by including the specification of who is

informed in the sample space, we can generically represent the information flow of
a trader by K D .Kt /tD1;:::;T0 . In this way, we can rewrite (10.23) as
Zt D ıt EŒvjK
Q t ; for every t D 1; : : : ; T0 :
Given the structure of the investors’ preferences, the dealer determines the bid
and the ask prices by adopting a zero expected profits condition. We denote by
D D .Dt /tD1;:::;T0 the information flow of the dealer. Given the investors’ behavior
described in (10.22)–(10.23) and the information Dt available to the dealer at date
t, the dealer’s expected profit at date t is given by
h ˇ i
E .pask
t v/1
Q fZt >pask C . v
Q /1 bid ˇDt ;
bid
t g p t fZt <pt g
which can be equivalently rewritten as

t E vjD
pask Q t ; Zt > pask
t P.Zt > pask
t jDt /
bid (10.24)
pt E vjD Q t ; Zt < pbid
t P.Zt < pbid
t jDt /:
The zero expected profits condition implies that (10.24) has to be equal to zero at all
dates t D 1; : : : ; T0 . Hence, the zero expected profits equilibrium (if it exists) at each
date t 2 f1; : : : ; T0 g can be described by a couple of random variables .pask t ; pt /
bid
satisfying

pask
t D E vjD
Q t ; Zt > pask
t and t D E vjD
pbid Q t ; Zt < pbid
t : (10.25)
In general, showing the existence of two processes .pask t /tD1;:::;T0 and .pt /tD1;:::;T0
bid
satisfying condition (10.25) at all t D 1; : : : ; T0 is not easy. In the special case where
the dealer’s information flow D contains the information flow G of the informed
traders (i.e., if Gt Dt , for all t D 1; : : : ; T0 ), then it holds that paskt D pbid
t D
EŒvjD
Q t , for all t D 1; : : : ; T0 . On the other hand, if the dealer’s information flow D
coincides with the information flow F of the liquidity trader (i.e., Dt D Ft , for all
t D 1; : : : ; T0 ), then it holds that
˚ ˚
pask
t D inf a W EŒvjF
Q t ; Zt > a a and pbid
t D sup b W EŒvjF
Q t ; Zt b ;
for all t D 1; : : : ; T0 . As pointed out in Glosten & Milgrom [790], according to

this notion of equilibrium, the dealer does not regret ex-post any trade that he is
committed to execute. The ask price corresponds to the revised expectation of the
fundamental value of the asset if the dealer is executing a buy order, while the bid
price corresponds to the revised expectation of the fundamental value of the asset if
the dealer is executing a sell order. In other words, equilibrium prices are set equal to
the dealer’s conditional expectations of the fundamental value of the asset given the
type of order submitted. In this sense, the type of order is interpreted by the dealer
as an informative signal and the bid and ask prices set by the dealer will incorporate
the information that each trade has revealed.
A first and fundamental property to be checked is that the bid price is less or equal
than the ask price and, moreover, that the conditional expectation of the fundamental
value of the asset (on the basis of all publicly available information) lies between
the bid price and the ask price. This is confirmed by the next proposition, which
corresponds to Glosten & Milgrom [790, Proposition 1] and the proof of which is
given in Exercise 10.10.
Proposition 10.8 In the context of the above model (see Glosten & Milgrom [790]),
suppose that there exist two processes .pask
t /tD1;:::;T0 and .pt /tD1;:::;T0 satisfying the
bid
equilibrium condition (10.25). Then it holds that
t EŒvjD
pbid Q t ^ Ft pask
t ; for all t D 1; : : : ; T0 ;
where Dt ^ Ft denotes the smallest -algebra generated by the sets belonging to

both Dt and Ft .
At each date t 2 f1; : : : ; T0 g, if a trade is being executed and it is a buy order,
then the transaction price at t will correspond to the ask price pask t . On the contrary,
if at date t a trade is executed and it is a sell order, then the transaction price at date t
will correspond to the bid price pbid t . Hence, in view of rule (10.22), if a transaction
happens at t, then the corresponding transaction price pt can be written as
pt D pask
t 1fZt >pask
t g
C pbid
t 1fZt <pbid
t g
:
In equilibrium, the ask price and the bid price are determined according to (10.25),
so that the last relation can be rewritten as
pt D EŒvjD
Q t ; Zt > pask
t 1fZt >pask
t g
C EŒvjD
Q t ; Zt < pbid
t 1fZt <pbid
t g
D EŒvjD
Q t ;
(10.26)
conditionally on the fact that a trade takes place at date t (a buy or a sell order).
Suppose that N trades take place between the initial date t D 0 and date T0 and
denote by .Tn /nD1;:::;N f1; : : : ; T0 g the collection of dates at which a transaction
has taken place. Then, relation (10.26) can be equivalently rewritten as
pTn D EŒvjD
Q Tn ; for all n D 1; : : : ; N; (10.27)
where the process .pTn /nD1;:::;N collects all the prices at which transactions have
occurred. As shown in the following simple proposition (corresponding to Glosten
& Milgrom [790, Proposition 2]), transaction prices form a martingale with respect
to the information flow of the dealer.
Proposition 10.9 In the context of the above model (see Glosten & Milgrom [790]),
suppose that there exist two processes .pask
t /tD1;:::;T0 and .pt /tD1;:::;T0 satisfying
bid
the equilibrium condition (10.25). Suppose that N trades take place at dates
.Tn /nD1;:::;N . Then the transaction price process .pTn /nD1;:::;N is a martingale with
respect to the information flow .DTn /nD1;:::;N , i.e., it holds that
pTn D EŒpTnC1 jDTn ; for all n D 1; : : : ; N 1:
Proof The claim simply follows from property (10.27) together with the tower
property of the conditional expectation:

pTn D EŒvjD Q TnC1 jDTn D EŒpTnC1 jDTn ;
Q Tn D E EŒvjD
for every n D 1; : : : ; N 1. t
u
A first implication of the above proposition is that transaction prices are
informationally efficient in the semi-strong sense (see Chap. 8), in the sense that
they are efficient with respect to the information flow of the dealer, which is assumed
to contain all publicly available information. The fact that the dealer is risk neutral
(or, analogously, that there is perfect competition among several dealers), implies
that there are no profit opportunities arising from the information known to the
dealer, due to the martingale property of transaction prices. A second implication
of the above proposition is that the first differences of transaction prices are serially
uncorrelated, as a direct consequence of the fact that martingale increments are
uncorrelated.
After having presented the general framework of the model of Glosten & Mil-
grom [790], let us illustrate some of its main implications in a very simple setting,
following O’Hara [1384, Section 3.3] and De Jong & Rindi [540, Section 4.1]. We
suppose that the random variable vQ representing the fundamental value of the asset
can take only two possible values v and v, with v > v, with probabilities and
1, respectively. As above, the probability is known to every market participant,
but only the insider traders observe the realization of v.
Q The dealer faces an informed
trader with probability ˛, while with probability 1˛ he faces a liquidity trader. The
liquidity trader randomly decides to submit buy or sell orders, i.e., the probability
that a liquidity trader decides to buy one unit of the asset is given by 1=2 and the
probability that he decides to sell one unit of the asset is also given by 1=2. On the
other hand, an informed agent trades by maximizing his expected profit, so that he
will submit a (one unit) buy order if he knows that the fundamental value is equal to
v and, on the contrary, a (one unit) sell order if the fundamental value is v.
Similarly as above, the dealer sets the bid and the ask prices for the asset,
according to the zero expected profits rule. More specifically, the bid and the ask
prices are set according to the rule (10.25), i.e., bid and ask prices represent the
expectations of the fundamental value of the asset (from the point of view of the
dealer), conditionally on the information revealed by the trade itself. Hence:
pask D EŒvjbuy
Q order and pbid D EŒvjsell
Q order;
where EŒvjbuy
Q order and EŒvjsell
Q order denote the conditional expectations of
the fundamental value of the asset, given the information that a buy or a sell
order has been submitted, respectively. In the present simple setting, as shown in
Exercise 10.11, the equilibrium bid and ask prices can be easily computed in closed
form, by making use of Bayes’ rule:
.1 C ˛/ .1 /.1 ˛/
pask D v Cv ;
1 ˛ C 2 ˛ 1 ˛ C 2 ˛
(10.28)
.1 ˛/ .1 C ˛/
pbid D v Cv :
1 C ˛ 2 ˛ 1 C ˛ 2 ˛
Moreover, if we furthermore assume that D 1=2, we then get
1C˛ 1˛ vCv ˛

pask D v Cv D C .v v/;
2 2 2 2
1˛ 1C˛ vCv ˛
pbid Dv Cv D .v v/
2 2 2 2
and the corresponding bid-ask spread is given by
pask pbid D ˛.v v/:
In this simple setting, we can observe that the bid-ask spread is increasing in
the proportion of informed traders in the economy and it is due to the presence of
adverse selection costs. Moreover, we can regard the difference vv as a measure of
the volatility of the fundamental value of the asset. In this sense, the above relation
shows that the bid-ask spread is increasing in volatility. The existence of a positive
bid-ask spread amounts to a compensation of the dealer for the risk of trading with
better informed investors. Of course, even in this simple setting, the transaction
prices exhibit the martingale property established in Proposition 10.9, thus implying
that transaction prices are semi-strong efficient from an informational point of view
(with respect to the information flow of the dealer).
The model of Glosten & Milgrom [790] yields several additional interesting
insights. For instance, markets exhibiting a large trading volume will have smaller
spreads and vice versa. Glosten & Milgrom [790, Proposition 4] show that, under
suitable technical conditions, the dealer’s conditional expectation of the fundamen-
tal value of the asset converges to the insider trader’s conditional expectation of the
fundamental value as the number of trading dates increases. Intuitively, this means
that the dealer learns over time by updating his beliefs on the basis of the information
revealed by the trades themselves. Glosten & Milgrom [790, Proposition 5] shows
that, at any given date t 2 f1; : : : ; T0 g, the bid-ask spread increases when, all other
things being equal, the insider traders’ information becomes more precise, the ratio
of informed to uninformed traders arrival rates at t increases and the elasticity of
uninformed supply and demand at date t increases. This is due to the fact that the
adverse selection problem is worse the greater the fraction of informed traders and
the better their information. These results have been confirmed in Madhavan [1272]:
if the adverse selection problem is severe, then the dealer may set prices in such a
way that no trade occurs (market breakdown).
The model of Glosten & Milgrom [790] considers standardized trades where the
traded quantity is fixed to one unit of the asset. By making the trade size endogenous,
several interesting implications can be obtained. A model in the spirit of Glosten &
Milgrom [790] but with possibly different trade sizes has been developed in Easley
& O’Hara [619], motivated by the fact that large trades are usually carried out at less
favorable prices than small trades. As a consequence, the size of an order also reveals
some information to the dealer. If investors are allowed to choose the size of their
trades, then the size of the trade can affect the bid-ask spread, since better informed
agents prefer to trade larger quantities in comparison to uninformed agents. In
Easley & O’Hara [619], it is shown that two types of equilibria may arise, depending
on the values of the parameters of the model (see Exercise 10.12): either informed
traders submit both small and large orders, so that they will not be separated from
liquidity traders (pooling equilibrium) or informed traders only submit large orders,
so that they can be separated from liquidity traders (separating equilibrium). In
particular, in correspondence of a separating equilibrium, the adverse selection
problem does not appear in correspondence of small market orders, since the latter
are only submitted by liquidity traders. Moreover, the ask price associated to a
large order in a separating equilibrium is greater than the ask price associated to
a large order in a pooling equilibrium. Again, the rationale of this phenomenon is
the adverse selection effect. More details on this model are given in Exercise 10.12.
Not surprisingly, competition among market dealers affects the bid-ask spread.
Inventory-based models predict that the bid-ask spread is a decreasing function of
the number of dealers in the market (see Amihud & Mendelson [58], Grossman
& Miller [846], Ho & Stoll [950], Biais [231]). For instance, in the context of
a model similar to that proposed by Stoll [1570], the bid-ask spread is inversely
proportional to the number of market dealers, as shown in Exercise 10.13. Moreover,
considering a financial market where multiple assets are traded, the presence of
substitutability among the traded securities in case of a monopolistic dealer leads
to a decrease of the bid-ask spread, as considered in Hagerty [873]. In the presence
of asymmetric information, the effects of the competition among dealers on the
bid-ask spread is more ambiguous. In Dennert [559], assuming risk neutral insider
investors, it is shown that an increase in the number of risk neutral market makers
leads to a higher risk exposure for each individual market maker and, therefore, to
larger bid-ask spreads and transaction costs. The rationale for this effect is that the
total amount of uninformed liquidity trading is limited and remains constant and,
therefore, the adverse selection problem for each individual dealer becomes more
pronounced as the number of dealers increases. Moreover, in a competitive dealers
market, the informational advantage of each dealer is smaller than in a monopolistic
dealer market (this effect is similar to that described in Glosten [787]). Under some
conditions, liquidity traders prefer a monopolistic market maker. A similar result
is obtained in an order-driven market by allowing agents to submit orders to more
than one risk neutral market maker (see Bernhardt & Hughson [206]). In Madhavan
[1272], tight bid-ask spreads are obtained in a setting with adverse selection and
competition among several dealers.
In Dutta & Madhavan [604], a game theoretic model has been developed in
order to analyse the dynamic pricing strategies of competing market makers in a
dealer market. In particular, the authors aims at studying whether competing market
makers can sustain bid-ask spreads that are above the competitive levels, even if they
do not explicitly cooperate to fix prices. It is shown that, under some conditions,
the market makers can act in a non-cooperative way in setting prices and, at the
same time, they may still set bid-ask spreads that are higher than the corresponding
bid-ask spreads obtained in a competitive environment. This is a form of implicit
collusion, which results from the dealers adopting Nash strategies. In some cases,
the dealers’ strategies under implicit collusion coincide with the strategies obtained
under explicit collusion. However, implicit collusion is not feasible in every market.
Indeed, if the dealers are sufficiently impatient or if there are no entry barriers, then
the only possible equilibrium of the economy is the competitive equilibrium (in this
regard, see also Grossman et al. [847]). The implicit costs represented by the costs
of establishing a reputation and forming relationships to gain access to the order
flow (order preferencing arrangements) are sufficient to reduce price competition
and to ensure implicit collusion. When dealers receive preferenced order flow, in
equilibrium the bid-ask spread increases. Furthermore, collusion is more frequent
in an actively traded market and when dealers are of a similar size.
Dealers provide immediacy to the market. Assuming that traders experience
liquidity needs, Grossman & Miller [846] develop an equilibrium model for the
number of dealers in the market. The number of market makers will adjust until in
equilibrium the dealers’ costs of maintaining a continuous presence are equal to the
expected returns of trading. As the number of market makers increases, the market
liquidity increases (see also Biais [231] for a related equilibrium analysis of the
number of dealers in a market). Wahal [1640] shows that the number of dealers is
increasing in the trading intensity.
According to the above analysis, three different types of costs are at the origin
of the bid-ask spread: order processing costs, inventory costs and adverse selection
costs. A cross-sectional analysis shows that volume, risk, price and firm size explain
most of the variability of the bid-ask spread or of the price impact of a trade. While
the relevance of order processing costs is widely recognized, the relevance of the
other two cost components is more debated. The relevance of adverse selection and
inventory effects has been pointed out in George et al. [768], Glosten & Harris
[789], Hasbrouck [911], Huang & Stoll [976, 978], Easley et al. [618], Lin et al.
[1219], Hasbrouck & Sofianos [913], Madhavan & Smidt [1279, 1280], Hendershott
& Seasholes [934], showing that the adverse selection effect increases with trade
size (see Bloomfield [249] for some experimental evidence). Madhavan & Sofianos
[1281] and Manaster & Mann [1292] suggest that dealers manage their inventory
by timing the size and the direction of their trades rather than by adjusting their
quotes. Note also that quote adjustments due to adverse selection and inventory
management go in the same direction (e.g., a buy order leads to an increase in the
10.4 Multi-Period Models of Market Microstructure 619
price), but quote adjustments due to inventory management tend to be reversed over
time (since the dealer wants to reconstruct the optimal inventory position), while
quote adjustments for adverse information are not. Therefore, inventory effects
induce negative serial correlation in orders, price changes and returns at high
frequencies (on the other hand, under asymmetric information the quote revision
is persistent). Confirming the results of Glosten & Milgrom [790] and Easley &
O’Hara [619], large trades have a large price impact (i.e., the price is shown to be
an increasing and concave function of the trade size) and widen the spread (see
Hasbrouck [911, 912]). The price impact is increasing in the degree of adverse
selection (see Koski & Michaely [1122]).
The positive relation detected empirically between bid-ask spreads and expected
returns (see Amihud & Mendelson [59], Chalmers & Kadlec [393], Datar et al.
[525], Brennan & Subrahmanyam [297], Eleswarapu [635], Amihud [57], Easley
et al. [615], Brennan et al. [294], Bekaert et al. [185]) can be explained in terms
of the adverse selection effect, leading to a higher probability of trades based on
superior information (and, therefore, higher expected returns) and widening the
bid-ask spread. A theoretical model accounting for this phenomenon has been
developed in Easley & O’Hara [620], while evidence in favor of the Kyle [1147]
microfoundation has been documented in Chordia et al. [433]. Easley et al. [618]
show that the probability of informed trading is lower for high volume stocks.
Moreover, there is a positive premium for informational trading. A decrease of the
bid-ask spread and of its adverse selection component is observed in Greene &
Smart [823] when the noise traders’ component is more significant in the market.
Although a secular relation between the reduction of transaction costs and trading
activity has been identified (see Chordia et al. [435]), no significant relation has
been observed between volume/trading activity and the bid-ask spread of stocks
(see Jones [1039] and Johnson [1037]).
10.4 Multi-Period Models of Market Microstructure
The seminal two-period model developed in Kyle [1147, Section 2] and discussed
at the beginning of Sect. 10.2 has been extended to a dynamic setting in Kyle [1147,
Section 3]. The dynamic version of the model preserves the fundamental features
already discussed in Sect. 10.2 and, in particular, at each date the equilibrium price
reflects the information contained in the past and present order flow and the insider
trader maximizes his expected profits, taking into account the effect of his actions
on the prices both at the current date as well as at the future dates.
More specifically, Kyle [1147] considers a multi-period economy where N
auctions take place sequentially over time, at dates t1 < : : : < tN , where tN D 1 and
with tn denoting the date at which the n-th auction takes place. Let tn WD tn tn1 ,
for all n D 1; : : : ; N, with t0 WD 0, denote the time step between two successive
auction dates. As in the two-period version of the model presented in Sect. 10.2,
we assume that a single risky asset is traded, with liquidation value d, Q where
dQ N .d; N d2 /. Let the process .Qzt /t2Œ0;1 denote a scaled Brownian motion, so
that the increment Qzn WD zQtn zQtn1 is distributed as a normal random variable
with mean zero and variance z2 tn , for every n D 1; : : : ; N, for some scaling
parameter z2 > 0. For every n D 1; : : : ; N, the random variable Qzn represents
the quantity of the asset traded by noise (liquidity) traders at the n-th auction. Note
that the independence of Brownian increments implies that the quantities traded
by noise traders in correspondence of different auctions are independent. It is also
Q Similarly
assumed that the process .Qzt /t2Œ0;1 is independent of the random variable d.
as in the version of the model considered in Sect. 10.2, the insider trader knows the
realization of the liquidation value dQ of the asset. We denote by xQ n the aggregate
position of the insider trader after the n-th auction, for all n D 1; : : : ; N, so that
Qxn WD xQ n xQ n1 represents the quantity traded by the insider at the n-th auction.
For each n D 1; : : : ; N, we denote by pQ n the market clearing price at the n-th auction
viewed as a random variable and by pn its generic realization.
In correspondence of each auction, the trading mechanism follows the two steps
described at the beginning of Sect. 10.2, suitably modified in order to include the
relevant information generated by the previous auctions. Namely, in correspondence
of each auction date tn , n D 1; : : : ; N:
1. the insider, knowing the realization of the liquidation value of the asset and the
transaction prices p1 ; : : : ; pn1 realized at the previous auctions, determines his
optimal position xQ n D Xn .p1 ; : : : ; pn1 ; d/,Q where Xn W Rn ! R is a measurable
function, for each n D 1; : : : ; N;
2. the market maker determines the price pQ n at which the market clears, observing
the current order flow Qxn C Qzn at date tn and observing also the past values of
the order flow. Accordingly, the market clearing price is determined by a rule of
the form pQ n D Pn .Qx1 C zQ1 ; : : : ; xQ n C zQn /, where Pn W Rn ! R is a measurable
function, for each n D 1; : : : ; N.
Note that the information flow of the insider trader includes the quantities he has
traded in the past. Similarly, the information flow of the market maker includes
the prices he has set in the past. For brevity of notation, let us define the family
of functions X WD .X1 ; : : : ; XN / and P WD .P1 ; : : : ; PN /. Following Kyle [1147],
the family X can be referred to as the insider trader’s trading strategy and P as the
market maker’s pricing rule.
Similarly as in the static version of the model presented in Sect. 10.2, it is
assumed that the market maker is risk neutral and sets the market price according
to the zero expected profits condition. The insider trader is also risk neutral and, at
each auction date tn , determines his demand of the asset by maximizing his expected
profit. At each date tn the profits obtained by the insider trader from the auctions
taking places at the dates ftn ; : : : ; tN g are given by
X
N X
N
Q n WD .dQ pQ k /Qxk D .dQ pQ k /Xk .p1 ; : : : ; pk1 ; d/;
Q
kDn kDn
with p1 ; : : : ; pk1 denoting the realizations of pQ 1 ; : : : ; pQ k1 . Recalling that the price
process .Qpn /nD1;:::;N is determined dynamically by the market maker according to
the pricing rule P, we write Q n D Q n .X; P/ in order to emphasize the dependence of
the profits of the insider trader on the family of functions X and P. Analogously,
following the notation of Kyle [1147], we also write pQ n D pQ n .X; P/ and xQ n D
xQ n .X; P/.
In this context, similarly as in Sect. 10.2, an equilibrium of the economy
(sequential auction equilibrium) is defined as a couple .X; P/ of families of
measurable functions such that, at every date tn , n D 1; : : : ; N:
(i) the insider trader maximizes his expected profits, given the observation of the
realization d of the liquidation value dQ and the realized past transaction prices,
i.e., the insider trader demand function X satisfies
EŒQ n .X; P/jp1; : : : ; pn1 ; d EŒQ n .X 0 ; P/jp1 ; : : : ; pn1 ; d; (10.29)
for any alternative family X 0 D .X10 ; : : : ; XN0 / of measurable functions such that
Xk0 D Xk , for all k D 1; : : : ; n1, and for all possible realizations p1 ; : : : ; pn1 ; d
Q
of the random variables pQ 1 ; : : : ; pQ n1 ; d;
(ii) the market is efficient, in the sense that the price pQ n set by the market maker
satisfies
Q 1 C z1 ; : : : ; xn C zn ;
pQ n D pQ n .X; P/ D EŒdjx (10.30)
for all realizations x1 C z1 ; : : : ; xn C zn of xQ 1 C zQ1 ; : : : ; xQ n C zQn .

Apart from the different information structures appearing in the above definition of
equilibrium, the interpretation is completely analogous to the notion of equilibrium
discussed in Sect. 10.2. In the present dynamic setting, it is important to remark that
the market efficiency condition (10.30) implies that the price process .Qpn /nD1;:::;N
is a martingale with respect to the information flow generated by the total order
flow of the market. This property immediately follows from the tower property of
the conditional expectation together with the fact that, for each n D 1; : : : ; N, the
price pQ n is given in (10.30) by the conditional expectation of the terminal random
variable d.Q
In the present dynamic setting, an equilibrium .X; P/ is said to be a recursive
linear equilibrium if each component of the family of functions .X; P/ is linear and
if there exist constants 1 ; : : : ; N such that
pQ n D pQ n1 C n . Qxn C Qzn /; for every n D 1; : : : ; N:

Similarly to Proposition 10.3, it can be shown that there exists a unique recursive
linear equilibrium, which furthermore admits an explicit characterization (see Kyle
[1147, Theorem 2]). In particular, for every n D 1; : : : ; N, it holds that
Qxn D ˇn .dQ pQ n1 / tn and Qpn D n . Qxn C Qzn /;
where the constants ˇ1 ; : : : ; ˇN and 1 ; : : : ; N are given as the unique solution to a

system of difference equations. The parameters ˇ1 ; : : : ; ˇN measure the sensitivity
of the demand of the insider trader with respect to his private information, while
the parameters 1 ; : : : ; N measure the depth of the market, with a small value of
corresponding to a deep market. In particular, note the similarity with the structure
of the linear equilibrium described in Proposition 10.3. In equilibrium, it holds that
n D ˇn ˙n =z2 , where ˙n is the variance of the liquidation value of the asset
conditionally on the observation of the market flow up to date tn . The quantity ˙n
measures how much of the insider trader’s information is not yet incorporated into
the market price (as estimated by the market maker). Referring to the original paper
Kyle [1147] for full details, we only mention that the proof of this result proceeds
by backward induction.
Many of the properties of the equilibrium discussed in Sect. 10.2 continue
to hold in the present dynamic setting. The parameter ˙n , which measures the
informativeness of the equilibrium price, is decreasing with respect to n, reflecting
the fact that, as time goes on, more of the private information of the insider trader
is reflected in the market price. Note, however, that ˙N > 0, meaning that the
insider trader’s information is only partially incorporated into the equilibrium price
at the terminal date. The parameters n , n D 1; : : : ; N, are inversely proportional
to the parameter z2 , representing the variance of the noise traders’ demand. This
means that increasing the amount of liquidity trading increases the depth of the
market, increases proportionately the profits of the insider trader and leaves the
informativeness of the equilibrium price unchanged, similarly as in the simple two-
period model discussed in Sect. 10.2.
Kyle [1147, Section 5] analyses the limiting behavior of the equilibrium when
the time step between two successive auction dates goes to zero. As tn converges
to zero, the stochastic processes of the price and of the order flow converge to a
system of stochastic differential equations driven by Brownian motion. In the limit,
the parameter does not depend on time and equals d =z , meaning that, in the
continuous time limit, the market depth is constant in equilibrium (see Kyle [1147,
Theorem 3]). On the other hand, the conditional variance of the liquidation value
of the asset decreases linearly with time and, moreover, by the end of trading, the
insider trader’s private information will be perfectly reflected into the equilibrium
price. The equilibrium price process is a martingale with constant instantaneous
variance.
The properties of the equilibrium characterized in Kyle [1147] reflect the fact that
an insider trader acts in a way that does not lead to an immediate revelation of his
private information, but rather acts in a strategic way and his private information is
gradually incorporated into the equilibrium price. Such a result provides a potential
explanation for the fact that market prices typically need some time in order to
incorporate new information arriving into the market (underreaction). Considering
endogenous liquidity traders in a market with long-lived asymmetric information,
the dissemination of information turns out to be slower, as shown in Mendelson &
Tunca [1324]. Moreover, as considered in Chau & Vayanos [414], if new private
information arrives at each trading date, then the insider trader acts aggressively
and his private information is quickly revealed to the market. Multiple insider
traders have been considered in Holden & Subrahmanyan [953], assuming that each
individual insider knows the liquidation value of the asset before starting to trade.
In this setting, the presence of competition among insider traders induces them to
trade aggressively and leads to a rapid incorporation of their private information
into the equilibrium prices. Moreover, the trade of an insider in the presence of
competition is larger than that of a monopolistic insider, the market quickly becomes
very deep and the conditional variance of the liquidation value of the asset quickly
converges to zero. Increasing the number of informed traders, the intensity of these
effects is even more pronounced. In the limit case where the time step between two
successive trading dates converges to zero, the price becomes immediately fully
revealing and the market very deep. These results also hold in markets with risk
averse insiders (see Holden & Subrahmanyan [954]). However, competition among
insiders can be much less intense if agents are endowed with heterogeneous (and
correlated) private information, as considered in Foster & Viswanathan [732]. When
the correlation among heterogeneous signals observed by different insiders is low,
then less information is revealed in the presence of multiple insiders than in the case
of a unique insider (see Back et al. [101]).
It is important to remark that, in the model of Kyle [1147], insider traders
never manipulate the market. Indeed, the trading strategies of the insiders are linear
and insiders always trade following the direction of their private information. The
model has been extended in Hong & Rady [959] by considering the case where
the market maker is better informed than the insider trader on liquidity trades, in
the sense that the market maker knows the variance of liquidity trades while the
insider trader knows it imperfectly. In this context, insider traders optimally take into
account the uncertainty on the variance and gradually learn about market liquidity
from past prices and trading volume. One of the main implications of this model
is that past prices and trading volume affect future trades via a learning effect.
Wang [1642] allows for heterogeneous prior beliefs (i.e., agents have different
distributional assumptions concerning an informed trader’s private signal) and
asymmetric information in the above model. In equilibrium, it turns out that volume
and absolute price changes-volatility are positively correlated and trading volume is
positively serially correlated.
In the above setting, no public disclosure of insider trades has been considered. In
many financial markets public disclosure of insider trades is mandatory, in the sense
that insiders associated with a firm (i.e., officers, directors and major shareholders)
must report to some market authority the trades they make in the stock of the
firm (see also Sect. 10.5). The rationale for this rule is that public disclosure of
insider trades should be beneficial to uninformed agents and, therefore, should
facilitate price discovery. Huddart et al. [989] confirm this effect in the context of
the Kyle [1147] dynamic model discussed above, with a single insider trader and
a market maker who observes insider trades at the end of each round of trading.
In comparison with the original model of Kyle [1147], in this setting the insider
trades by adding a random noise component to his demand, except in the last trading
round. Intuitively, the insider trader tries to dissimulate his private information in
order to limit its dissemination in the market. Public disclosure of insider trades
accelerates the price discovery process and lowers the trading costs with respect
to the non-disclosure case and, moreover, the insider trader’s expected profits are
lower in comparison with the classical setting of Kyle [1147]. These effects become
stronger as the number of trading rounds increases. However, these results are not
necessarily confirmed in the presence of several insider traders. In Huddart et al.
[988], it is shown that public disclosure of insider trades may inhibit the price
discovery process by dampening competition among insiders. Assuming repeated
(short-lived) private information arrivals, public disclosure of insider trades induces
a tacit coordination by allowing insiders to monitor each others’ trades. Insider
traders coordinate among themselves and trade less aggressively in order to exploit
their private information. In the case of delayed disclosure, Cheng et al. [427] have
considered the possibility of price manipulation.
Under the assumption of a perfectly competitive market, we have seen that
informed traders trade more aggressively than uninformed agents and trading
volume is increasing in the precision of the private signals. This result is only
partially confirmed when agents trade strategically. Indeed, in Kyle [1147], Admati
& Pfleiderer [20], Foster & Viswanathan [729], it is shown that informed agents
act strategically, revealing to the market their private information in a gradual way
and trying to dissimulate their trades in order to better exploit their informational
advantage. However, the empirical literature does not provide a clear conclusion
on this aspect. On the one hand, Hasbrouck [912], Easley et al. [616], Seppi
[1518] show that trade size has an effect on price adjustments (in the sense that
large trades are more informative), while in Barclay & Warner [161], Chan &
Fong [404], Chakravarty [390] it is shown that medium size trades are the more
informative (stealth trading) and in Jones et al. [1041], Easley et al. [617], Chan
& Fong [404] it is shown that transaction numbers rather than volume determine
volatility. In a related direction, Meulbroek [1333] and Cornell & Sirri [499] show
that insider traders use medium size trades in order to better conceal their infor-
mation. Block trades can also be motivated by liquidity (uninformative) reasons:
taking this observation into account, Seppi [1518] shows that the informational
content is increasing in trade size. Whether an insider trader acts competitively
or strategically has been investigated empirically. According to the Kyle [1147]
model, in a strategic perspective the trade size should be inversely related to market
illiquidity (as measured by the parameter introduced above) and positively related
to trading volume. These results have been confirmed empirically in Brennan &
Subrahmanyam [298].
Intraday and Intraweek Regularities
Market microstructure models have also been considered as a possible explanation

of some well-known intraday regularities appearing in asset prices time series.
Indeed, in some financial markets, it has been observed that volume, volatility and
bid-ask spreads follow a U-shaped pattern during a trading day. They are at the
highest point at the opening of the trading day, decrease rapidly to lower levels
during the day and then rise again towards the end of the day. These empirical
regularities have been reported for instance in Goodhart & O’Hara [812], Foster &
Viswanathan [731], Brock & Kleidon [307], Chan et al. [403], Werner & Kleidon
[1652], Chordia et al. [434] and are difficult to explain in the context of classical
asset pricing theory. While this U-shaped pattern is by itself a puzzle, one of the
main issues is represented by the fact that trading volume is high when trading costs
are large. From a theoretical point of view, it is difficult to explain such a relation,
while, on the other hand, the relation between price volatility and spread can be
supported by theoretical models based on information asymmetry.
A model allowing to explain the intraday patterns described above has been
proposed in Admati & Pfleiderer [20]. The main feature of the model is that noise
traders can decide the timing of their trades during a time span (day, week, month).
In Admati & Pfleiderer [20], the authors consider a multi-period economy, with
trading dates t D 1; : : : ; T, where a single risky asset is traded. The value of the
asset at the terminal date T is exogenously given and equal to
X
T
dQ D dN C "Qt ;
tD1
where the family .Q"t /tD1;:::;T is composed of independent random variables with
zero mean. While at the terminal trading date T every market participant has full
information on the liquidation value dQ of the asset, at previous dates the information
on dQ is partially revealed through public and private sources. More specifically, it is
assumed that, at each date t D 1; : : : ; T, the random variable "Qt becomes publicly
known, but there also exist informed agents who observe a signal yQ t of the form
yQ t D "QtC1 C Q t ; for each t D 1; : : : ; T 1;

2;
with Var.t / D t . This represents the fact that informed agents receive a signal
about the information that will become publicly known at the successive trading date
(or, according to an alternative interpretation, informed agents are able to process
the public information in a faster or more efficient way). Note that in this model
the private information becomes useless at the subsequent trading date, since it
will become part of the publicly available information. In this sense, the private
information is short-lived and informed traders, therefore, have no incentive to
restrict their trading activity in order to have a larger informational advantage in
the future, unlike in most of the models mentioned above.
In the economy, all the traders are risk neutral, similarly as in Kyle [1147]. At
each date t D 1; : : : ; T, it is assumed that there are Nt informed traders (in Admati
& Pfleiderer [20] the model is also extended by making the decision of becoming
informed endogenous). Differently from the models considered above, the economy
comprises two types of liquidity traders: in each period, there are non-discretionary
liquidity traders, who must trade a given number of shares in that period, and
discretionary liquidity traders, who have liquidity demands that do not need to
be satisfied immediately but only before some future date. At each trading date
t D 1; : : : ; T, non-discretionary liquidity traders have a total demand represented by
the random variable zQt , while discretionary liquidity traders determine their demand
by minimizing their expected trading costs. It is assumed that, in each period, there
exists a constant number M of discretionary liquidity traders.
Similarly as in Kyle [1147], at each trading date the market price is set by a
risk neutral market maker who sets the market price according to the zero expected
profits condition (forced by competition). The market price set by the market maker
is given by the expectation of the liquidation value of the asset, conditionally on
the publicly available information and the observation of the market flow up to the
current date. Similarly as in the models considered above, under the assumption that
all the random variables appearing in the model are normally distributed, it can be
shown that the equilibrium of the economy is characterized by a linear pricing rule
of the form
X
t
pt D dN C "Qt C t !Q t ; for every t D 1; : : : ; T;
sD1
where !Q t denotes the aggregate market demand (order flow) at date t (see Admati &
Pfleiderer [20, Section 1.2]). Moreover, in equilibrium the demand xQ t of an informed
trader at date t is given by
xQ t D ˇt yQ t ; for every t D 1; : : : ; T;
where
s s
t Var.Q"tC1 / Nt
ˇt D 2;
and t D 2;
;
Nt .Var.Q"tC1 / C t / Nt C 1 t .Var.Q"tC1 / C t /
where t is the variance of the aggregate liquidity trades at date t (i.e., the variance
of the total demand by non-discretionary liquidity traders and discretionary liquidity
traders), see Admati & Pfleiderer [20, Lemma 1].
The coefficient t is decreasing with respect to Nt and t . Similarly as in the
model of Kyle [1147], the reciprocal of the parameter t measures the market
depth at date t. This implies that, as the number of informed agents increases,
the competition among them becomes more intense, since all the informed agents
are assumed to observe the same private signal, and this leads to an increase of
the depth of the market. A similar effect is observed as the (endogenous, due to
the presence of discretionary liquidity traders) variance of noise traders’ aggregate
demand increases. As far as the insider trader’s trading intensity is concerned (as
measured by the coefficient ˇt ), it increases with respect to the liquidity traders’
aggregate trading volume and to the precision of the private signal, while it is
decreasing with respect to the number of insider traders.
In equilibrium, each discretionary liquidity trader follows a trading policy aiming
at minimizing the expected trading costs, subject to the constraint of meeting the
liquidity need at some future date. It can be shown that this induces discretionary
liquidity traders to trade in the period where the market is more deep (i.e., when
the parameter t takes the smallest value). Therefore, there always exist equilibria
where discretionary liquidity trading is concentrated in the same period (see Admati
& Pfleiderer [20, Proposition 1]). As shown above, the parameter t is small when
the variance t is large. In turn, the large variance of liquidity trading allows insider
traders to more easily conceal their trades in the aggregate liquidity trades and
this phenomenon is reflected in the expression for the coefficient ˇt given above.
In other words, the informed traders trade more aggressively in the period when
all the discretionary liquidity traders trade together. Similarly as in Kyle [1147],
the variance of price changes and the amount of information incorporated into the
equilibrium price do not depend on the variance of the aggregate liquidity trades
(the rationale for this effect is the same as in the Kyle [1147] model), as long as the
number of informed traders is kept fixed and the precision of information is constant
over time.
Admati & Pfleiderer [20] also consider an extended version of their model
where the decision to become informed is endogenous, so that the number of
informed traders becomes an endogenous variable and is determined as part of the
equilibrium. In this case, the concentration of trades becomes even more intense:
the fact that at some date there is more liquidity trading attracts more informed
trading, which can be more easily concealed, and this makes even more attractive
for liquidity traders to trade in the same period. In the presence of endogenous
information acquisition, the clustering of trades increases the informativeness of the
equilibrium price and the variance of price changes over time. Traders collect more
information when there is more liquidity trading. The model of Admati & Pfleiderer
[20] can explain the relation between large trading volume and large price changes
observed empirically. These results are in part confirmed when information is
heterogeneous. When the private information is long-lived, liquidity trades (volume)
and volatility of prices are still related (see Back & Pedersen [102]).
The model of Admati & Pfleiderer [20] presented above does not explain the
fact that large trading volume is often associated with large bid-ask spreads. Indeed,
one would expect the opposite relation to hold, with a large trading volume being
associated to a tight bid-ask spread. Information-based models suggest that bid-ask
spread and volatility decline monotonically over the day, as a consequence of the
presence of a learning process reducing the information asymmetry, with the largest
volatility being at the opening of the day. This intraday volatility pattern has been
confirmed for many financial markets (see Gerety & Mulherin [772] and Stoll &
Whaley [1571]): the variance of open-to-open returns is larger than the variance
of close-to-close returns. The result can be attributed to the procedure for opening
stocks or to private information released by prices. Moreover, prices tend to reverse
around the opening of the trading day. Madhavan et al. [1278] argue that, while
information costs decrease over the day (as a consequence of the fact that agents
learn from the trading process itself), inventory costs increase and generate the U-
shaped pattern of the bid-ask spread.
The intraday pattern of trading volume and bid-ask spreads has been explained
by introducing market closure inside the dynamic optimal investment-consumption
problem without assuming private-asymmetric information (see Brock & Kleidon
[307] and Gerety & Mulherin [771]). Market closure induces agents to trade with
more intensity and less price elasticity for hedging reasons towards the beginning
and the end of the trading period. As information is diffused without trading
opportunities, portfolios at the opening differ in general from optimal positions and
typically involve large trades. Similarly, in preparation for an overnight non-trading
period, agents will trade more. Heterogeneity across agents regarding preferences
and risk tolerance leads to high volume in the opening and before the market closure.
Under some conditions, also the volatility increases and a monopolist dealer will
perform price discrimination, resulting in a wide bid-ask spread associated with
large trading volume.
Hong & Wang [965] allow for trades motivated by both informational and
hedging reasons with market closures. Market closures preclude not only trades
but also learning, so that information asymmetry increases during market closures.
There is a time-varying hedging effect and a time-varying information asymmetry
effect. Under some conditions, these two effects generate U-shaped patterns in
volume, mean and volatility of returns. Moreover, open-to-open returns are more
volatile than close-to-close returns and returns over trading periods are more volatile
than returns over non-trading periods.
Intraweek regularities have been also observed in asset prices (see Harris [898]
and Hawawini & Keim [916]). In particular, returns earned during the weekend
are typically negative (see French [739]). A model in the spirit of Admati &
Pfleiderer [20] allowing to describe this type of regularities has been proposed in
Admati & Pfleiderer [21]. The key ingredient is that in equilibrium buy and sell
liquidity trades and, therefore, buying and selling trading volume are concentrated in
different periods. When buy liquidity traders trade in the market, prices are high and
vice versa, so that patterns in order imbalance emerge. All discretionary liquidity
buyers (sellers) trade in the same period, but liquidity buyers and liquidity sellers
are concentrated in different periods. Moreover, systematic patterns emerge due to
the presence of privately informed traders. By assuming that information remains
private for more than one trading period and that liquidity traders act strategically (in
the sense that they can postpone their trades), intraday regularities can be generated
as shown in Foster & Viswanathan [729]. Confirming the empirical observations
(see Foster & Viswanathan [731]), if the insider has more information on Monday,
then the variance of price changes and trading costs (bid-ask spreads and the market
10.5 Market Abuse: Insider Trading and Market Manipulation 629
maker’s sensitivity to order flow) on Monday are highest and volume is lower than
on Tuesday (liquidity traders do not trade on Monday).
10.5 Market Abuse: Insider Trading and Market

Manipulation
Market abuse refers to a misuse of the market by its participants. This definition is
of course quite vague and for a more precise definition one has to refer to specific
financial market regulations. There are two main classes of market abuse: insider
trading and market manipulation. Insider trading refers to the fact that an agent
(insider trader) exploits his privileged information when trading in the market,
thereby earning positive excess returns. Market manipulation consists of many
different actions, some of them identified by the financial markets regulation. Of
course, insider trading and market manipulation are two strictly related phenomena
and it is difficult to establish a precise boundary between the two. However, one can
identify three main differences. First, insider trading is always based on privileged
information, while market manipulation is not necessarily based on privileged
information. Second, market manipulation affects market prices, whereas insider
traders typically do not want to affect market prices. Third, trades based on inside
information are typically followed by other trades based on information, while
trades originated by market manipulation are more likely to be reversed.
Insider Trading
The notion of insider trading depends on the regulation adopted in a specific

financial market. Typically, a class of market participants is identified by the
regulator as potential insider traders (e.g., CEOs and agents having a privileged
relation with a firm). In Bhattacharya & Daouk [223], it is shown that among 103
countries having a stock market, 87 have adopted an insider trading regulation.
Insider trading has effects on the market and on the functioning of the firms. In the
literature, benefits and costs of insider trading have been identified. The following
arguments have been put forward in favor of a light insider trading regulation,
pointing out some of the benefits generated by the presence of insider traders in
the market:
• there is more diffusion of information in the market, market prices are more
efficient, price discovery is quicker and, therefore, agents have better risk sharing
opportunities;
• because of reduced risk, asset prices are higher and the cost of capital is lower;
• the opportunity to engage in profitable insider trading activities provides a way
to compensate the managers of a firm;
• the presence of insider traders in the market leads to lower information acquisi-
tion costs for the outsiders;
• greater informational efficiency in the market should encourage efficient invest-
ment decisions by the firms.
On the other hand, the following arguments have been put forward in favor of
a more pervasive insider trading regulation, pointing out the possible drawbacks of
the presence of insider traders in the market:
• insider trading generates an adverse selection problem and, therefore, low
liquidity, large bid-ask spreads, high costs of capital and, under some conditions,
a market breakdown can even occur;
• insider trading generates a cost to the shareholders since it exploits information
which is not publicly available to outsiders, resulting in a lower efficiency of the
firms;
• managers may have an incentive to mostly consider their insider trading profits.
Summing up, on the one hand insider trading increases the diffusion of information
in the market, on the other hand it induces an adverse selection effect which may
discourage agents from trading in the market.
In previous sections of this chapter, we have discussed the effects of the presence
of insider traders in the market by considering market indicators such as efficiency,
liquidity and volatility. In the following, we concentrate on the welfare implications
of insider trading. In this direction, Bhattacharya & Nicodano [222] show that the
presence of insiders can have beneficial effects on the outsiders’ welfare when
the equilibrium trades of the insiders are small compared to the liquidity based
trades of the outsiders and when the adverse selection effect is not too strong. In
their model, Bhattacharya & Nicodano [222] assume inflexible ex-ante aggregate
investment choices by the agents. In this setting, the improvement of risk sharing
with more efficient prices (due to insider trading) compensates the losses due to the
adverse selection effect. On the contrary, when the adverse selection is strong and
investment choices are flexible (i.e., outsiders reduce their investment because of
the presence of insider traders), a prohibition of insider trading may lead to a Pareto
improvement (see also Ausubel [89]). Dow & Rahi [586] show that the welfare
effects of insider trading depend on how the revelation of information changes the
risk sharing opportunities in the market. Fishman & Hagerty [709] show that insider
trading may make a (non-perfectly competitive) security market less efficient if
outsiders gather less information. In a model with endogenously determined and
price-sensitive investment, Leland [1181] shows that, if insider trading is permitted
then stock prices, expected real investments and firm’s profits are higher and the
market is less liquid, more volatile and more efficient (in the sense that prices better
reflect the information). Insider trading decreases both the expected return and the
risk of outsiders’ investments. The total welfare may increase as well as decrease
in the presence of insider trading, while outside investors and liquidity traders
suffer a welfare loss. On the contrary, insiders and owners of the company benefit
from insider trading. In the model of Leland [1181], the key factor in determining
the welfare consequences of insider trading is represented by the sensitivity of

investment of the company to the current price. In the presence of a high sensitivity,
insider trading turns out to be beneficial.
The empirical analysis of insider trading phenomena concerns three main
aspects: market effects (and, in particular, diffusion of privileged information),
insider traders’ profits and the informational content of insider trades.
The capability of insider trading to quickly disseminate private information has
been tested with positive evidence in Damodaran & Liu [512], Cornell & Sirri
[499], Meulbroek [1333], Aktas et al. [37]. It is shown that illegal insider trading
is associated with quick price discovery: the stock market detects insider trades
and incorporates a large fraction (about 50%) of their information in prices before
the information is made public. Trading days where insider trading activities have
occurred are characterized by a high volume, but the proportion of that volume
due to insiders is small, thus implying that there are many market participants who
just follow the insider traders. Moreover, Bhattacharya et al. [225] have shown
that in some emerging markets there is little market response to public market
announcements, since the new information was already incorporated in market
prices, due to the presence of unrestricted insider trading. On the other hand,
Chakravarty & McConnell [392] find that insider trading does not affect market
prices in a significant way.
Analysing legal transactions by corporate insiders, Jaffe [1007], Seyhun [1520,
1521], Lakonishok & Lee [1158], Jeng et al. [1030] show that insider trading
is profitable (i.e., there is evidence of abnormal returns) and that insider traders
are actually informed. However, negative abnormal returns have been reported
in Eckbo & Smith [624]. Insiders’ purchases (sales, respectively) are a good
(bad, respectively) news and predict positive (negative, respectively) future returns.
Furthermore, insider purchases are more informative and earn more excess returns
than insider sales. Seyhun [1520] and Rozeff & Zaman [1482] show that, taking
into account transaction costs, outsiders do not earn excess returns when they try to
imitate insiders. Confirming asset overreaction, insider buying tends to increase as
a stock changes from a growth stock to a value stock (see Rozeff & Zaman [1483]).
The insider trading regulation can take different forms. In what follows, we shall
concentrate on the following types of regulation:
• public disclosure of trades by insiders and trade restrictions;
• company-level regulation of trades by insiders;
• disclosure of information by firms;
• legal prosecution.
In most financial markets of developed economies, public disclosure of insider
trades is mandatory. The effects of public disclosure on insider trading have been
considered in Huddart et al. [989], in the context of the Kyle [1147] model, as well
as in Huddart et al. [988], as we have already mentioned in Sect. 10.4. Furthermore,
in some financial markets the financial regulation prohibits short sales by insiders
and the opportunity of profiting from short term offsetting transactions.
Concerning the second element in the above list, company-level regulation of

insider trading is widespread: typically, there is a blackout period during which the
company prohibits trading by its insiders (e.g., around quarterly earnings announce-
ments, mergers, takeover bids). Corporate policies are designed to minimize the
costs for the company, given the constraints imposed by the financial regulation,
improve the liquidity of the market and maximize the value of the firm. During a
blackout period, it is shown in Bettis et al. [218] that insider trading is lower than in
normal days and this fact induces a greater market liquidity. Note also that, a single
company and the society as a whole may disagree on the opportunity of imposing
insider trading restrictions, since they exhibit different cost-incentives (see Khanna
et al. [1088]).
Public disclosure of inside information by firms precludes profits by insider
trading. A manager may decide not to disclose some information in order to
exploit it by trading in the market. An equilibrium model with a manager whose
compensation is increasing in the stock price is presented in Narayanan [1371].
It is shown that, depending on the pay-performance sensitivity of the manager,
full disclosure, nondisclosure or partial disclosure can be observed in equilibrium.
Partial disclosure implies that the manager does not disclose bad news and only
discloses good news. If the manager’s pay-performance sensitivity is low, then the
manager will not disclose bad and good news, while full disclosure occurs when
the manager’s sensitivity is high. Leuz & Verrecchia [1202] show that disclosure by
firms reduces the bid-ask spread and increases volume.
Legal prosecution is made against trades motivated by non-public information.
While legal prosecution is adopted in almost all developed countries, it is not easy
to prosecute an agent for insider trading and in some countries legal prosecutions
are indeed rare events. The enforcement of insider trading regulations is costly, in
the sense that there is a social cost in enforcing insider trading regulations (costly
investigation). DeMarzo et al. [552] model the optimal enforcement of insider
trading regulations by maximizing the expected utility of uninformed traders. It
turns out that a non-random threshold policy is optimal, i.e., the regulator starts
an insider trading investigation if and only if the market volume exceeds some
threshold and the optimal penalty is a high fixed penalty for trading above a critical
value and nothing below. Therefore, from a social welfare point of view, it is
optimal to allow for limited insider trading. Shin [1539] reaches a similar conclusion
by minimizing the liquidity traders’ expected trading costs when insider trading
regulation improves the precision of the information of market professionals.
Estimating undetected insider trading before acquisition by means of the abnor-
mal volume before the announcement, Bris [304] shows that insider trading
enforcement increases both the incidence and the profitability of insider trading.
However, there is a negative relationship between the insider trading profits and
the toughness of the law. During the ’80s, the sanctions to insider trading behavior
have been increased in the U.S. market. However, according to Seyhun [1521],
the increase in the sanctions did not induce a decline in insider trading activity.
The cost of equity is not affected by the introduction of insider trading laws, but
decreases (and the price informativeness increases) after the first prosecution (see
Bhattacharya & Daouk [223, 224], Fernandes & Ferreira [682]).
Market Manipulation
Following Allen & Gale [44] and Kyle & Viswanathan [1149], we can identify
three different forms of market manipulation: trade based manipulation, action
based manipulation, and information based manipulation. Trade based manipu-
lation occurs when some market participant buys and/or sells stocks over time
by trading against his information or preferences, with the purpose of inducing a
price movement that will allow for profitable trading opportunities in the future.
Trade based manipulation can be based on privileged information or not, by simply
exploiting market power (squeezes, corners). Note also that an agent can manipulate
the market by exploiting the false opinions of other agents who believe that he is
informed. Action based manipulation consists in actions that change the actual or
the perceived value of an asset, while information based manipulation consists in
releasing false information or spreading false rumors in the market.
Trade based manipulation based on market power has been analysed in Jarrow
[1020], assuming the presence in the market of a large trader (i.e., a trader whose
trades change market prices, with no private information). In this context, a market
manipulation strategy is a strategy which generates positive wealth with no risk.
Sufficient conditions under which the large trader cannot manipulate the market
are established. Under general conditions, if the price process depends on the past
sequence of the large trader’s holdings (as opposed to only his current holdings),
then market manipulation is possible.
The possibility of trade based manipulation based on private information has
been analysed in several papers. Allen & Gale [44] show that a uninformed trader
may manipulate the market provided that other agents in the market assign a
positive probability to the manipulator being an informed trader. Allen & Gorton
[47] show that manipulation can occur in a market microstructure model similar to
the model of Kyle [1147] by introducing asymmetric noise traders (i.e., liquidity
sales are more likely than liquidity purchases). Information dissimulation by an
insider trader in the Kyle [1147] model may become manipulation (meaning that
the informed trader trades against his information) if the market maker does not
perfectly know that an informed agent exists in the market and does not know the
nature of his information (see Chakraborty & Yilmaz [388, 389]). When agents
become informed sequentially, they trade aggressively in the initial period and
then reverse partially their trades. Brunnermeier [318] shows that the insider may
manipulate the market if he is allowed to trade twice before and after the information
is made public. John & Narayanan [1034] and Fishman & Hagerty [710] have shown
that a trade disclosure rule may induce the insider to manipulate the market by
trading against his information. This result is due to the inability of the market to
observe the motive for trading of the insider (information based or liquidity based).
Mandatory disclosure leads to profitable trading opportunities for the (informed or

uninformed) insider. Manipulation occurs when there are many liquidity traders or
when the informational advantage is small. In this case, market efficiency is reduced
by mandatory disclosure and mandatory disclosure can make insiders better off
and outsiders worse off. The presence of competition among insiders reduces the
likelihood of manipulation.
Models of action based manipulation have been developed in Vila [1623], Bag-
noli & Lipman [107], Gerard, & Nanda [770]: agents trade and make profits in
correspondence of a takeover bid announcement or of seasoned equity offerings by
depressing the market before the offer. Models of information based manipulation
have been proposed in Vila [1623] and Benabou & Laroque [189]. In this case,
an agent manipulates the market by releasing false information. If an agent has a
reputation of being informed about a stock and the information is noisy (i.e., it
is difficult to verify), then he can successfully manipulate the market by releasing
distorted information.
10.6 Market Design
A financial market is not an abstract entity: in this chapter as well as in Chap. 8,

we have seen that the functioning of a market and, in particular, its efficiency
(from a welfare and from an informational point of view) depend on its institutional
features. In this section, we address market design issues, focusing on the following
topics: transaction taxes, concentration-fragmentation, order/quote-driven market,
continuous/discrete time market, market transparency, and circuit breakers. We refer
to Madhavan [1275] for a survey on market microstructure and market design.
Transaction Taxes
A highly integrated financial system may go under pressure because of temporary

imbalance between supply and demand. Such an order imbalance can be due to
the fact that traders are highly reactive to the arrival of new information in the
market (aggressive speculation), to noise trading by uninformed agents, or to the
adoption of portfolio insurance strategies by institutional traders. We have already
mentioned in Sect. 9.3 that speculation (a trading strategy aiming to revert asset
prices towards the corresponding fundamentals) does not necessarily stabilize the
market. In order to limit the impact of speculation and of noise trading and to reduce
market volatility, some authors have suggested the introduction of transaction taxes,
see Tobin [1597], Summers & Summers [1579], Stiglitz [1569]. Supporters of
transaction taxes follow Keynes [1087]: “It is usually agreed that casinos should
in public interest be inaccessible and expensive. And perhaps the same is true of
stock exchanges”. In several cases, taxes are paid on stocks held for a short period
10.6 Market Design 635
against short term speculation and noise trading. There are other motivations for the
introduction of a transaction tax: (short term) speculation may come from privileged
information (insider trading) and the revelation of information may also induce
a cost for the society (restricted risk sharing opportunities due to the Hirshleifer
effect, see Chap. 8). Stiglitz [1569] points out that financial analysts invest too
many resources in financial research, thereby inducing an inefficient outcome: in
this case, the introduction of a transaction tax may help to reduce this inefficiency. A
tax on short term trading induces agents to extend their trading horizon and reduces
the number of myopic traders. However, it has to be noted that transaction taxes
increase the required rates of return and, therefore, the cost of capital. In this sense,
the introduction of transaction taxes can lead to a deterioration of the liquidity and
of the efficiency of the market. Moreover, it is not easy to assess the distributional
effects and costs of a transaction tax. We refer to Schwert & Seguin [1512] for a
survey on the benefits and the costs generated by the introduction of transaction
taxes.
Subrahmanyam [1577] has addressed the effect of transaction taxes on financial
markets, showing that, in the presence of several informed traders, a transaction
tax leads to a decrease of market liquidity and of the profits of informed traders
(informed traders trade less aggressively). However, the opposite conclusion is
obtained in a market with a monopolist informed trader. A positive effect of
a transaction tax is that agents invest less resources to get early acquisition of
information. A transaction tax induces agents to acquire long term information
rather than short term information. Recall that, as we have discussed in Chap. 8, the
revelation of information by market prices leads to two competing effects: on the
one hand, the revelation of information concerning the agents’ risk exposure leads
to a welfare loss because agents have limited risk sharing opportunities (Hirshleifer
effect), on the other hand, the revelation of information on extra risk factors can
generate a welfare benefit, allowing for more efficient hedging possibilities. Under
some conditions, Dow & Rahi [587] show that a speculator may be better off in the
presence of a transaction tax. Moreover, a tax may lead to a Pareto improvement.
Indeed, a transaction tax always reduces the informativeness of market prices: in this
perspective, the introduction of a transaction tax may also induce a positive welfare
effect if the Hirshleifer effect is very strong.
The empirical evidence is mostly against the introduction of transaction taxes
(see, e.g., Schwert & Seguin [1512]). Considering the Swedish stock market,
Umlauf [1604] shows that transaction taxes do not reduce volatility, while Jones
& Seguin [1044], Aliber et al. [40] show that a reduction in transaction costs has
led to an increase in volume and a decrease in volatility in the NYSE and exchange
rate markets. Song & Zhang [1557] have developed a theoretical model showing
that a transaction tax may produce either an increase in market volatility or a
decrease. Hsieh & Miller [970] and Chowdhry & Nanda [442] have shown that
the introduction of margin requirements does not produce effects on volatility and,
under some conditions, margin requirements may destabilize the market.
Concentration and Fragmentation
The concentration/fragmentation of exchanges in an economy with asymmetric-

private information and liquidity traders is a complex issue. In general, market
depth and liquidity are positively related to trading volume and to the agents’ rate of
participation in the market. However, this does not suffice to induce all the agents to
trade in a single market. In support of market concentration, there are microstructure
models (see in particular Admati & Pfleiderer [18] and Kyle [1147]) showing that
liquidity traders tend to concentrate their trades in order to reduce transaction costs
(high volume is associated with high liquidity and low transaction costs) and that the
insider trader’s intensity of trade is positively related to the liquidity of the market
and, therefore, to the trading volume generated by liquidity traders. Liquidity traders
create volume, which is further amplified by the demand of informed traders, as
already discussed above. On the other hand, there are many examples of assets
simultaneously traded in more than one market (cross listing) and of financial
markets organized through more than one trading system (with different features),
depending on the order size and/or on the agents.
The analysis of market concentration/fragmentation depends on the presence of
informed traders in the market. Let us first consider the case of agents endowed
with homogeneous information. In Pagano [1394], considering an imperfectly
competitive market with hedgers having a stochastic initial asset endowment and
without private information, it is shown that exchanges tend to concentrate in a
single market (in the absence of differential transaction costs). Volume and a greater
number of agents in the market induce high liquidity and low transaction costs and,
therefore, all the agents trade in the same market. If transaction costs differ, then in
equilibrium an asset can be traded in two markets. In this case, there is a separation
phenomenon: traders will cluster together according to the size of their transactions.
In the more expensive market there are more traders and greater diversity in the
initial endowment. One of the two markets can be organized through a direct search
of the trading partner. In general, concentration of trades is Pareto superior to two-
market fragmentation. In Mendelson [1323], it is shown that fragmentation reduces
the expected trading volume, increases price variance and decreases expected gains
from trade, while it improves the quality of price signals. The existence of a two-
market equilibrium has been shown in Chowdhry & Nanda [441] in the context
of an economy populated by agents with short-living information and by liquidity
traders, assuming that some of them are discretionary liquidity traders who are
allowed to place their orders in both markets (looking for the market with lower
transaction costs). In equilibrium, it is shown that discretionary liquidity traders
place their orders in the market which has more non-discretionary liquidity traders,
thereby creating a dominating market. In turn, the mechanism described in Admati
& Pfleiderer [20] works and the dominating market will also attract more informed
traders.
When two or more locations compete for listing decisions by firms, then two
different effects emerge: liquidity traders prefer disclosure requirements with low
trading costs, whereas corporate insiders (those who actually make the listing
decision) prefer low disclosure requirements. Huddart et al. [987] show that
locations engage in a race for the top with an increase in disclosure requirements
and low trading costs: again, the preferences of liquidity traders drive the decisions
of insiders, since the latter prefer to trade in a deep market. Under some conditions,
liquidity traders flow to low disclosure exchanges. However, there are other driving
forces behind the decision to list in more than one market (for instance, the desire
to increase the set of potential shareholders, see Karolyi [1074]). By analysing the
listings in the NYSE, Doidge et al. [580] show that companies want to increase the
level of protection of shareholders.
In many financial markets, a separation is observed between a market for small
and medium size orders (downstairs market) and a market for block trades (upstairs
market). In Seppi [1517], a market is analysed where a large institution trades either
to exploit private information or for liquidity reasons. The institution can execute a
large order or can split it in a sequence of small orders. It is shown that the institution
trading for liquidity reasons always trades a large order, while the institution may
prefer to trade a sequence of small orders when trading for informational reasons.
By trading a block, a trader signals that there is no informational content in his
trade. However, in some cases a block order can also contain information (see Seppi
[1518]). Order separation mitigates the adverse selection problem. A separating
equilibrium is observed in many financial markets: for instance, in the NYSE there
is a downstairs market for medium-small orders and an upstairs market for block
trading. The downstairs market works through a continuous intraday market, while
the upstairs market works through a search-brokerage mechanism where prices are
determined through negotiation. The downstairs market is more efficient, while the
upstairs market is deeper. Keim & Madhavan [1081] show that as a block order
is sent upstairs, information leakage occurs before the trade is being executed and
the temporary price impact or liquidity effect is a concave function of the order
size (upstairs intermediation costs). Madhavan & Cheng [1276] confirm the result
of Seppi [1517]: in the NYSE, a large part of block trades are executed in the
downstairs market and only traders who can signal in a credible way that they are
trading for liquidity reasons trade in the upstairs market. Brokers in the upstairs
market having some information about the agent and the motivation of a trade
(liquidity or information) help this separation between liquidity and informed block
trading. As the upstairs market is not anonymous, an agent can develop a reputation
of being an informationless trader. The price impact of a large block trade in the
upstairs market is smaller than that of trades executed downstairs.
An alternative explanation for the existence of an upstairs and a downstairs
market has been provided in Grossman [843]. The motivation is that agents do not
participate continuously in all the markets and intermediaries are repositories of
information about unexpressed demands of non-participating customers. They may
know what states of nature are likely to induce agents to trade and have information
about unexpressed order flow. This knowledge increases the effective liquidity of
the upstairs market, making it well suited for block trades. On the other hand,
customers bear search costs. Under some conditions, the downstairs and the upstairs
markets coexist in equilibrium. The empirical evidence provided in Bessembinder

& Venkataraman [217] supports the analysis of Grossman [843] and Seppi [1518].
Trading Systems
A comparative analysis of different trading systems requires a welfare analysis,

focusing in particular on the effects on uninformed-liquidity trading. As we have
already discussed in the previous sections of this chapter, some indirect measures of
welfare are represented by market liquidity, price impact, bid-ask spread, volatility
and informational efficiency.
Madhavan [1272] compares the performance of a continuous quote-driven
market (i.e., dealers set prices before order submission) with that of a continuous
order-driven market (i.e., traders submit orders before prices are set). In this
context, continuity means that an order is executed immediately upon submission.
In the model, dealers are assumed to be risk neutral and investors trade for both
informational and hedging reasons. When the adverse selection is strong, in the
sense that the private information is very precise, then an equilibrium does not
exist (market breakdown). It is shown that a dealer market is more robust than
a continuous auction market with respect to adverse selection. Indeed, for a high
degree of adverse selection, the continuous order-driven market does not admit
an equilibrium, while an equilibrium exists in a dealer market. In the dealer
market, transaction prices follow a martingale process and prices are efficient in
a semi-strong form. Moreover, prices are less variable than those corresponding
to a continuous time market. However, these properties are not satisfied in the
continuous auction market. Furthermore, allowing for the free entry of market
makers in the continuous auction market, the two price systems turn out to coincide.
The bid-ask spread is increasing in the degree of adverse selection, while it is
decreasing in the risk aversion of the agents. However, it has to be noted that a
discrete time system does not ensure immediate order execution and provides less
information to the investors. The different robustness of discrete and continuous
auction systems with respect to adverse selection is due to an argument similar to
that proposed in Glosten [787]. A discrete call auction is less risky for uninformed
traders, since the negative effects associated with trading with an insider are weaker
than in a continuous auction market with pooled orders.
A continuous auction market is an expensive trading mechanism and is typically
less deep than a discrete time call auction market. The disadvantages of a discrete
time call auction are that information diffusion is slow with high volatility and that
there is a loss of immediacy and, therefore, risk management opportunities can be
negatively affected. In this sense, there is a trade-off in organizing an auction market
in continuous time or in discrete time. In Schnitzlen [1507], the performance of a
continuous and of a discrete call auction have been evaluated through laboratory
experiments. Confirming theoretical results, a discrete call auction is shown to be

more robust than a continuous auction with respect to adverse selection. Liquidity is
also greater and noise traders incur in smaller transaction costs than in a continuous
market. Amihud et al. [60], Amihud & Mendelson [59] have empirically shown that
shifting from a daily call auction to a repeated continuous trading session improves
the liquidity and the efficiency of the market.
Market Transparency and Disclosure
Financial markets can exhibit different degrees of transparency, depending on the

extent to which market participants (i.e., market makers, brokers, dealers and
traders) know the order flows and the quotes of the other agents. Even though market
transparency represents a highly debated topic, in recent years there has been a gen-
eral trend towards a greater transparency in most financial markets. Transparency
helps uninformed traders to learn private information and, therefore, helps to reduce
adverse selection problems. However, in a transparent market informed investors
trade less aggressively and, therefore, this leads to a reduction of the information
disseminated in the market by insider traders. Moreover, in a transparent market
agents have a smaller incentive to buy information and the Hirshleifer effect may
produce negative welfare effects. In this sense, a greater transparency can reduce the
risk sharing opportunities. This effect is particularly strong for dealers who have less
time to manage their inventory position. In a transparent market, transaction costs
for a dealer who has to execute a large uninformative trade or for an informed trader
can be high. In general, a dealer market is more (pre-trade) quote-transparent than
an auction market and a continuous auction market is more order-transparent than a
dealer market. Market transparency has many dimensions and we can distinguish
between order (imbalance) transparency, trader anonymity, pre- and post-trade
quote disclosure.
Order transparency has been investigated in Pagano & Roell [1395], where the
authors propose the following ranking in terms of order transparency: (i) transparent
auction (a batch auction where agents know all the individual orders); (ii) continuous
time auction (where agents observe all the individual past orders); (iii) dealer market
(where each dealer does not observe the orders of other dealers). Two specifications
of a dealer market are given, under the assumption that dealers are risk neutral:
immediate last trade publication and no trade publication. The latter specification
represents the less transparent setting, while the former has a transparency similar
to that of a continuous time auction. Note that a batch auction cannot be ranked
with respect to the above market specifications: indeed, in a batch auction market
the dealer sets the price observing the total order flow and not the individual orders.
In a model with an insider trader, Pagano & Roell [1395] show that the expected
transaction costs for uninformed traders are lower in a transparent auction than
in a dealer market. The costs in the batch and in the continuous auction cannot
be compared with the others. These results are obtained by assuming that the
insider trader adopts the same strategy in all three types of market. Under suitable
conditions, this ranking in transaction costs also holds when the insider trader
is allowed to modify his trading strategy depending on the market. In general,
transparency is associated with lower transaction costs.
Anonymity of trading is strictly related to market transparency. In Forster &
George [727] it is shown that, when market makers have some information about the
trading motives of the investors, execution costs for liquidity traders and gains for
informed traders are lower if there is enough competition among informed agents. If
brokers have information about the order flow motivation (informed/ uninformed),
then the bid-ask spread decreases (see Benveniste et al. [194]). Specialist markets
are less anonymous than dealer markets (as the NASDAQ), in particular when orders
are channelled through brokers as in the NYSE (see Benveniste et al. [194] and
Garfinkel & Nimalendran [758]).
Madhavan [1274] analyses information disclosure about the order flow and the
order imbalance. In particular, the author analyses the effect of publicizing the order
imbalance due to price inelastic demands (liquidity trades) prior to trading. This
demand can be interpreted as the order submitted by a large uninformed trader and
transparency can be interpreted as sunshine trading: the large trader announces his
trade declaring that the trade is not based on an informational motive. In the model, it
is assumed that agents are endowed with symmetric private information. Disclosure
of an order imbalance due to liquidity trades (or uninformative trades) should
have a stabilizing effect. It is shown that in a transparent market prices are more
informative, volatility is reduced and market liquidity is increased, provided that the
market is sufficiently large and liquid. However, if this is not the case, then market
transparency may increase the volatility generated by transitory order imbalances
and may lead to less liquidity and higher transaction costs. Moreover, if the agents’
information is very precise, in a transparent market there is no equilibrium, even
though an equilibrium would exist in an opaque market. The rationale of this result
is that market transparency may lead to a market failure: agents endowed with very
precise information prefer not to trade in a transparent market. Madhavan et al.
[1277] have empirically investigated the pre-trade transparency about investors’
latent demands present in the limit order book by relying on data from the Toronto
stock exchange, reporting that greater transparency (i.e., public disclosure of the
limit order book) is associated with larger execution costs and volatility. A welfare
analysis of sunshine trading has been carried out in Admati & Pfleiderer [23]. In the
model, some liquidity traders can pre-announce the size of their orders, declaring
also their nature (uninformative). It is shown that such agents have lower trading
costs, but the costs to informed and liquidity traders who cannot pre-announce their
trades increase. Sunshine trading increases the informational content of prices, while
the effect on volatility is not clear a priori.
In Biais [231], in the context of a model without asymmetric-private informa-
tion, it is shown that information about quotes, trades and inventories of other
intermediaries may affect the bid-ask spread. In the model, the number of (risk
averse) dealers-market makers is endogenous. In a centralized (auction) market,
there are market makers and limit order traders, who can monitor the positions-
quotes of their competitors. In a fragmented market, there are dealers who cannot
monitor the positions of other intermediaries. The expected bid-ask spread in a
centralized market is equal to the expected bid-ask spread in a decentralized market
where dealers do not know positions. However, the spread is more volatile in
centralized than in fragmented markets. Note that fragmentation (associated with
low transparency) provides the dealer with a monopolistic surplus (in the sense
of wide bid-ask spreads), which is decreasing in the number of traders. Assuming
private-asymmetric information, risk averse dealers may prefer a non-transparent
inter-dealer market. In Lyons [1261], it is observed that the Hirshleifer effect works
against dealers in a transparent market: greater order flow transparency accelerates
the diffusion of information by prices, leaving dealers less time to manage their
inventory position. In this context, market transparency can limit the risk sharing
possibilities. The optimal level of transparency for dealers is at an intermediate
level, while both high and low levels of transparency restrict risk sharing. In this
sense, dealers prefer imperfect market transparency (i.e., the market order flow can
be observed up to a noise component).
Madhavan [1273] has analysed the post-trade disclosure of prices by dealers
in the context of an economy populated by liquidity and informed traders. Large
liquidity and informed traders prefer a market with no price disclosure because
in that case they can more easily conceal their trades, thus reducing the execution
costs. Non-transparency allows dealers to set wide spreads since there is less price
competition. Moreover, non-transparency generates an informational advantage
and, therefore, large gains for the dealer. Because of these effects, dealers, informed
and large traders will not disclose their trades-prices unless disclosure is mandatory.
Porter & Weaver [1428] show that dealers in the NASDAQ delay large trades
disclosure and the delay is associated with the informational content of the trade. In
this setting, transparency reduces volatility and increases price efficiency. In a model
with asymmetric information, the welfare effects of post-trade price disclosure
for a risk averse investor are ambiguous (see Naik et al. [1368]). Disclosure of
trades reduces the adverse selection effects but increases the price revision risk for
the dealer. The liquidity effects of price disclosure in a dealer market have been
empirically investigated in Gemmill [766] on the basis of London Stock Exchange
data. It is shown that transparency (as measured by the delay in the publication of
trades) does not affect execution costs, liquidity, price adjustment and the diffusion
of information.
In Bloomfield & O’Hara [252, 253] and Flood et al. [717], market transparency
has been evaluated by means of laboratory experiments. In Bloomfield & O’Hara
[252], it is shown that the disclosure of trades and quotes increases the informational
efficiency of transaction prices, but also increases the bid-ask spread by reducing
competition among dealers. As a result, market makers benefit from trade disclosure
at the expense of liquidity and informed traders who cannot time their trades.
Trade disclosure can have important effects on the market behavior, whereas
quote disclosure does not. Flood et al. [717] investigate quote disclosure (pre-
trade transparency) in a quote-driven market where trade information is never
revealed. Quote transparency reduces opening spreads and increases volume, but
price discovery is much faster in a opaque market and prices are also more efficient.
This is due to the low competition among dealers in a transparent market, leading
to a trade-off between liquidity and price efficiency. Market transparency affects
competition between different trading systems. Note that liquidity traders prefer
transparent markets, but informed traders prefer opaque markets. In Bloomfield &
O’Hara [253], it is shown that when there are two markets, the less transparent
one tends to become the dominant location. Less transparent dealers are more
aggressive in the early rounds of trading, attract order flow and informative trades
and use the resulting informational advantage to quote narrower spreads and
attract liquidity trades. Bloomfield & O’Hara [253] support these findings with
laboratory experiments. It is also shown that most dealers prefer low transparency.
However, the existence of a more transparent market is ensured by the fact that the
informational advantage of non-transparent dealers decreases with the number of
dealers.
Circuit Breakers, Trading Halts and Price Continuity
After the 1987 market breakdown, the Presidential Task Force on Market Mech-
anism was nominated to investigate the crash. Among other policy measures, the
committee recommended the adoption of circuit breakers in order to close the
market when it does not work “properly”: “circuit breaker mechanisms (such as
price limits and coordinated trading halts) should be formulated and implemented
to protect the market system” (Presidential Task Force on Market Mechanism
[1434]). The possibility of adopting circuit breakers is a highly debated issue from a
theoretical and an empirical point of view. Indeed, the effects of circuit breakers
are typically ambiguous. A trading halt is called by specialists when a unusual
order imbalance that cannot be matched occurs or by market officials when news
with a strong price impact are announced or expected. During a market closure,
the specialist engages in price exploration, receives commitments to trade and then
reopens the market through a call auction matching the commitments to trade. In
some financial markets, the stock exchange authority can call for a trading halt and
then reopen the market when it is less under pressure. Bhattacharya & Spiegel [228]
show that suspension occurs at the NYSE when a firm announces impending news
or when there is a severe order imbalance.
A circuit breaker is needed because of a malfunctioning of the market or in the
presence of strong adverse selection/poor information transmission by prices. In
Chap. 8 as well as in the present chapter, we have seen that uninformed traders can
refuse to trade when they recognize that there is a strong adverse selection effect in
the market. According to this interpretation, a circuit breaker is designed to remove
a (technical) market malfunctioning and/or to improve the diffusion of information
and the price efficiency in the market by weakening adverse selection effects and
execution risk. As a consequence, a circuit breaker should reduce volatility and
volume and prevent a chaotic market behavior. To this end, the characteristics of
the market opening system are crucial. Note also that a circuit breaker should not
affect the long run tendency of the market.
Theoretical arguments are mostly against the adoption of circuit breakers. In a
non-fully revealing rational expectations model with private-asymmetric informa-
tion and noise supply, if agents have the possibility of observing intertemporally
trading volume and price, then agents are able to detect private information
and prices reveal more information (see Blume et al. [257], Brown & Jennings
[310], Grundy & Martin [851]). In this setting, circuit breakers would have a
negative effect. The effects of circuit breakers have also been discussed in studies
explaining intraday patterns of asset prices, showing that high volatility, volume
and spreads typically occur before a natural market closure and also that volatility is
high at the opening of the market (see Brock & Kleidon [307], Gerety & Mulherin
[771, 772], Stoll & Whaley [1571] and see also Sect. 10.4). When expecting a
market closure, similar phenomena should be observed. In Subrahmanyam [1576],
it is shown that agents sub-optimally make trades in advance when they expect
a trading halt and, therefore, a circuit breaker may actually increase the trading
volume, liquidity, price variability and the probability of the price crossing the
circuit breaker bound and decrease price efficiency. Moreover, in a two-market
setting, if a circuit breaker is adopted in the liquid dominant market, both price
variability and liquidity and volume migrate to the satellite market.
In support of circuit breakers, Greenwald & Stein [824] show that when there
is a large (liquidity or information based) trade and uncertainty on the number
of traders aiming to compete for it, traders face a large transactional risk in an
auction (i.e., agents do not know the transaction price and fear cascade dynamics).
The uncertainty on the number of traders may be due to adverse selection in the
market or to price uncertainty/volatility. In general, there is a high transactional risk
when prices are uninformative. This transactional risk restrains agents from trading
and in equilibrium the price will drop substantially. In these situations, circuit
breakers will help a trader by reducing the uncertainty about the traders’ behavior,
transactional risk and adverse selection. Stein [1567] proposes a formal model
showing that, during periods of price uncertainty, continuous trading can decrease
the informational content of prices, while informative halts can improve market
quality. However, note that a market closure preventing trades always induces
hedging-liquidity costs, thus exacerbating liquidity problems. Instead of closing the
market, it may be a good solution to switch to a trading system that is more robust
with respect to adverse selection. In this sense, Madhavan [1272] suggests to switch
to a periodic auction market. Edelen & Gervais [627] show that trading halts serve
as a discipline device for market makers. Indeed, in some cases market makers may
have an incentive to quote a privately optimal pricing schedule (e.g., not executing
orders immediately) and this type of behavior has a negative externality effect for
the market as a whole in terms of loss of order flow in the future when the market
enjoys a reputation for execution quality. A trading halt called by the specialist or
by the exchange commission can act as a coordination device preventing this type
of behavior and increasing exchange profits.
The empirical evidence on the adoption of circuit breakers is ambiguous. While
some positive effects of circuit breakers have been detected in Lauterbach & Ben-
Zion [1169] (low opening order imbalance), in Lee et al. [1173], Edelen & Gervais
[627], Bhattacharya & Spiegel [228] it is observed that NYSE firm-specific trading
halts increase both post-halt volume and volatility. A possible explanation of these
effects is that the reopening mechanism is inefficient and limits the diffusion of
information. Corwin & Lipson [500] confirm the results of Lee et al. [1173] and
show that NYSE trading halts reduce order book depth around the event, increase
the spread but, on the other hand, allow traders to effectively reposition their trades.
The experimental results of Ackert et al. [12] confirm that an imminent trading halt
increases the pre-halt trading volume.
Price continuity requirements are designed to stabilize the market. According
to these requirements, upper limits are placed on absolute price changes during
trading hours. The upper limit depends on past price evolution and/or on the volume
or liquidity of the market. Dealers/market makers set prices in order to satisfy
these requirements. The effect of a bound on the absolute price change given by
a multiple of past order flow has been analysed in Dutta & Madhavan [603]. It
is shown that, under stringent continuity requirements, informed traders will trade
more aggressively and, as a consequence, the price efficiency is unaffected or even
increased if more agents buy private information. This type of rule creates a trend
in asset prices. There is a loss for dealers/market makers and a gain for liquidity
traders. Moreover, price limits might increase market stability (see Chowdhry &
Nanda [442] as well as Sect. 9.5). Kim [1094] empirically observes that volatility
does not usually decrease when price limits are made more restrictive.
We refer the reader to Madhavan [1275], O’Hara [1384], Biais et al. [236], De
Jong & Rindi [540], Foucault et al. [735] for detailed surveys on financial markets
microstructure .
In the Kyle [1147] model discussed in Sect. 10.2, if the linearity condition or
the normality assumption are not imposed, then multiple equilibria arise. However,
regardless of the normality assumption, the uniqueness of the equilibrium holds
if the insider trader can observe noise trades (see Rochet & Vila [1455]). The
model has been extended to a multi-security setting in Caballé & Krishnam [324].
Considering a generalized version of the model of Kyle [1147], Vives [1629] shows
that the presence of the market maker contributes p to the convergence towards the
fundamental value: the rate of convergence is 1= n, where n is the number of
rounds in the adjustment process. On the other hand, p in the absence of a market
marker, the rate of convergence is much slower (1= n1=3 ). A continuous time
extension of the Kyle [1147] model has been developed in Back [97] and Back
& Baruch [99]. In the context of equilibrium models under imperfect competition,
it is also worth to mention that non-fully revealing equilibrium prices have been
obtained in Laffont & Maskin [1155]. In Biais & Hillion [237], in the presence of
imperfect competition, it is shown that adverse selection and market breakdown can
be avoided by introducing derivative securities.
It is interesting to consider the possibility that information can be bought and
sold in the financial markets. In this direction, Admati & Pfleiderer [19] investigate
the direct and indirect sale of information in the context of the Kyle (1985) model:
an insider trader can sell information to other agents or exploit it by trading the
asset. The private information consists in the observation of a signal about the
fundamental value of the asset. In the market, there are N potential buyers and noise
traders. The insider trader sells his information without manipulating its content.
Agents who do not buy the information quit the market because they know that
the price being determined according to the market efficiency condition (10.11)
will be unfair. Therefore, the number of agents in the market, excluding noise
traders, is equal to the number of those who decide to buy information. Agents
have a constant coefficient of absolute risk aversion. If the information owner is risk
neutral, then he does not want to sell his information to any other trader. On the other
hand, if the information owner is risk averse, then he can have an incentive to sell
his private information. Competition with other insider traders reduces the profits
associated with speculation but creates risk sharing opportunities (with a positive
effect for the insider’s welfare). There is a trade-off between competition among
insider traders and risk sharing possibilities. For plausible values of the parameters
of the model, the profit of informed agents is a unimodal function of the number
of agents acquiring information and, therefore, there is a unique optimal number
of insider traders. Typically, if the risk aversion of the information owner is low,
then he does not sell the information, preferring to fully exploit it by trading in the
market. Furthermore, if potential buyers are weakly risk averse then the information
owner sells the information and commits not to trade. For intermediate degrees of
risk aversion, the information owner trades in the market and at the same time sells
the information to other agents. Less precise information is sold to more traders
and leads to lower profits. The indirect sale of information through a mutual fund
is also analysed. By setting appropriate fees (a fixed fee and a per share fee), the
information owner can control the effects of competition among informed traders
and can increase his profits.
The strategy of a market maker endowed with private information has been
analysed in Gould & Verrecchia [814]. The market maker sets a price and then a
trader chooses the quantity he wants to trade. Both the market maker and the trader
in the market can be privately informed. If the market maker is privately informed,
then he will introduce a noise component in the price with respect to the one that
maximizes his expected profits. In Madrigal & Scheinkman [1282], the model has
been extended to an economy populated by noise traders and risk averse insider
traders endowed with private and heterogeneous information. In this setting, the (bid
and ask) prices quoted by the market maker aggregate information with some noise.
The risk neutral market maker sets prices to maximize expected profits conditionally
on the observation of market orders. Prices always reveal information, but never
fully. The equilibrium price is a discontinuous function of market orders, with the
consequence that a slight change in the parameters of the model can lead to a crash
of the market.
Limit order book models and an analysis of the trade-off between market and
limit orders have been developed in Foucault [733], Foucault et al. [734], Hasbrouck
& Schwartz [914], Parlour [1402], Goettler et al. [794], Chakravarty & C. Holden
[391], Rosu [1471], Back & Baruch [100]. For empirical studies on the functioning
of limit order books we refer to Harris & Hasbrouck [900], Biais et al. [239], Harris
& Venkatesh [902], Hollifield et al. [955, 955] (see also Bloomfield et al. [254]
for an experimental analysis). Comparing the NYSE with the NASDAQ, Chung
et al. [450, 451] confirm that limit orders decrease execution costs with respect to
a pure quote-driven market. Insufficient depth in the limit order book relative to
theoretical predictions is reported in Sandas [1494]. In some markets, a specialist or
a dealer faces competition from limit order traders and competition contributes to
lower the bid-ask spread and to increase price efficiency (see Seppi [1519], Brown
& Zhang [312], Baruch [174], Hendershott & Mendelson [933]). In Wahal [1640],
the author provides empirical evidence in favor of a reduction of the bid-ask spread
as competition among market makers increases. A model on dealers’ competition
with payments for order flow and preferencing arrangements has been developed in
Kandel & Marx [1065].
In Chan [400], the author suggests that financial markets microstructure can
generate cross-autocorrelation among stock returns, explaining for example the
coexistence of weak serial correlation of asset returns and positive serial correlation
of index returns and the lead-lag effect. Let us consider a monopolist specialist
market with many assets, where the value of a stock is given by the sum of a
market-wide component and a stock-specific component. A market maker observes
a noisy signal about the value of his stock without instantaneously observing signals
about the value of other stocks. Each signal contains market-wide information and
uncorrelated noise. In a multi-period setting, if the market maker revises his quotes
by observing previous price changes, then stock returns will be serially uncorrelated
and positively cross-autocorrelated. It is shown that the lead-lag effect between large
and small firms can be explained by this argument. Foster & Viswanathan [730]
show that if the relevant random variables have an elliptically contoured distribution,
then volume and price variance are correlated and volume is serially correlated.
Chordia & Subrahmanyan [438] analyse and provide empirical evidence on the
relation between order imbalance and returns, showing that the price pressure due to
autocorrelated imbalances generates a positive relation between lagged imbalances
and returns.
In Chap. 8, we have remarked that in a model with heterogeneous beliefs
or opinions, trading volume is positively related to the degree of heterogeneity.
However, this is not necessarily true in a dealer market model with transaction costs
when liquidity trading is elastic with respect to transaction costs (see George et al.
[769]). Trading volume can increase or decrease with respect to the informational
asymmetry and the dispersion of beliefs: volume increases with respect to the
informational asymmetry (or diversity in beliefs) if and only if liquidity trading

decreases in transaction costs at an increasing rate.
Brokers and, in general, financial intermediaries can be allowed or forbidden to
trade for their own account. The practice of trading both for customers and for their
own account is known as dual capacity. The effects of dual capacity have been
analysed in Fishman & Longstaff [712] and Roell [1456]. On the one hand, the dual
capacity can introduce a conflict of interest between brokers and their customers. On
the other hand, the dual capacity can reduce the costs of trading. The effects of the
dual capacity on market depth, liquidity and welfare are ambiguous in a model with
heterogeneous agents (see Locke et al. [1241]). In a model with asymmetric-private
information, the observation by the broker of the customers’ order flow can have an
informational content.
Insider trading may either increase or decrease the value of a firm, see Manne
[1299] and Manove [1300] for results in favor of the two different conclusions.
Depending on the trade-off between diffusion of information and adverse selection
and its consequences on firm investment, positive or negative welfare consequences
of insider trading have been obtained in Bernhardt et al. [205]. Repullo [1446]
shows that the positive effects of insider trading detected in Leland [1181] are
not robust to the introduction of a noise component in the insider’s information
(see also Medrano & Vives [1317]). As pointed out in Sect. 10.5, allowing for
insider trading can represent a way to compensate managers, since they can use
their private information to earn profits by trading in the market. However, by
allowing for this possibility, firm managers can operate in a detrimental way for
other shareholders and this effect would in turn discourage corporate investment,
thus decreasing the efficiency of the firm. Moreover, the possibility of insider trading
affects the choice among investment projects, since insider traders tend to choose
riskier investment projects in order to take advantage of the greater volatility (see
Bebchuk & Fershtman [184]). In Fischer [704], it is shown that insider trading does
not affect the firm if there are no other agency-related or asymmetric information
problems. Otherwise, prohibiting or imposing delayed registration of insider trades
may be valuable.
According to asymmetric information models of corporate finance, managers
should issue common stocks or convertible debts when stock prices are overvalued
(i.e., managers have a negative private information). This hypothesis is confirmed by
insider trades analysis, showing that insider sales increase and purchases decrease
prior to issues of convertible debt and equity and firms with abnormal insider selling
underperform in the long run (see Kahle [1057]). Noe [1378] shows that managers
exploit market reaction to management earnings forecasts: insider sales (purchases)
increase after a positive (negative) price reaction. Insiders trade after voluntary
disclosures (when the information asymmetry with outsiders is low) and not before.
These trades earn abnormal returns and are not correlated with management forecast
errors. In a signalling model, insider trades complement dividends: dividends
convey positive information (positive price response) when they are joined by
insider buying and negative information when they are joined by insider selling
(see John & Lang [1035] for a theoretical model and empirical evidence). Bagnoli
& Khanna [106] show that insider trades may preclude signaling through observable
actions (e.g., dividend announcements). A relation between information asymmetry
and insider trades is also detected in Huddart & Ke [990], Ke et al. [1077], Huddart
et al. [991]: insiders possess and trade on the basis of specific and significant
information about the firm (e.g., forthcoming accounting disclosures). The evidence
is consistent with the signaling role of insider trades.
A particular type of manipulation can arise by simultaneously trading in deriva-
tives and in the underlying assets. Kumar & Seppi [1142] show that a uninformed
trader may manipulate the market through a sequence of trades in the spot and
in the future markets against informed agents (see also Kyle [1146] on future
markets manipulation). Assuming the presence of a large trader in a model similar
to that proposed by Jarrow [1020] with traders having symmetric information and
complete markets, Jarrow [1021] shows that allowing for trading in a derivative
security allows for manipulation even in a model which otherwise would not admit
manipulation without the derivative security. Manipulation is avoided by excluding
corners through quantity limits and by facilitating information diffusion concerning
trades in two markets. It is generally shown that manipulation increases volatility,
liquidity and returns (we refer to Merrick et al. [1326], Allen et al. [48], Aggarwal
& Wu [27] for an empirical analysis of manipulation by trading on derivatives).
The shareholders’ monitoring activity is affected by market liquidity, since
shareholders can trade in the stock market as an informed agent after having
monitored the company, see Maug [1312]. If monitoring is costly, a liquid market
motivates a large shareholder to monitor the firm and trade on private information.
A liquid market mitigates free riding by small shareholders on large shareholders.
In some cases, it is optimal for the large shareholder to not exert pressure on the
management of the firm and to exploit (negative) information on the firm to trade in
the stock market (see Kahn & Winton [1058] and Faure-Grimaud & Gromb [681]).
Concerning the trading mechanisms of real financial markets, the minimum
tick size has been reduced over the years in many markets. On the one hand, a
reduced tick size should reduce the bid-ask spread through dealers’ competition
and, therefore, benefit liquidity traders. On the other hand, a small tick size could
be detrimental to liquidity providers by decreasing their profits (see Grossman &
Miller [846] and Chordia & Subrahmanyan [437]). The tick size reduction from
eighths to sixteenths in U.S. financial markets has been analysed in several papers,
showing that spreads and depths declined, liquidity traders trading small quantities
are better off but traders placing large orders are damaged (see Jones & Lipson
[1043], Goldstein & Kavajecz [798], Van Ness et al. [1606]). A large tick size on
the NYSE generates the payment for order flow practice and precludes competition
(see Chordia & Subrahmanyan [437]).
Christie & Schultz [446], Christie et al. [445], Barclay [157], Bessembinder
[212], Kandel & Marx [1064], Barclay et al. [159] have remarked the absence
of odd-eighth quotes for 70 of the 100 most capitalized NASDAQ stocks (the
phenomenon is not observed in the NYSE) and that spreads are quite large (at
least 0:25$). There is a positive relation between spread and price rounding (see
Harris [899], Barclay [157], Bessembinder [212], Kandel & Marx [1064], Simaan
et al. [1544], Bessembinder [214]) and internalization-preferencing of the order
flow (see Huang & Stoll [977], Godek [793], Chung et al. [449]). Some authors
suggest that dealers tacitly collude to set wide spreads. Bloomfield & O’Hara [251]
show through experiments that order preferencing can increase bid-ask spreads and
also reduce informational efficiency. Information ambiguity also affects the bid-
ask spread: Routledge & Zin [1475] show that information ambiguity (or model
uncertainty, see Sect. 9.1) increases the bid-ask spread set by an uncertainty averse
monopolist market maker and reduces the market liquidity.
In Huang & Stoll [977] and Bessembinder & Kaufman [216], it is shown that
execution costs on the NASDAQ (before the 1997 reform) were higher than on
the NYSE. The difference can be attributed to higher order processing costs (see
Affleck-Graves et al. [26]), avoidance of odd-eighth quotes and lack of competition
in the NASDAQ market (collusion) (see Christie et al. [445] and Barclay [157])
or order preferencing arrangements (see Huang & Stoll [977]). The reform of
the NASDAQ market and the reduction of the tick size have decreased the (still
positive) difference between execution costs on the NASDAQ and on the NYSE
(see Bessembinder [213], Barclay et al. [158], Weston [1658], Bessembinder [214]).
Based on an analysis of dual listed stocks traded both in Paris (automated market)
and in London (dealer market), De Jong et al. [539] show that effective spreads
are lower in Paris than in London. In Kim et al. [1098], analysing how NYSE
and NASDAQ prices incorporate new information (analyst recommendations), it is
shown that a call auction market incorporates the new information before the dealer
market does. However, information asymmetry can create a loss of liquidity in a call
auction market (i.e., time is needed to find a price) which is instead not observed in
a dealer market. Lamoureux & Schnitzlen [1164] compare through an experimental
investigation the performance of a dealer market when (informed and liquidity)
traders can decide to opt for bilateral search. It is shown that this opportunity induces
dealers to trade more aggressively and that their profits decrease substantially, while
on the contrary price discovery increases.
In Barclay et al. [158] and Chung & Van Ness [452], the 1997 NASDAQ reform
has been analysed. The reform was designed to offer investors more competitive
quotes through the mandatory display of limit orders when they are better than
quotes posted by market makers and the dissemination of superior prices placed
in proprietary trading systems. In this way, the public can compete directly through
limit orders with dealers. This reform has reduced fragmentation, increased market
transparency and introduced some auction features in the market. It is shown that
spreads have tightened by approximately 30%. Chung & Van Ness [452] show that
the decline is particularly large during midday and this can be attributed to the
intraday variation in competition among limit order traders. According to Weston
[1658], the reform has also increased competition and reduced the dealers’ profits.
10.8 Exercises
Exercise 10.1 Consider the model proposed by Kyle [1148] and described in
Sect. 10.1. By making use of the multivariate normality assumption on the random
Q yQ 1 ; : : : ; yQ N ; zQ/, prove Proposition 10.1 (compare also with Kyle [1148,
variables .d;
Appendix A]).
Exercise 10.2 Consider the model proposed by Kyle [1147] and described in
Sect. 10.2.
(i) Suppose that the functions X and P are linear:
Q D ˛ C ˇ dQ
X.d/ and P.Qx C zQ/ D C .Qx C zQ/;
for some constants ˛, ˇ, and to be determined. Show that the profit

maximization constraint (10.10) implies that
1
ˇD and ˛ D ˇ:
2
(ii) By relying on the result obtained in the first step, prove that the market
efficiency condition (10.11) implies that
ˇd2
D and dN D .˛ C ˇ d/:
N
ˇ 2 d2 C z2
Deduce that Proposition 10.3 holds.

Exercise 10.3 In the context of the model introduced in Kyle [1147] and discussed
in Sect. 10.2, let the couple .X; P/ represent the linear equilibrium, as stated in
Proposition 10.3. Prove the following claims:
(i) Letting pQ denote the equilibrium price, it holds that
Q p/ D d2
Var.djQ I
2
(ii) The optimal profits of the insider trader (i.e., the profits associated to his optimal
demand X.d/), conditionally on the observation of the private signal dQ D d, are
given by
N 2
.d d/
4
Q the expected profits
Ex-ante (i.e., before the observation of the realization of d),
of the insider trader are equal to d2 =.4/.
10.8 Exercises 651
Exercise 10.4 Suppose that all the assumptions of the model of Kyle [1147] as
presented in Sect. 10.2 are satisfied, apart from the fact that informed agents do not
observe perfectly the liquidation value dQ of the asset, but rather an imprecise signal
yQ of the form
yQ D dQ C "Q;
where "Q is a normally distributed random variable independent of all the other
random variables appearing in the model, with mean zero and variance Var.Q"/ D "2 .
Show that, under this assumption, the economy admits a unique linear equilibrium
.X; P/, where X W R ! R and P W R ! R are two measurable functions such that
N
X.Qy/ D ˛ C ˇ.Qy d/ and P.Qx C zQ/ D C .Qx C zQ/;
where the constants ˇ and are explicitly given by

s s
z2 1 d4
ˇD 2
and D
d C "2 2 .d2 C "2 /z2
and where the constants ˛ and are given by
ˇ 2 "2 Cz2
ˇ 2 .d2 C"2 /Cz2
D dN and ˛D :
ˇd2 1 2
1 ˇ 2 .d2 C"2 /Cz2 2
In particular, recalling that the market depth is measured by the quantity 1=, this
implies that market depth is increasing with respect to the variance of the error
term appearing in the informed agent’s observation (or, equivalently, decreasing
with respect to the precision of the informed agent’s signal). Similarly, the intensity
of trading of the informed agent, as measured by the coefficient ˇ, is increasing with
respect to the precision of his signal, as measured by the quantity 1="2 .
Exercise 10.5 Consider the model developed in Subrahmanyam [1575] and pre-
sented in Sect. 10.2. In this exercise, we are going to prove Proposition 10.4.
(i) Start from the conjecture that the risk neutral market maker adopts a linear
pricing rule, i.e., letting N xQ C zQ denoting the total market demand, the market
maker sets a price pQ D P.N xQ C zQ/ given by
P.N xQ C zQ/ D dN C .N xQ C zQ/;
for some constant to be determined. Consider the expected utility maximiza-

tion problem of an informed trader, conditionally on the observation of the
private signal yQ D "Q C .
Q Assume that each informed trader conjectures that
the other informed traders have an asset demand of the form xQ D ˇ.Q" C /,
Q for
some constant ˇ to be determined. Denoting by xQ the optimal demand of an

arbitrary informed trader, show that
"Q C Q .N 1/ˇ.1 C 2 /.Q" C /

Q
xQ D 2 2 2 2
:
.1 C /.2 C a z / C a .1 C /.2 C a z / C a2
2 2 2
(10.31)
(ii) By relying on (10.31), show that, for a given value of the constant , the
parameter ˇ corresponding to a linear equilibrium of the economy is given
by
1
ˇD :
.1 C 2 /.N C 1/ C a.2 C 2 z2 .1 C 2 //
(iii) By relying on the equation satisfied in equilibrium by the coefficient ˇ defining

the demand of the informed traders, prove that the coefficient is given by the
solution to the quintic equation

2 2 1 2 N
N .1 C / C 2 z D ;
ˇ ˇ
thus completing the proof of Proposition 10.4.

Exercise 10.6 Consider the model proposed in Subrahmanyam [1575] and pre-
sented in Sect. 10.2. In the setting of Proposition 10.4, suppose that all the N
informed traders are risk neutral, in the sense that a ! 0. Show that, in this case,
the equilibrium value of is given by
s
1 N
D :
NC1 1 C 2
Exercise 10.7 Consider the model developed in Spiegel & Subrahmanyam [1558]
and presented in Sect. 10.2. In this exercise, following Spiegel & Subrahmanyam
[1558, Appendix A], we are going to prove Proposition 10.5.
(i) Assume that each informed trader k 2 f1; : : : ; Kg conjectures that
• every uninformed agent n 2 f1; : : : ; Ng demands a quantity wQ n D eQ n of
the security, for some constant to be determined;
• every other informed trader l 2 f1; : : : ; Kg n fkg demands a quantity xQ l D
ˇ.Q" C Q l / of the security, for some constant ˇ to be determined;
10.8 Exercises 653
• the market maker sets a price pQ according to the rule

!
X
K X
N
pQ D dN C xQ k C wQ n ;
kD1 nD1
for some constant to be determined.

Show that the optimal quantity xQ k of the informed agent k 2 f1; : : : ; Kg,
conditionally on the observation of the private signal "Q C Q k , is given by
1 .K 1/ˇ
xQ k D "2 .Q" C Q k /:
2."2 C 2 /
Deduce that, in equilibrium, the constant ˇ must satisfy
"2
ˇD : (10.32)
..1 C K/"2 C 22 /
(ii) Consider then the expected utility maximization problem of each uninformed
trader. Suppose that each uninformed trader n 2 f1; : : : ; Ng conjectures that
• every informed trader k 2 f1; : : : ; Kg demands a quantity xQ k D ˇ.Q" C Q k /
of the security;
• each other uninformed agent m 2 f1; : : : ; Ngnfng demands a quantity wQ m D
eQ m of the security, for some constant to be determined;
• the market maker sets a price pQ according to the rule
!
X
K X
N
pQ D dN C xQ k C wQ n ;
kD1 nD1
for some constant to be determined.

Show that the optimal quantity wQ n demanded by the risk averse uninformed
agent n 2 f1; : : : ; Ng, conditionally on his endowment eQ n , is given by
a"2 .1 Kˇ/Qen
Q n D
w : (10.33)
2 C a "2 .1 Kˇ/2 C K2 2 ˇ 2 C 2 .N 1/ 2 e2
Deduce that, in equilibrium, the constant must satisfy the equation

a2 3 .N1/e2 C aK2 2 ˇ 2 C a"2 .1 Kˇ/2 C 2 Ca"2 .1Kˇ/ D 0:
(10.34)
(iii) By relying on the zero expected profits condition for the risk neutral market
maker, together with the linear price setting rule, deduce that in equilibrium
the parameters and satisfy the equation
2
K"4 ."2 C 2 / D N 2 2 e2 .1 C K/"2 C 22 (10.35)
and, as a consequence, is given by

q
K"4 ."2 C 2 /
D p :
Ne2 ..1 C K/"2 C 22 /
(iv) On the basis of the previous steps, deduce the result of Proposition 10.5.
Exercise 10.8 Consider the model proposed in Vives [1628] and presented in
Sect. 10.2. In this exercise, we are going to prove the existence of a unique linear
symmetric equilibrium of the economy, characterized as in Proposition 10.6.
(i) As usual, start from the conjecture that there exists a linear equilibrium. A first
consequence of the linearity of the equilibrium, together with the distributional
assumptions of the model, is that the equilibrium price will be normally
distributed. Consider then the expected utility maximization problem of an
arbitrary informed trader i 2 Œ0; 1. Show that, conditionally on the observation
of the realization p of the equilibrium price pQ and of the realization yi of the
private signal yQ i , the demand schedule of the i-th informed trader satisfies
Q yi p
EŒdjp;
Xi .p; yi / D :
Q yi /
a Var.djp;
R1
(ii) For any price p, let L.p/ WD 0 Xi .p; yi /di C zQ be the aggregate demand in
correspondence of the realization yi of the private signal yQ i . Suppose that
Xi .p; yi / D ˛yi C bp C c; for every i 2 Œ0; 1;
for some constants ˛, b and c, for every realization yi . Show that the equilib-
Q
rium price set by the risk neutral market maker via the rule pQ D EŒdjL./ is
given by
pQ D .1 ˛/dN C .˛ dQ C zQ/;
where D ˛z =.˛ 2 z C d /.

(iii) By making use of the results established in the two previous steps of the
exercise, deduce that the optimal demand schedules of the informed traders
are given as in (10.18), so that Proposition 10.6 holds.
10.8 Exercises 655
Exercise 10.9 Let us consider the two-period model of a market dealer economy
originally proposed in Stoll [1570] and presented in Sect. 10.3. Prove Proposi-
tion 10.7.
Exercise 10.10 Consider the model proposed in Glosten & Milgrom [790] and
presented in Sect. 10.3. In this exercise, by relying mainly on the tower property
of the conditional expectation and following the original arguments of Glosten &
Milgrom [790], we are going to prove Proposition 10.8. Suppose that there exist
two processes .pask t /tD1;:::;T0 and .pt /tD1;:::;T0 satisfying the equilibrium condition
bid
(10.25). We shall prove the inequality stated in Proposition 10.8 for the ask price,
the proof of the inequality for the bid price being completely analogous. Let
t 2 f1; : : : ; T0 g and, for simplicity of notation, define the event Ct by
˚
Ct WD fZt > pask
t g D EŒvjK
Q t > pask
t =ıt ;
where the second equality simply follows by the definition of Zt , using the
information flow K D .Kt /tD1;:::;T0 introduced after (10.23). Note that
pask
t D EŒvjD
Q t ; Ct :
For brevity of notation, in the remaining part of the exercise, we shall denote by
Et Œ the conditional expectation EŒjDt ^ Ft , for t D 1; : : : ; T0 .
(i) By relying on the tower property of the conditional expectation, prove that
pask
t D Et ŒvjC
Q t :
(ii) By relying on condition (10.21), show that

h ˇ i
Q t jDt ^ Ft _ .ıt /; Ct ˇCt :
Q t D Et E EŒvjK
Et ŒvjC
(iii) Prove that, for any random variable X with a finite expectation, it holds that
EŒXjX > a EŒX;
for any a 2 R. Moreover, argue that the same relationship also holds for a
conditional expectation of the form EŒXjA ; X > a, where A is an arbitrary
-algebra.
(iv) By relying on the results established in steps (ii)–(iii) of the exercise, show that
h ˇ i
Et ŒvjC Q t j.ıt / ˇCt
Q t Et Et EŒvjK
and deduce that
pask
t Et Œv;
Q
thus completing the proof of the inequality for the ask price stated in
Proposition 10.8.
Exercise 10.11 Consider the simple version of the model of Glosten & Milgrom
[790] described in Sect. 10.3, where the fundamental value vQ of the asset is a random
variable taking the two possible values v and v with probabilities and 1 ,
respectively, and where the probability that an agent is informed is ˛. Suppose
furthermore that an uninformed agent is equally likely to submit a buy or a sell
order. By making use of Bayes’ rule, show that the equilibrium bid and ask prices
set by the dealer by applying the zero expected profits condition are explicitly given
as in (10.28).
Exercise 10.12 Consider the model of Easley & O’Hara [619], where traders are
allowed to post market orders of different sizes, as presented in De Jong & Rindi
[540, Section 4.2]. More specifically, consider the same setting of the simplified
Glosten & Milgrom [790] model presented in Sect. 10.3, characterized by the
following assumptions:
• the fundamental value of the asset is represented by the random variable vQ which
can take the two possible values 1 and 0 with equal probabilities;
• in the economy, there are informed and liquidity traders, with informed traders
knowing the exact realization of v;Q
• there is an exogenous probability ˛ that an order is being submitted by an
informed trader and a probability 1 ˛ that an order is being submitted by a
liquidity trader;
• liquidity traders randomly submit buy and sell orders with equal probability;
• informed traders submit buy or sell orders by maximizing their expected profit
(i.e., knowing the realization of v,
Q they submit a buy order if vQ D 1 and, on the
contrary, a sell order if vQ D 0).
Assume furthermore that buy orders can be of two different sizes B1 and B2 , with
0 < B1 < B2 , and sell orders can be of sizes S1 and S2 , with 0 < S1 < S2 . We
let ˇ=2 and .1 ˇ/=2 be the probabilities that an uninformed trader makes large
(i.e., B2 or S2 ) orders and small (i.e., B1 or S1 ) orders, respectively. For simplicity,
suppose that
B1 D S 1 D 1 and B2 D S2 D 2:
In this model, the dealer is risk neutral and has the possibility of fixing bid and ask
prices for the asset depending on the size of the trade, i.e., the dealer sets pask .1/,
pask .2/, pbid .1/ and pbid .2/, where pask .i/ and pbid .i/ represent respectively the ask
and bid prices for an order of size i, with i 2 f1; 2g.
(i) Suppose that informed traders submit large orders with probability 2
.0; 1/ and small orders with probability 1 . By arguing similarly as in
Exercise 10.11, show that the equilibrium ask prices pask .1/ and pask .2/ are
10.8 Exercises 657
given by
˛.1 / C .1˛/.1ˇ/
2
˛ C .1˛/ˇ
2
pask .1/ D and pask .2/ D :
˛.1 / C .1 ˛/.1 ˇ/ ˛ C .1 ˛/ˇ
(ii) Suppose furthermore that the following condition holds:

2 1 pask .2/ D 1 pask .1/;
meaning that the profit realized by the informed agent when submitting a
large buy order coincides with the profit realized by the informed agent when
submitting a small buy order (of course, this is a necessary condition for the
insider trader to post orders of both sizes). Show that the probability is then
endogenously determined as
ˇ.1 ˇ/
DˇC
˛.1 C ˇ/
and, as a consequence, the equilibrium ask prices pask .1/ and pask .2/ are given
by
1 ˇ C ˛.1 C ˇ/ 3 ˇ C ˛.1 C ˇ/
pask .1/ D and pask .2/ D :
2 4
The ask prices pask .1/ and pask .2/ then corresponds to a pooling equilibrium of
the economy, where informed traders have incentive to submit both small and
large market orders. In this case, the equilibrium bid prices can be computed
in an analogous way. In the remaining part of the exercise, we are going
to characterize the equilibrium ask prices corresponding to a separating
equilibrium of the economy, where informed traders only submit large market
orders.
(iii) Suppose that

2 1 pask .2/ 1 pask .1/;
so that informed traders have incentive to submit only large orders (i.e., in the
above notation, it holds that D 1). Show that in this case the equilibrium ask
prices pask .1/ and pask .2/ are given by
1 .1 ˛/ˇ C ˛
pask .1/ D and pask .2/ D
2 2.1 ˛/ˇ C 2˛
and deduce that a necessary condition for a separating equilibrium to exist is

ˇ ˛=.1 ˛/.
Exercise 10.13 In this exercise, taken from De Jong & Rindi [540, Section 5.2.1]
and related to the models discussed in Sect. 10.3, we present a simple model
of an economy where several risk averse dealers operate. Consider an economy
where a single risky asset with liquidation value dQ is traded. It is assumed that
dQ N .d;
N d2 /. Suppose that the economy is populated by the two following types
of agents:
• M risk averse dealers submitting limit orders to the market, with each dealer
m 2 f1; : : : ; Mg being endowed with Em units of the asset. The utility function of
each dealer is assumed to be negative exponential, with a common risk aversion
parameter a.
• N liquidity traders, submitting market orders. The aggregate demand of liquidity
traders is represented by the random variable zQ, supposed to be normally
distributed with mean zN and variance z2 .
(i) For each m 2 f1; : : : ; Mg, let Xm W R ! R represent the demand schedule of
the m-th dealer, representing the number of units of the asset demanded as a
function of the current market price. Show that
dN pQ
Xm .Qp/ D Em ; for every m D 1; : : : ; M:
ad2
(ii) By imposing market clearing, deduce that the equilibrium price p satisfies

zQ N
pQ D E ad2 C d;
M
P
where E WD M mD1 Em =M, with the consequence that the bid and ask prices
associated to an aggregate quantity z traded by liquidity traders are given by
a 2
pask D dN Ead2 C d jzj; if z > 0;
M
ad2
pbid D dN Ead2 jzj; if z < 0;
M
with a bid-ask spread given by
2ad2
pask pbid D jzj:
M
Observe that the term ad2 =M represents the price impact of a trade: the greater
the dealers’ risk aversion and the volatility of the asset, the greater the price
impact. Increasing the number M of dealers leads to a reduction of the price
impact.
10.8 Exercises 659
(iii) By using the results established in steps (i)–(ii) of the exercise, show that the
quantity traded by each dealer is given by
z
xm D E Em ;
M
where z is the realization of the random variable zQ.
Chapter 11
Solutions of Selected Exercises
In this appendix, we provide the detailed solutions to a selection of the exercises

proposed at the end of the chapters. The solutions to the exercises that are not solved
in this appendix can be found in the solutions manual.
11.1 Exercises of Chap. 2

P
Solution of Exercise 2.1 Let the expected utility function U.Qx/ D SsD1 s u.xs /
represent the preference relation .M ; R/ and let b; c be strictly positive numbers.
Q x/ WD b C cU.Qx/ and uQ .xs / WD b C cu.xs /, for every s D 1; : : : ; S. Then,
DefinePU.Q
since SsD1 s D 1, we have that
X
S X
S
Q x/ D b C cU.Qx/ D b C c
U.Q s u.xs / D s uQ .xs /;
sD1 sD1
thus showing that the function U Q has the expected utility form. Moreover, since
b; c > 0, we have that, for every xQ 1 ; xQ 2 2 M ,
Q x1 / U.Q
xQ 1 R xQ 2 ” U.Qx1 / U.Qx2 / ” U.Q Q x2 /:
Solution of Exercise 2.3 For k D 0, the random variable xQ coincides with its
expectation, so that xQ EŒQx. In this case, for any utility function u W R ! R,
it holds that u EŒQx D u.Qx/ D EŒu.Qx/. Definition 2.2 directly implies that
u .Qx; 0/ D 0. The second claim can be proved by differentiating with respect to
k condition (2.4) which defines the risk premium:
@u .Qx; k/
u0 EŒQx u .Qx; k/ D EŒu0 .Qx/Q ;
@k

662 11 Solutions of Selected Exercises
so that

@u .Qx; k/ E u0 EŒQx C k Q Q
D 0 :
@k u EŒQx u .Qx; k/
Since EŒQ D 0 and u .Qx; 0/ D 0, we get

@u .Qx; k/ ˇˇ u0 EŒQx EŒQ
ˇ D 0 D 0:
@k kD0 u EŒQx
Solution of Exercise 2.7 Let ua .Qx/ and ub .Qx/ denote the risk premia of agents
a and b, respectively, in correspondence of a random variable xQ . Let us consider
the random variable xQ which takes values x0 ˙ with equal probabilities. Then, by
definition of risk premium:
1 1
x0 ua .Qx/ D ua EŒQx ua .Qx/ D E ua .Qx/ D .x0 / C .x0 C a /
2 2
thus showing that ua .Qx/ D .1 a/=2. Analogously, in the case of agent b, we
have that ub .Qx/ D .1 b/=2. Agent b is more risk averse than agent a only if
ub .Qx/ ua .Qx/, i.e., only if b a.
Note that in the context of the present exercise the expected utility of a generic
random variable xQ can be equivalently rewritten as follows, for k D a; b:

EŒuk .Qx/ D E xQ 1fQxx0 g C x0 1fQx>x0 g C k E .Qx x0 /C ;
where, for any event E, the symbol 1E denotes the function (indicator function) that
takes value one if event E occurs and zero otherwise and ./C denotes the positive
part function. Due to the above expression, if a b, then agent a will always accept
a gamble if agent b does, when both agents start from the same initial wealth. Hence,
agent b is more risk averse than agent a.
Solution of Exercise 2.8
(i): The claim follows directly by differentiating the absolute risk aversion
coefficient rua .x/:
d a u000 .x/u0 .x/ .u00 .x//2

ru .x/ D :
dx .u0 .x//2
(ii): An increasing concave function u satisfies pau .x/ > rua .x/, for every x 2 R, if
and only if the following condition holds:
u000 .x/ u00 .x/

< ; for every x 2 R:
u00 .x/ u0 .x/
11.1 Exercises of Chap. 2 663
Equivalently, the above condition holds if and only if u000 .x/u0 .x/ > .u00 .x//2 ,
for every x 2 R. The claim then follows from the proof
of part a).

ru .x/ D rua .x/ rua .x/ pau .x/ .
d a
(iii): The claim follows by noting that dx

Solution of Exercise 2.9 It suffices to note that rua .x/ D dx d
log u0 .x/ , for every
x 2 R.
(i): The claim can be readily verified by differentiating rua .x/ D 1=.a C bx/ and
rur .x/ D x=.a C bx/ with respect to x.
(ii): Let us consider the three different cases: b D 0, b D 1 and b … f0; 1g.
If b D 0, then u00 .x/=u0 .x/ D 1=a and, as a consequence (compare with
Exercise 2.9),
u0 .x/ D kex=a and u.x/ D a kex=a C h; for every x 2 R;
for suitable constants k > 0 and h 2 R. Hence, up to an increasing linear

transformation, we have u.x/ D a exp.x=a/, where the utility function is
defined for every x 2 R.
If b D 1, then u00 .x/=u0 .x/ D 1=.a C x/ and, as a consequence,
u0 .x/ D k.a C x/1 and u.x/ D k log.x C a/ C h for every x > a;

transformation, we have u.x/ D log.xCa/, where the utility function is defined
for x > a.
In the remaining cases (i.e., if b ¤ 0 and b ¤ 1), an analogous reasoning
leads to
k 1
u.x/ D .a C bx/11=b C h; for every x 2 R;
b 1 1b

1 b1
transformation, we have u.x/ D b1 .a C bx/ b . Let us now determine the
domain of the utility function u. If b > 0, which corresponds to assuming
that u is of the DARA type, we must have x > x WD a=b. If b < 0, which
corresponds to assuming that u is of the IARA type, then the function u is
defined for every x 2 R (however, in order to exclude negative risk aversion,
we should restrict the domain to x < x WD a=b).
(i): It holds that rur .x/ D b, for some b 2 R, if and only if rua .x/ D b=x. As shown
in Sect. 2.2, this implies that the coefficient of relative risk aversion is constant
if and only if the utility function u belongs to the HARA class and has the form
u.x/ D ˛x1b C ˇ, with b ¤ 1, for some ˛ > 0 and ˇ 2 R.
(ii): As for part (i), we have rur .x/ D 1 if and only if rua .x/ D 1=x or, equivalently,
if the utility function u belongs to the HARA class and has the form u.x/ D
˛ log.x/ C ˇ for some ˛ > 0 and ˇ 2 R.
Solution of Exercise 2.13 The first part follows directly from the definitions. To
show that the converse implication does not necessarily hold, it suffices to consider
two agents with exponential utility functions: ua .x/ D eax and ub .x/ D ebx
with a > b. In this case, agent a is more risk averse than agent b. Take x1 ; x2 2 R
such that x1 x2 > log alog
ab
b
. Then, it is easy to check that:
00 00 0 0
ua .z/ ua .x1 / a2 a ua .x2 / ua .z/
inf b00 b00 D 2 e.ab/x1 < e.ab/x2 D b0 sup b0 :
z2R u .z/ u .x1 / b b u .x2 / z2R u .z/
This shows that agent a is not strongly more risk averse than agent b.
Solution of Exercise 2.14 We shall always assume that we can interchange
expectation and differentiation. By differentiating the function V with respect to
its first argument we get
@V
.; 2 / D EŒu0 . C zQ/ D EŒu0 .Qx/ 0;
@
since the function u is assumed to be increasing. This shows that V is increasing

with respect to the first argument. Let us now differentiate V with respect to 2 :
@V 1
2
.; 2 / D EŒu0 . C zQ/Qz:
@ 2
Let us denote by ' W R ! R the density function of the standard Normal random
variable zQ. Then
Z C1
0
EŒu . C zQ/Qz D u0 . C z/ z '.z/dz
1
Z C1 Z 0
D u0 . C z/ z '.z/dz C u0 . C z/ z '.z/dz
0 1
Z C1 Z C1
D u0 . C z/ z '.z/dz u0 . z/ z '.z/dz;
0 0
where we have used the symmetry of the density function '. Using the fact that the
function x 7! u0 .x/ is decreasing (since the function u is assumed to be concave),
we obtain
Z C1
0
EŒu0 . C zQ/Qz D u . C z/ u0 . z/ z '.z/dz 0;
0
thus showing that @V.; 2 /=@ 2 0.

Solution of Exercise 2.15 It suffices to compute the ratio (2.14) for the specific
cases of an exponential and a quadratic utility function, respectively, using the
Laplace transform of a normal random variable,
h i
E eaQx Qx
1 da
d
EŒeaQx
EŒeaQx

D D aI
EŒeaQx EŒeaQx
h xQ i h i
E .1 bQx/
b E xQ Qx

b
. 2 C 2 / b 2 b

D D D :
EŒ1 bQx EŒ1 bQx 1 b 1 b
Solution of Exercise 2.16 Let Fi be the distribution function of the random variable
xQ i , for i D 1; 2. Due to Proposition 2.10, in order to show that xQ 1
SSD xQ 2 , we need
to prove that
Z y
G.y/ D F1 .z/ F2 .z/ dz 0; for every y 2 R:
1
As a preliminary, note that, if 2 .Qx1 / 2 .Qx2 /, then F1 .x/ F2 .x/ for every x
and F2 .x/ F1 .x/ for every x , with F1 ./ D F2 ./. Note also that the
function y 7! G.y/ is negative and decreasing for y , reaches its minimum for
y D and is increasing afterwards. Hence, to prove our claim it suffices to show that
limy!C1 G.y/ D 0. Using integration by parts and the fact that EŒQx1 D EŒQx2 D ,
we get
Z C1
lim G.y/ D F1 .z/ F2 .z/ dz
y!C1 1
Z C1
ˇC1
D z F1 .z/ F2 .z/ ˇ1 z '1 .z/ '2 .z/ dz D EŒQx2 EŒQx1 D 0;
1
where 'i W R ! R denotes the density function of the random variable xQ i , for
ˇC1
i D 1; 2, and where z F1 .z/ F2 .z/ ˇ1 D 0 follows from the fact that F1 .1/
F2 .1/ D F1 .1/ F2 .1/ D 0 together with the fact that the tails of a normal
distribution decay exponentially fast. The converse implication can be shown by
applying Definition 2.9 to the concave function x 7! .x /2 .
(i): Due to Proposition 2.8, we have xQ 1
FSD xQ 2 if and only if F1 .x/ F2 .x/
for every x 2 Œ0; 1. In the present case, this condition holds if and only if
2 Œ0:3; 1 and x 1.
R y to Proposition 2.10, we have xQ 1
SSD xQ 2 if and only if EŒQx1 D EŒQx2 and
(ii): Due
0 F1 .z/ F2 .z/ dz 0 for all y 2 Œ0; 1. In the present context, the first
condition amounts to x.1 / D 0:7. For the second one, note that:
8
ˆ
ˆ0:3 ; for y 2 Œ0; x/I
<
F1 .y/ F2 .y/ D 0:3 1; for y 2 Œx; 1/I
ˆ
:̂0; for y D 1:
Ry
Hence, the condition 0 F1 .z/ F2 .z/ dz 0, for all y 2 Œ0; 1, holds if and
only if 0:3.
2.12, we have xQ 1
SSD xQ 2 if and only if EŒQx1 EŒQx2 and
M
(iii): RDue to Proposition
y
0 F1 .z/ F2 .z/ dz 0 for all y 2 Œ0; 1. In the present context, analogously
as in part b), these conditions hold if and only if x.1 / 0:7 and 0:3 .
(iv): We have xQ 1
MV xQ 2 if and only if EŒQx2 D .1 /x 0:7 D EŒQx1 and
2 .Qx2 / D x2 .1 / 0:21 D 2 .Qx1 /.
Ry
Solution of Exercise 2.18 It is easy to show that 1 F1 .z/ F2 .z/ dz 0, for all
y 2 Œ1; 1, if and only if . The claim then follows by Proposition 2.10.
Solution of Exercise 2.19 The first implication follows directly from Proposi-
tion 2.8. For a counterexample, it suffices to take any two non-negative random
variables xQ 1 and xQ 2 with EŒQx1 EŒQx2 and EŒQx12 EŒQx22 . Since the function u W
RC ! RC defined by u.x/ WD x2 is non-decreasing, we have EŒu.Qx1 / EŒu.Qx2 /,
thus showing that xQ 1
FSD xQ 2 does not hold.
Solution of Exercise 2.20 The answer to both questions is negative. Indeed, it
suffices to consider two random variables xQ 1 and xQ 2 such that xQ 1
M SSD x Q 2 but
2 .Qx1 / > 2 .Qx2 /. For instance, let xQ 1 be an arbitrary random variable and
Q a non-
negative random variable such that Cov.Qx1 ;
/ Q > 2 .
/=2
Q Q Then,
and let xQ 2 WD xQ 1
.
due to Proposition 2.12, we have xQ 1
SSD xQ 2 but
M
Q C 2 .
/
2 .Qx2 / D 2 .Qx1 / 2 Cov.Qx1 ;
/ Q < 2 .Qx1 /:
Solution of Exercise 2.21 Let us consider the first order conditions for the solution
of the minimization problem (2.20):
@L.w1 ; : : : ; wN ; /
D 2 wn 2 .Qxn / D 0; for every n D 1; : : : ; NI
@wn
@L.w1 ; : : : ; wN ; / XN
D wn 1 D 0:
@ nD1
Since 2 .Qxn / > 0, for every n D 1; : : : ; N, the above first order conditions are both
necessary and sufficient for the optimal solution to the problem. Hence:

wn D ; for every n D 1; : : : ; N:
2 2 .Qx n/
PN
Together with the condition nD1 wn D 1, this leads to:
2
D PN ;
nD1 1= 2 .Qxn /
which immediately implies that the optimal solution wn is given by expres-
sion (2.21), for every n D 1; : : : ; N.
Solution of Exercise 2.22 Let define the random variables
xQ n1 D f0; nI 1 1=n2 ; 1=n2 g; for all n 2 N;
and let xQ 2 be the trivial random variable which always takes value zero. It is easy
to show that EŒQxn1 D 1=n and 2 .Qxn1 / D 1 1=n2 . For n ! 1, it holds that
EŒQxn1 ! 0, while 2 .Qxn1 / ! 1. Hence, since the function V is assumed to be
continuous, increasing with respect to the mean and decreasing with respect to the
variance, we have that, for sufficiently large n,

EŒu.Qxn1 / D V EŒQxn1 ; 2 .Qxn1 / < V.0; 0/ D EŒu.Qx2 /:
However, it is clear that P.Qxn1 xQ 2 / D 1 and P.Qxn1 > xQ 2 / > 0, since the random
variable xQ n1 satisfies xQ n1 xQ 2 and P.Qxn1 D n > 0/ D 1=n2 > 0.
Solution of Exercise 2.23 The claim easily follows by considering the random
variables xQ n1 and xQ 2 introduced in the previous exercise. Indeed, if Assumption 2.3
was satisfied, then it would imply that the sure payoff of zero is preferred to the sure
payoff of n, thus yielding a clear contradiction with the strictly increasing feature of
the mean-variance utility function V.
Solution of Exercise 3.1 By the first order optimality condition, for an agent to
invest all his initial wealth w0 (or more) in the risky asset, it must be that
EŒu0 .w0 rQ /.Qr rf / 0:
Since the utility function u is supposed to be twice differentiable, we can apply a

Taylor approximation of the marginal utility function u0 up to the first order:
u0 .w0 rQ / u0 .w0 rf / C u00 .w0 rf /w0 .Qr rf /:
By combining the two above expressions, the claim follows directly.

Solution of Exercise 3.3 We already know that, by the concavity of the utility
e solving problem (3.2) is unique. Arguing by
function u, the optimal wealth W
contradiction, suppose that there exist two distinct portfolios w1 ; w2 2 RN such that
X
N X
N
e D w0 rf C
W w1n .Qrn rf / D w0 rf C w2n .Qrn rf /:
nD1 nD1
PN 1
PN
If nD1 wn ¤ nD1 w2n , then the last relation implies that
PN
.w1n w2n /Qrn
rf D PnD1
N
;
1 2
nD1 .wn wn /
thus
Pcontradicting
P the assumption that there are no redundant assets in the economy.
If NnD1 w1n D NnD1 w2n , then the existence of two distinct portfolios w1 ; w2 allows
to obtain an analogous contradiction.
Solution of Exercise 3.4 Let w 2 RN denote the optimal portfolio, with
P N
kD1 wk D 1, for some given initial wealth w0 . Then, for n 2 f1; : : : ; Ng, consider
the following auxiliary maximization problem:
2 0 13
6 B X
N X
N
C7
max E 6 B
4u @ wk rQk C Qrn C .wn / k rQk C7
A5 : (11.1)
2R
kD1 kD1
k¤n k¤n
Due to the optimality of w 2 RN , problem (11.1) is solved by D wn . Whether

wn > 0 holds depends on the sign of the derivative of the function to be optimized
in problem (11.1) with respect to , evaluated at D 0. Such a derivative can be
computed as
2 0 10 13
6 0B X
N
CB X
N
C7
E6 B
4u @ .wk C wn k /Qrk C B
A @rQn k rQk C7
A5
kD1 kD1
k¤n k¤n
2 0 1 3 2 0 13
6 0 BX
N
C 7 6 BX N
C7
D E6 u
4 @
B .w C w k /Qrk C Qn 7 D E 6u0 B .w C w k /Qrk C7 EŒQ n ;
k n A 5 4 @ k n A5
kD1 kD1
k¤n k¤n
where the last equality follows by iterated conditioning and (3.55), since condi-
tion (3.55) implies that
EŒ f .Qr1 ; : : : ; rQn1 ; rQnC1 ; : : : ; rQN /Q n D EŒ f .Qr1 ; : : : ; rQn1 ; rQnC1 ; : : : ; rQN /EŒQ n

for any integrable function f W RN1 ! R. Since the utility function u is assumed
to be strictly increasing, we see that the sign of the derivative is determined by the
expected value EŒQ n , thus proving the claim.
Solution of Exercise 3.5 The claim follows from Exercise 3.4, by writing rQn D rf C
Qn and noting that, under the present assumptions, the random variable Qn satisfies
condition (3.55), for all n D 1; : : : ; N.
Solution of Exercise 3.7 The reasoning is analogous to that used in the proof of
e w0 rf
Proposition 3.3. Indeed, since w .w0 / > 0 for all w0 2 RC , it holds that W
e w0 rf on the event fQr rf g. Hence, since the
on the event fQr rf g, while W
mapping x 7! rr .x/ is decreasing, we have that
00 e 00
e u .W / D rr .W
W e / rr .w0 rf / D w0 rf u .w0 rf / ; on fQr rf g;
e /
u 0 .W u0 .w0 rf /
with the converse inequality holding on the event fQr rf g. Hence:

00
EWe u .W
e /.Qr rf / rr .w0 rf /E u0 .W
e /.Qr rf / D 0;
where the last equality follows from the optimality condition (3.4).
Solution of Exercise 3.10 Note that, as a consequence of Theorem 3.12, the two
portfolios w1 and w2 admit the representation
B.V 1 1/ A.V 1 e/ C.V 1 e/ A.V 1 1/

wi D C EŒQrwi ; for i D 1; 2:
D D
As a consequence, it holds that
Cov.Qrw1 ; rQw2 / D w1> Vw2

>
B.V 1 1/ A.V 1 e/ C.V 1 e/ A.V 1 1/
D C EŒQrw1 V
D D

B.V 1 1/ A.V 1 e/ C.V 1 e/ A.V 1 1/
C EŒQrw2
D D
>
B.V 1 1/ A.V 1 e/ C.V 1 e/ A.V 1 1/
D C EŒQrw1
D D

B1 Ae Ce A1
C EŒQrw2 :
D D
Developing the product, using the relation D D BC A2 and simplifying,

relation (3.24) follows.
Solution of Exercise 3.11 Since the minimum variance portfolio wMVP belongs
to the portfolio frontier, it suffices to apply formula (3.24) for w1 D wMVP and
w2 D w . Indeed, recalling that EŒQrwMVP D A=C, the first term on the right-hand
side of formula (3.24) becomes null, thus proving the claim.
Solution of Exercise 3.12 As can be easily verified from formula (3.24), if there
exists a portfolio wzc such that Cov.Qrw ; rQwzc / D 0, then its expected return must be
necessarily given by the expression on the right-hand side of (3.30). Since, due
to Theorem 3.12 there exists a unique frontier portfolio for any given expected
return 2 R, the zero correlation portfolio exists and is given by the unique
frontier portfolio with expected return given by the expression on the right-hand
side of (3.30).
Solution of Exercise 3.13 Given any frontier portfolio (other than the minimum
variance portfolio wMVP ), the expected return of its zero correlation portfolio is given
by expression (3.30). Hence, due to (3.25), the variance of the return of the zero
correlation portfolio wzc with respect to a frontier portfolio w is given by
2
C A 2 1 C D=C2 1
2 .Qrwzc / D EŒQrwzc C D C :
D C C D EŒQrw A=C C
Hence, the frontier portfolio w satisfies 2 .Qrw / D 2 .Qrwzc / if and only if the
following equality holds, due to (3.25):
2
C A 2 1 C D=C2 1
EŒQrw C D C :
D C C D EŒQrw A=C C
p
The solution to the above equation is given by EŒQrw D A=CC D=C. Then,p due to
Theorem 3.12, the unique frontier portfolio w satisfying EŒQrw D A=C C D=C
is given by
p !
A D
w DgCh C ;
C C
where g and h are defined as in Theorem 3.12.

Solution of Exercise 3.14 Recall that the minimum variance portfolio wMVP is
represented in the variance-expected return plane by the point .1=C; A=C/, while,

due to expression (3.25),
2
the frontier
portfolio w is represented by the 2point
C=D.EŒQrw A=C/ C1=C; EŒQrw . In the variance-expected return plane . ; /,
the line connecting the two points satisfies

A EŒQrw A=C 2 1
D C :
C .EŒQrw A=C/2 C=D C
Hence, the intersection of the above line with the vertical axis is found by letting
2 D 0, yielding
A 1 EŒQrw A=C A D=C2

D 2
D D EŒQrwzc :
C C .EŒQrw A=C/ C=D C EŒQrw A=C
In the standard deviation - expected return plane .; /, by taking the total
differential of
(3.25), we obtain that the slope of the tangent to the portfolio frontier
at the point 2 .Qrw /; EŒQrw is given by
d .Qrw /D
D :
d CEŒQrw A
Hence, the intercept of the tangent with the vertical axis is given by
d A D=C2
EŒQrw .Qrw / D D EŒQrwzc ;
d C EŒQrw A=C
where the first equality follows from (3.25) and the second from (3.30).
Solution of Exercise 3.15 The portfolio wq is given as the solution to the following
problem:
1 >
min w Vw; subject to w> Vwp D 0 and w> 1 D 1:
w2RN 2
The above problem can be solved by optimizing the following Lagrangian:
1 >
L.w; ; / D w Vw w> Vwp .w> 1 1/:
2
The first order conditions give
@L.w ; ; /
D Vw Vwp 1 D 0;
@w
so that
w D wp C V 1 1;
where the optimal values for the Lagrange multipliers are given by
1 2 .Qrwp /
D and D ;
1 C 2 .Qrwp / 1 C 2 .Qrwp /
using the notation introduced in Theorem 3.12. Hence, the optimal portfolio wq
which solves the above problem can be represented as
1 2 .Qrwp /
wq D 2
wp V 1 1
1 C .Qrwp / 1 C 2 .Qrwp /
1 C 2 .Qrwp /
D 2
wp wMVP ;
1 C .Qrwp / 1 C 2 .Qrwp /
recalling that the minimum variance portfolio wMVP is given by wMVP D V 1 1=C.
The expected return EŒQrwq can be easily computed as
1 C 2 .Qrwp / A EŒQrwp A 2 .Qrwp /

EŒQrwq D EŒQ
r w p D :
1 C 2 .Qrwp / 1 C 2 .Qrwp / C 1 C 2 .Qrwp /
As can be easily checked (compare Exercise 3.14), the value EŒQrwq coincides
with
2 the intersection with the vertical axis of the line connecting the two points
.Qrwp /; EŒQrwp and 2 .QrwMVP /; EŒQrwMVP in the expected return - variance plane.
Solution of Exercise 3.17 The claim follows directly from formula (3.27) by noting
that, if EŒQrwq D EŒQrwp , then it holds that
Cov.Qrwq ; rQwp / D Cov.Qrwp ; rQwp / D 2 .Qrwp /:
Solution of Exercise 3.18 Let us consider the frontier portfolio w 2 PF with
expected return WD EŒQrw , given by (see Proposition 3.14)
rf
w D V 1 .e rf 1/ ;
K
where K D B 2Arf C Crf2 > 0. Consider an arbitrary linear combination (with

weights ˛ and 1 ˛) of the tangent portfolio we and the risk free asset, respectively,
i.e., consider the portfolio w.˛/ WD ˛we C .1 ˛/w0 , with w0 WD 0 2 RN . Due to
Proposition 3.15, such a portfolio is characterized by
e rf 1
w.˛/ D ˛we D ˛V 1 : (11.2)
1> V 1 .e rf 1/
Due to equation (3.39), the expected return associated to the portfolio w.˛/ is given
by
Arf B
EŒQrw.˛/ D .1 ˛/rf C ˛ :
Crf A
Hence, by choosing ˛ D . rf /.A Crf /=K, we have that EŒQrw.˛/ D EŒQrw . As

can be easily verified, if we replace this value of ˛ into equation (11.2), we recover
the portfolio w , thus proving the claim.
Solution of Exercise 3.19 Fix an arbitrary value ˛ > and consider the following
auxiliary problem:
w> e
max p over all w 2 N such that w> e D ˛: (11.3)
w> Vw
Clearly, the solution to this auxiliary optimization problem corresponds to the

solution to Problem (3.14) with respect to the fixed expected return ˛, with optimal
solution
w .˛/ D g C h˛;
where g and h are defined as in Theorem 3.12. In view of Property 2 of the portfolio
frontier (see formula (3.25)), the optimal value of problem (11.3) is then given by
˛
q ;
C˛ 2 2A˛CB
D
Optimizing over ˛ > the last expression gives the optimal value
A B
˛ D ;
C A
so that the optimal solution to the original problem (3.56) is given by
A B
w D w .˛ / D g C h
C A
1 V 1 1 1 V 1 e
D .BC AB A2 C AB/ C .AC BC AC C A2 /
D C A D C A
V 1 .1 e/ V 1 .e 1/
D D > 1 ;
C A 1 V .e 1/
thus proving the claim.
Solution of Exercise 3.20 By Proposition 3.15, the tangent portfolio we is given by
e rf 1
we D V 1 :
1> V 1 .e
rf 1/
If the risky asset returns are uncorrelated then the matrix V 1 is diagonal, with
1 N
diagonal elements .Vn;n /nD1 . Hence, it holds that, for all n D 1; : : : ; N,
1
Vn;n .en rf /
wen D PN 1
:
iD1 Vi;i .ei rf /
Since en > rf for all n D 1; : : : ; N, this directly implies that the tangent portfolio is
diversified, i.e., wen 2 .0; 1/ for all n D 1; : : : ; N.
Solution of Exercise 3.24 If the portfolio w satisfies the one fund separation
property, then the random variable rQw dominates (in the sense of second order
stochastic dominance) the random return rQw of any portfolio w 2 N . Hence, due
to Proposition 2.10, it holds that EŒQrw D EŒQrw for all w 2 N . In particular, this
implies that all portfolios have the same expected return. Moreover, as discussed
after Proposition 2.10, we also have 2 .Qrw / 2 .Qrw / for any portfolio w 2 N ,
thus showing that w is the minimum variance portfolio.
Solution of Exercise 3.25 Define the random variables
1 X
N
rQ WD .Qrn rf / and Qn WD rQn rQ rf ; for all n D 1; : : : ; N:
N nD1
P
Let ai D 1=N for all i D 1; : : : ; N. Then, it is easy to check that NiD1 ai Qi D 0
PN
and iD1 ai D 1. Moreover, since the returns .Qr1 ; : : : ; rQN / are i.i.d., the conditional
expectation EŒQ n jQr does not depend on n and, hence,
" #
1X
N
1 X
N ˇ
ˇ
EŒQ 1 jQr D EŒQ n jQr D E rQn rQ ˇQr rf D 0:
N nD1 N nD1
Since the random vector rQ can be equivalently written as rQ D rf 1 C bQr C Q ,

where Q D .Q 1 ; : : : ; QN /, Proposition 3.18 then implies that the two fund monetary
separation property holds.
Solution of Exercise 3.26 Note first that, for any portfolio w 2 N , we can write
rQw D rQwMVP C Qw , for some random variable Qw with EŒQ w D 0. Moreover, due to
2
equation
(3.28),
it holds that Cov Q
r ;
w wQ
r MVP D .Q
r w MVP /. In turn, this implies that
Cov Qw ; rQwMVP D 0. Hence, due to the normal distribution of rQw , rQwMVP and Qw and
since uncorrelated normal random variables are independent, we have that
EŒQ w jQrwMVP D EŒQ w D 0:
Then claim then follows by the same arguments used in the proof of Proposi-
tion 3.20.
Solution of Exercise 3.28 The first order condition (3.46) for the optimal consump-
tion problem gives
1
1
p1 1
x1 D x2 :
p2
If p1 =p2 > =.1 /, this directly implies that x1 < x2 .
Solution of Exercise 3.29 From the proof of Proposition 3.24, let us consider the
first order condition evaluated at w D 0, using the basic properties of the covariance:

E u0 .w0 xQ / xQ .1 C /EŒQx D Cov xQ ; u0 .w0 xQ / EŒu0 .w0 xQ /EŒQx:
Hence, the first order condition will be negative for all values > . In
correspondence of such values, the optimal insurance demand will be w =0.
Solution of Exercise 3.30 The insurance problem can be formulated as follows:

1 h i
max E exp a w0 .1 w/Qx w.1 C /EŒQx ;
w2RC a
or, equivalently, using the moment generating function of the normal distribution,
a
max w0 .1 w/ w.1 C / .1 w/2 2
w2RC 2
and, hence, the optimal value w is given by

w D 1 2 _ 0:
a
Hence, under the assumption that > 0, it immediately follows that w < 1 if (and
only if) > 0.
Solution of Exercise 3.31 Consider an alternative insurance contract with the same
price p and with a different indemnity function I./. N Since EŒI .Qx/ D p=.1 C / D
N N
EŒI.Qx/, the expected value of I.Qx/ must be equal to the expected value of I .Qx/ D
maxf0I xQ Kg. Hence, if the indemnity is increased for the loss xj , for some j 2
f1; : : : ; Ng, then it must be decreased for the losses xi , with i ¤ j. So, consider an
increase j > 0 of the indemnity in correspondence of the loss level xj , for some
j 2 f1; : : : ; Ng, and a decrease i 0 of the indemnities Pin correspondence to the
loss levels xi , for all i 2 f1; : : : ; Ng with i ¤ j, so that i¤j i i D j j , in order
N x/. Note that, since the indemnity cannot be reduced
to ensure that EŒI .Qx/ D EŒI.Q
to negative values, we must decrease the indemnity only for the loss levels xi > K.
N x/ can be described as follows:

Summing up, the indemnity I.Q
N j / D maxf0I xj Kg C j ;
I.x
N i / D xi K i ;
I.x for all i ¤ j such that xi K C i :
Let denote by w1 .Qx/ the net wealth at time t D 1 associated to the insurance contract
with indemnity I .Qx/ and price p and, analogously, denote by wN 1 .Qx/ the net wealth
at time t D 1 associated to the insurance contract with indemnity I.Q N x/ and price p.
N
Note that, due to the definition of the new indemnity I.Qx/, it holds that
wN 1 .xj / D w1 .xj / C j ;

(11.4)
wN 1 .xi / D w1 .xi / i 1fxi KC i g ; for all i ¤ j:
Note that, if i ¤ j and xi K C i , we have that
w1 .xi / D w0 p xi C I .xi / D w0 p K

(11.5)
w0 p minfxj I Kg D w0 p xj C I .xj / D w1 .xj /:
From equations (11.4)–(11.5), we can observe that the indemnity scheme I.Q N x/
decreases the wealth at time t D 1 for small levels of wealth, while it increases
the wealth at time t D 1 for large levels of wealth. This implies that the original
N x/ according to the second order stochastic
indemnity scheme I .Qx/ dominates I.Q
dominance criterion, thus establishing the result.
Solution of Exercise 3.32 Since the background risk yQ is assumed to be independent
of the return rQ of the risky asset, the portfolio optimization problem (3.50) can be
rewritten as follows, using the tower property of (conditional) expectations:

max E u w0 rf C yQ C w.Qr rf /
w2R
h i
D max E E u w0 rf C yQ C w.Qr rf / jQr
w2R

D max E v w0 rf C w.Qr rf /
w2R
where the utility function v./ is defined by v.x/ WD EŒu.x C yQ /, for all x 2 R. At
this stage, in view of Proposition 3.2, in order to show that w < w , it suffices to
show that the function v corresponds to a greater risk aversion in comparison with
the utility function u. In other words, we must check whether
v 00 .x/ EŒu00 .x C yQ / u00 .x/

rva .x/ D 0
D 0
0 D rua .x/: (11.6)
v .x/ EŒu .x C yQ / u .x/
It holds that
EŒu00 .x C yQ / D EŒrua .x C yQ /u0 .x C yQ /

D Cov rua .x C yQ /; u0 .x C yQ / C EŒrua .x C yQ /EŒu0 .x C yQ /
EŒrua .x C yQ /EŒu0 .x C yQ /

rua EŒx C yQ EŒu0 .x C yQ / D rua .x/EŒu0 .x C yQ /;
where the first two inequalities use the assumption that u is concave and that the
mapping x 7! rua .x/ is decreasing and convex, respectively, together with Jensen’s
inequality (for the last inequality). In particular, the fact that the covariance between
rua .x C yQ / and u0 .x C yQ / is positive follows from the monotonicity of the functions rua
and u0 (see, e.g., Thorisson [1589]). This implies that condition (11.6) is satisfied.
Finally, note that, if the mapping x 7! rua .x/ is decreasing, then it holds that
drua .x/=dx 0, i.e.,
2
drua .x/ u000 .x/u0 .x/ u00 .x/
D 2 0;
dx u0 .x/
which can be easily seen to imply that pau .x/ rua .x/, for all x 2 R.
Solution of Exercise 3.33 :Since the utility function u is assumed to be strictly
concave, the denominator of (3.53) is always strictly negative. Hence, in order to
study the behavior of s with respect to changes in the risk free rate rf , it suffices to
study the sign of the numerator of (3.53), which can be rewritten as follows:
u00 .w1 C s rf /s rf u0 .w1 C s rf /

00
0 u .w1 C s rf /
D u .w1 C s rf / 0 .w1 C s rf / 1 C u00 .w1 C s rf /w1

u .w1 C s rf /

D u0 .w1 C s rf / rur .w1 C s rf / 1 C u00 .w1 C s rf /w1 ;
where rur denotes the coefficient of relative risk aversion associated to the utility
function u. By the strict concavity of u, it is clear that the above quantity is negative
if rur .w1 C s rf / < 1.
Solution of Exercise 3.34 Recall that the coefficient of relative risk aversion is
given by rur .x/ D xu00 .x/=u0 .x/. By differentiating, we get

drur .x/ u00 .x/ u000 .x/ u00 .x/
D 0 1 C x 00 x 0 D rua .x/ 1 xpu .x/ C rur .x/ :
dx u .x/ u .x/ u .x/
If the mapping x 7! rur .x/ is decreasing, then the above expression is negative.
Hence, if rur .x/ > 1 for all x 2 R, this implies that xpu .x/ > 2, for all x 2 R, thus
proving the claim.
Solution of Exercise 4.1 From the analysis of Sect. 4.1, we know that an ex-ante
Pareto optimal allocation fxa
1 ; x2 ; x1 ; x2 g satisfies
a b b
0 0
ua .xa
1 / ub .xb
1 /
0 D ;
.1 /u .x2 /
a a
.1 /u 0 .xb
b
2 /
0 0 0 0
so that ua .xa
1 /=u .x2 / D u .x1 /=u .x2 / DW k. It holds that k < 1. Indeed,
a a b b b b
0 a0 a
suppose on the contrary that k 1. In that case, one would have ua .xa
1 / u .x2 /
0 0
and ub .xb
1 / u .x2 /, which in turns implies that x1 x2 and x1 x2 , due
b b a a b b
the concavity of the utility functions. But then:
ea1 C eb1 D xa

1 C x1 x2 C x2 D e2 C e2 ;
b a b a b
thus yielding a contradiction with the hypothesis of the exercise. Hence, if . p1 ; p2 /

are the prices of the contingent goods in correspondence of a Pareto optimal
allocation it holds that
0
p1 ua .xa
1 /
D 0 a D k< ;
p2 .1 /u .x2 /
a 1 1

Solution of Exercise 4.2 If condition (4.4) holds, then
0 j j0 j
ui .xi
s /D u .xs /; for all i; j D 1; : : : ; I and s D 1; : : : ; S:
i
Hence, for every i; j D 1; : : : ; I and s; r D 1; : : : ; S, it holds that

0 0 0
s ui .xi
s / s j i u j .xsj / s u j .xsj /
D D ;
r ui0 .xi
r / r i j u j0 .xrj / r u j0 .xr /
j
thus proving that condition (4.1) holds. Conversely, if condition (4.1) holds, then
condition (4.4) can be obtained by taking
1
i WD ; for all i D 1; : : : ; I;
ui0 .xi
1 /
so that
s 0 s 0
i ui .xi
s /D j u j .xsj /; for all i; j D 1; : : : ; I and s D 1; : : : ; S;
1 1
thus proving that the Borch condition (4.4) holds.

y .e/ 2 i
Solution of Exercise 4.3 The sharing rule is linear if and only if d de 2 D 0 for all
i D 1; : : : ; I. Hence, in view of Condition (4.11), this means that the sharing rule is
linear if and only if
dyi .e/ X
I
d X
I

tu0 i yi .e/ tu j y j .e/ D tui yi .e/ tu j y j .e/ ;
de jD1 de jD1
for all i D 1; : : : ; I. Substituting condition (4.11) in the above expression and

simplifying, we get
d X j
I
tu0 i yi .e/ D tu j y .e/ ;
de jD1
for all i D 1; : : : ; I. Since the right hand side of the last expression does not depend
on i, the claim is proved.
Solution of Exercise 4.4 We follow Huang & Litzenberger [971, Section 5.14].
Suppose that the sharing rule fyi W RC ! RC I i D 1; : : : ; Ig is linear in
correspondence of every possible Pareto optimal allocation. Note first that, since
Pareto optimal allocations are parameterized by the weights a D .a1 ; : : : ; aI /, the
sharing rule also depends on the weights a, so that we get the representation
yi .e/ D ˛i .a/ C ˇi .a/e; for all i D 1; : : : ; I;
for some constants ˛i .a/ and ˇi .a/ depending on the weights a, for every possible
realization e of the aggregate endowment. In the present context, condition (4.6)
implies that
0 0
ai ui yi .e/ D a j u j y j .e/ ; (11.7)
for all i; j D 1; : : : ; I. By differentiating (11.7) with respect to e we obtain
00 dyi .e/ 00 dy j .e/

ai ui yi .e/ D a j u j y j .e/ ;
de de
for all i; j D 1; : : : ; I. Moreover, by differentiating (11.7) with respect to ai and using
the linear form of the sharing rule, we get
0 00 00
ui yi .e/ C ai ui yi .e/ ˛ii .a/ C ˇii .a/e D a j u j y j .e/ ˛ji .a/ C ˇji .a/e ;
d˛ .a/
for all i; j D 1; : : : ; I, where ˛ii .a/ WD d˛dai .a/
i and ˛ji .a/ WD daj i and similarly for
ˇii and ˇji . Combining the last two expressions and making use of the linear form
of the sharing rule, we get
0
ui yi .e/
i00 i D Ai C Bi yi .e/; (11.8)
u y .e/
for all i D 1; : : : ; I, where Ai and Bi are constants. Now, as follows from

Exercise 4.3, the linearity of the sharing rule implies that, for every agent i D
1; : : : ; I, the first derivative of the risk tolerance of the utility function ui computed
in correspondence of the Pareto optimal allocation yi .e/ does not depend on i. This
implies that Bi D Bj for all i; j D 1; : : : ; I. Since we assumed that the sharing
rule is linear for all possible Pareto optimal allocations and since the latter are
parameterized by the weights a D .a1 ; : : : ; aI /, equation (11.8) must hold for all
possible weights a, thus showing that condition (4.15) has to be satisfied.
Solution of Exercise 4.5 In the case of generalized power utility functions,
condition (4.4) leads to
b
i ı i
.i C bxi
s / D j C bxs ;
j
for all i; j D 1; : : : ; I and s D 1; : : : ; S:
j ı j
Summing over j, we obtain

PI
bes C jD1 j i
xi
s D PI ; for all i D 1; : : : ; I and s D 1; : : : ; S;
b.i ıi / b
jD1 .j ıj /
b b
thus proving the linearity of the Pareto optimal sharing rule.

In the case of exponential utility functions, condition (4.4) leads to
i ıi exp.xi
s =i / D j ıj exp.xs =j /;
j
for all i; j D 1; : : : ; I and s D 1; : : : ; S:
Taking logarithms and summing over j, we obtain, for all i D 1; : : : ; I and s D

1; : : : ; S,
X
I
i
xi
s D i log.i ıi / C PI es j log.j ıj / :
jD1 j jD1
Solution of Exercise 4.6 Let the price-allocation couple .q I x1 ; : : : ; xI / corre-
spond to an Arrow-Debreu equilibrium. Suppose that at time t D 1 the state of the
world !s is realized, for some s D 1 : : : ; S, and suppose that (spot) markets are open
at time t D 1, so that agents have the possibility to trade at t D 1 the L consumption
goods at prices qs 2 RLCC . By definition of Arrow-Debreu equilibrium, for every
i D 1; : : : ; I, the allocation xi solves Problem (PO1) for agent i, subject to the
s 2 RCC maximizes
L
feasibility constraint (4.16). In particular, this implies that xi
the function x 7! U .x1 ; : : : ; xs1 ; x; xsC1 ; : : : ; xS / subject to the constraint
i i i i i
X
L X
L
qls .x eils / qls .xi
ls els /:
i
lD1 lD1
This means that agent i has no incentive to trade at time t D 1 in correspondence

of the equilibrium prices q . Since i D 1; : : : ; I and s D 1; : : : ; S are arbitrary, the
claim is proved.
Solution of Exercise 4.7 We proceed as in the proof of Proposition 4.15. The first
order conditions of (4.29) yield, for all s D 0; 1; : : : ; S and i D 1; : : : ; I,
X
I
s =i / D s
ai exp.xi and s D es :
xi
iD1
s /,
Hence, solving for exp.xi
i i i
s / D s .a /
exp.xi ;
which, by multiplying over all i, gives that, for all s D 0; 1; : : : ; S,
PI Y
I
iD1 i
exp.es / D s .ai /i ;
iD1
and, solving for s ,

!
es Y
I
PI i
s D exp PI .ai / iD1 i :
iD1 i iD1
Hence, the utility function u of the representative agent is defined by

!
e0 Y
I
PI i X
I
u0 .e0 / D exp PI .ai / iD1 i i ;
iD1 i iD1 iD1
and similarly for u1 .es /, for all s D 1; : : : ; S. Hence, by condition (4.32), the
equilibrium price vector q of the S Arrow securities is given by
P
ıs exp.es = IiD1 i /
qs D P ; for all s D 1; : : : ; S;
exp.e0 = IiD1 i /
thus proving the claim. Note that, as in the case of Proposition 4.15, the equilibrium
price vector q 2 RS does not depend on the resource allocation but only on the
aggregate endowment of the economy.
Solution of Exercise 4.10 For i D a; b, let us denote by .xi1 ; x2 / the optimal
i
consumption plan of agent i. If the two agents have homogeneous beliefs, then the
optimality conditions give
xi
2 p1
D ; for i D a; b;
.1 /x1i p2
p1 1 i
2 D p2 x1 , for i D a; b. For each agent i D a; b, the budget
thus implying that xi
constraint then implies that
p1 i
˛ i . p1 e1 C p2 e2 / D p1 xi
1 C p 2 x2 D
i
x :
1
1 =x1 D ˛ =˛ and an analogous reasoning

In turn, the last condition implies that xa b a b
allows to show that x2 =x2 D ˛ =˛ . Since es =es D ˛ a =˛ b , for s D 1; 2, and the

a b a b a b
s C xs D es C es , for s D 1; 2, it easily follows

feasibility constraint implies that xa b a b
that xj D ej , for i D a; b and j D 1; 2, thus showing that the optimal choice of the
i i
two agents is simply given by consuming their initial endowments.

Solution of Exercise 4.13 Under the present assumptions, the optimal consumption
Problem (PO3) is given by
XS
1 b1 b1
max . C bx0 / b Cı s . C bxs / b ;
.x0 ;x1 ;:::;xS /2RSC1
C
b1 sD1
subject to the budget constraint (note that the budget constraint can be expressed as
an equality since the utility function is strictly increasing)
X
S X
S
x0 C p s xs D e0 C ps es D eN :
sD1 sD1
Denoting by .x0 ; x1 ; : : : ; xS / the optimal consumption plan for Problem (PO3), the
optimality conditions give
. C bx0 /1=b D and ıs . C bxs /1=b D ps ;
for all s D 1; : : : ; S, so that

1=b
C bxs
ıs D ps ; for all s D 1; : : : ; S:
C bx0
Solving for C bxs and using the budget constraint, we get
b X
S !
ıs
C bxs D C b eN
p r xr ;
ps rD1
so that, multiplying by ps and summing over all s, we obtain
X
S X
S X
S X
S X
S X
S
ps Cb ps xs D ı b sb ps1b CbNeı b sb ps1b bı b ps xs sb ps1b :
sD1 sD1 sD1 sD1 sD1 sD1
In turn, the last equation implies that

P P P
X
S
ı b SsD1 sb ps1b SsD1 ps ı b SsD1 sb ps1b
eN x0 D ps xs D P C P eN ;
sD1
b 1 C ı b SsD1 sb ps1b 1 C ı b SsD1 sb ps1b
thus proving that the saving eN x0 is given by an affine function of eN .

Solution of Exercise 4.17 Let z 2 RN be an arbitrage opportunity of the second
kind. By Definition 4.17, we have that Dz 0. Since the pricing functional Q is
positive, we have that Q.Dz/ 0 and, since Dz 2 I.D/, it also holds that
0 Q.Dz/ D V.Dz/ D p> z;
thus showing that z cannot be an arbitrage opportunity of the second kind.

Solution of Exercise 4.18 Due to the equivalence between statements (i)-(ii) of
Proposition 4.18, the failure of the Law of One Price implies the existence of a
portfolio zO 2 RN such that DOz D 0 and p> zO ¤ 0. This implies that the price of the
payoff c can be arbitrarily modified by adding multiples of the zero-payoff portfolio
zO to the portfolio z such that Dz D c.
Solution of Exercise 4.19 Note first that, due to part (iii) of Proposition 4.18, the
absence of arbitrage opportunities implies that p> z > 0 and p> z0 > 0, so that
0 0
c
p> z
and p>c z0 are well defined. Arguing by contradiction, suppose that p>c z > p>c z0 .
p> z 0
Define then the portfolio zN WD z p> z0
z. Then, the portfolio zN satisfies p> zN D 0
>
and DNz D c pp> zz0 c0 > 0, thus showing that zN is a riskless arbitrage opportunity
(see Proposition 4.18). This contradicts the assumption that there are no arbitrage
0
opportunities. The case p>c z c z0 can be treated in an analogous way.
Solution of Exercise 4.20 Suppose that pNC1 > p> zc . Then, consider the portfolio
z WD .zc ; 1/ 2 RNC1 . At time t D 0, the value of the portfolio z is given by
p0> z D p> zc pNC1 < 0, while, at time t D 1, the associated payoff is given by
D0 z D Dzc c D 0. Hence, the portfolio z is an arbitrage opportunity (of the second
kind) in the extended market represented by . p0 ; D0 /. An analogous argument can be

applied in the case pNC1 zc , thus showing that pNC1 D p> zc must necessarily
hold in the absence of arbitrage opportunities.
(i): Let zu 2 RN be a portfolio such that Dzu c and, analogously, let zl 2 RN be
a portfolio such that Dzl c. Then, if there are no arbitrage opportunities, we
must have
p> zu p> zl : (11.9)
Indeed, the inequality p> zu zl would easily imply that the portfolio zN WD
zu zl is an arbitrage opportunity, since DNz 0 and p> zN < 0. By taking the
minimum and the maximum on the left- and right-hand sides of (11.9), the
claim follows. In particular, note that it is enough to exclude the existence of
arbitrage opportunities of the second kind in order to establish the claim.
(ii): If c 2 I.D/, there exists a portfolio zc such that Dzc D c. The claim then
follows from the previous step, noting that qu .c/ p> zc ql .c/ and recalling
that V.c/ D p> zc .
(iii): By the first step, it suffices to prove that c … I.D/ implies that qu .c/ ¤ ql .c/.
Arguing by contradiction, suppose that qu .c/ D ql .c/, so that there exist two
portfolios zu ; zl 2 RN such that
Dzu c and Dzl c;
with p> zl D p> zu . Moreover, since c … I.D/, at least one of the above two
inequalities must be a strict inequality. Then, the portfolio zN WD zu zl satisfies
p> zN D 0 and DNz > 0, i.e., zN is an arbitrage opportunity. Since the market is
assumed to be free of arbitrage opportunities, this shows that c … I.D/ implies
that qu .c/ > ql .c/.
Solution of Exercise 4.22 This property is a consequence of the Law of One
0 0
Price. Suppose that there exist two portfolios zc ; zc 2 RN such that Dzc D Dzc ,
0
while p> zc > p> zc . In this case, it would be possible to create wealth at time
0
t D 0 out of nothing, by simply observing that the portfolio zN WD zc zc satisfies
DNz D 0 while p> zN > 0. In this case, at the initial time t D 0, an agent could
0
invest in the portfolio zc and sell short the portfolio zc . Such an investment strategy
yields a strictly positive amount of wealth at time t D 0 and does not affect an
agent’s consumption plan at t D 1, i.e., the portfolio zN is an arbitrage opportunity.
Clearly, any non-satiated agent would invest in an unlimited way into this strategy
and markets could not clear, thus contradicting the existence of an equilibrium. Of
0
course, a similar reasoning holds in the case p> zc zc , thus showing that we
0
must have p> zc D p> zc .
Solution of Exercise 4.23 Suppose that pNC1 qu .c/ (the case pNC1 ql .c/ can
be treated in an analogous way). In view of Definition 4.23, there exists a portfolio
z 2 RN such that Dz c and p> z D qu .c/. In particular, since c … I.D/, the

inequality Dz c cannot be an equality, so that Dz > c. Consider then the portfolio
z0 WD .z; 1/ 2 RNC1 . Then, the value at time t D 0 of such a portfolio is given by
p> z pNC1 0, while the payoff at time t D 1 is given by Dz c > 0. We have
thus shown that the portfolio z0 is an arbitrage opportunity in the extended market
represented by . p0 ; D0 /.
Solution of Exercise 4.31 Arguing by contradiction, suppose that zN 2 RN is an
arbitrage opportunity. If zN is an arbitrage opportunity of the first kind (or if p> zO 0,
so that the portfolio zO is an arbitrage opportunity of the first kind), the claim follows
exactly as in the proof of Proposition 4.27, since ui is strictly increasing in its second
argument. Suppose that p> zO > 0 and that zN is an arbitrage opportunity of the second
>
kind, so that p> zN < 0. Consider the portfolio zQi WD zi C zN pp> zzNO zO. Such a portfolio
satisfies
p> zN
p> zQi D p> zi and Dz0 D Dzi C DNz DOz > Dzi :
p> zO
Since ui is strictly increasing in its second argument, this contradicts the optimality
of zi .
(i): This statement follows from Exercise 4.30 together with the same argument
used in the proof of part (i) of Proposition 4.18.
(ii): This statement follows from Exercise 4.31 together with Proposition 4.18.
Solution of Exercise 4.33 In view of Definition 4.17, it is straightforward to
check that there are no arbitrage opportunities for any price p 2 .0; 1/. Since the
utility function u is strictly increasing, the budget constraint (4.41) is satisfied as an
equality in correspondence of an optimal portfolio, so that the portfolio optimization
problem can be rewritten as
e1 e2
max e0 C C C z.1 p/ :
z2R 2 2
Clearly, since p 2 .0; 1/ and consumption is not restricted to be non-negative, this

problem does not admit an optimal solution.
Solution of Exercise 4.34 We can apply the representative agent analysis, denoting
by u0 and u1 the representative agent’s utility functions for consumption at time
t D 0 and at time t D 1, respectively. The price q1 of the first Arrow security is equal
to the corresponding state price m1 , which in turn is given by, due to equation (4.52),
u01 .e1 / 1 u01 .e1 /

m1 D ı D P ;
u00 .e0 / rf SsD1 s u01 .es /
where the second equality follows from equation (4.50). Since the aggregate
endowment in correspondence of the different states of the world satisfies
e1 : : : eS , it follows that m1 1=rf , due to the concavity of u1 .
Solution of Exercise 4.35 If markets are complete and there is no aggregate
risk, then any equilibrium allocation is Pareto efficient and characterized by an
equal optimal consumption in correspondence of any state of the world at time
t D 1 (i.e., the risk is perfectly diversified). The claim then follows directly from
equation (4.48), since the covariance with respect to a constant random variable is
zero.
(i): From the analysis of Sect. 4.3, since the agents’ endowments are expressed in
terms of units of the traded securities, we know that the equilibrium allocation
of the original economy with I agents is Pareto optimal and characterized by
a linear sharing rule (see Proposition 4.5). As a consequence, the equilibrium
allocation can be obtained as the no-trade equilibrium of a single represen-
tative agent. Moreover, as shown in the proof of Proposition 4.15, the utility
function of the representative agent is of the generalized power form.
(ii): In view of equation (4.52), in the representative agent economy, the stochastic
discount factor valuation rule, applied to the payoff of the Call option
(recall that, as explained at the end of Sect. 4.4, the stochastic discount
factor valuation rule (4.52) can be also applied to payoffs not traded in
the financial market, compare also with (4.49)), together with the explicit
form of the representative agent’s utilityP function obtained in the proof of
Proposition 4.15, gives that (recall that IiD1 i D 0)

" 1b #
Q u01 .Qe/ Q eQ
p call;n
D ı E maxfdn kI 0g 0 D ı E maxfdn kI 0g :
u0 .e0 / e0
(iii): Thanks to the distributional assumptions, the expected value (4.64) can be
explicitly calculated. Indeed,
Z 1 Z 1
k
pcall;n D pn e y
ex f .x; y/dy dx
1 log pn
k p n
Z 1Z 1 Z 1Z 1
D pn e
xCy
f .x; y/dy dx k ex f .x; y/dy dx;
1 log k
pn 1 log pkn
(11.10)
1
where f denotes the joint density of the random variables .log.ı. eeQ0 / b /;
Q
log. pdnn //. Recall that, from the basic properties of the bivariate normal distribu-
tion, the conditional distribution of log.ıQe=e0 /1=b given that log.dQ n =pn / D y,
with density denoted by f .xjy/, is normal with mean e C .e =n /.y n /
and variance .1 2 /e2 . This allows us to compute the second term of (11.10)
as follows:
Z 1Z 1 Z 1 Z 1
e f .x; y/dydx D f .y/ ex f .xjy/dx dy
x
1 log k
pn log k
pn 1
Z 1 .12 /e2
e C.e =n /.yn /C
D e 2 f .y/dy
log pkn
2 Z 1 2 e2 .yn /2
ee Ce =2 .e =n /.yn / 2
2n2
D p e dy
2n log pkn
2 Z 1 .yn e n /2
ee Ce =2
2n2
D p e dy
2n log pkn
2 Z 1
ee Ce =2 y2
D p . /
log pk n
e 2 dy
2n n
n e
pn !
e Ce2 =2
log C n
De N k
C e :
n
(11.11)
Similarly, for the first term of (11.10) we have

Z 1 Z 1 Z 1 Z 1
exCy f .x; y/dy dx D ey f .y/ ex f .xjy/dxdy
1 log pkn log pkn 1
Z 1 .1 2 /e2
D eyCe C.e =n /.yn /C 2 f .y/dy
log pkn
e2 Z 1 2 e2 .yn /2

ee C 2 yC.e =n /.yn /
2n2
D p e 2
dy
2n log k
pn
e2 C2e n Cn2 Z 1 .yn e n n2 /2

ee Cn C 2
2n2
D p e dy
2n log k
pn
!
e Cn C
e2 C2e n Cn2 log. pkn / C n
De 2 N C e C n :
n
(11.12)
Note that
" 1b #
e C
e2 eQ 1
e 2 D ıE D ; (11.13)
e0 rf
since the above expression represents the arbitrage free price of a risk free
asset which pays one unit of wealth in correspondence of every possible state
of the world. Similarly, applying formula (4.52) to the underlying security of
the Call option we get
" 1#
e Cn C
e2 C2e n Cn2 dQ n eQ b
e 2 D ıE D 1: (11.14)
p n e0
Hence, putting together (11.10), (11.11), (11.12), (11.13) and (11.14), we get
!
log. pkn / C n k log pkn C n
p call;n
D pn N C e C n N C e
n rf n

log. pkn / C log.rf / n k log. pkn / C log.rf / n
D pn N C N ;
n 2 rf n 2
where the last equality follows from (11.13)–(11.14), which imply that n C
e n C n2 =2 D log.rf /. This yields the explicit formula (4.65).
Solution of Exercise 4.37 Consider first the case of an unleveraged firm subject to
taxation. In this case, at date t D 1, a tax equal to V 1 .!s / has to be payed and,
similarly to (4.58), the firm value at the initial date t D 0 is then given by
X
S
V0un;tax D .1 / ms V1 .!s /;
sD1
for any choice of the state price vector m 2 RSCC . On the other hand, the value at
t D 0 of a leveraged firm with debt of nominal value K and subject to taxation is
given by
X
S X
S
V0lv;tax D .1 / ms maxfV1 .!s / KI 0g C ms minfV1 .!s /I Kg
sD1 sD1
X
S X
S
D .1 / ms V1 .!s / C minfV1 .!s /I Kg
sD1 sD1
D V0un;tax C B0 ;
where the last equality makes use of relation (4.59), thus proving the claim.
Solution of Exercise 5.1 Observe first that

ıu01 .Qxm / m ı ı
Cov ; rQ D Cov u01 .Qxm /; rQ m D Cov u01 .wm
0rQ m /; rQm ;
u00 .xm
0/ u00 .xm
0/ u00 .xm
0/
PI PN
where wm 0 WD nD0 en pn denotes the value at the initial date t D 0 of the
i
iD1
market portfolio. Since the representative agent’s utility function is strictly concave
0
and wm0 > 0, the map x 7! u1 .w0 x/ is strictly decreasing. It then follows that, using
m
the definition of covariance,

m
Cov u01 .wm
0 rQ /; rQ
m m
D E u01 .wm
0rQ m / EŒu01 .wm0 rQ / .Q
m
r EŒQrm /

D E u01 .wm
0rQ m / u01 .wm
0 EŒQrm / C u01 .wm rm / EŒu01 .wm
0 EŒQ 0rQ m /
.Qrm EŒQrm /

D E u01 .wm Q m / u01 .wm
0r 0 EŒQrm / .Qrm EŒQrm /

D E u01 .wm Q m / u01 .wm
0r 0 EŒQrm / .Qrm EŒQrm / 1fQrm >EŒQrm g

C E u01 .wm 0rQ m / u01 .wm0 EŒQ
rm / .Qrm EŒQrm / 1fQrm EŒQrm g
0;
from which the claim then follows directly.

Solution of Exercise 5.2 Let us denote by w 2 RN a portfolio, parameterized in
terms of proportions of initial wealth invested in the N risky assets (with 1 w> 1
being the proportion of wealth invested in the riskless asset). The corresponding
random return is given by
rQ w D rf C w> .Qr rf /;
where rQ D .Qr1 ; : : : ; rQn /> . Hence:

Corr u01 .Qxm /; rQ w D Corr u01 .Qxm /; w> rQ :
Consider then the minimization problem

E u01 .Qxm / EŒu01 .Qxm / w> .Qr EŒQr/
min Corr u01 .Qxm /; w> rQ D min p p :
w2RN w2RN Var.u01 .Qxm // Var.w> rQ /
The first order condition for the above problem yields

> >
Cov u01 .Qxm /; rQn Var .w/
O rQ Cov u01 .Qxm /; .w/
O > rQ Cov .w/
O rQ ; rQn D 0;
for all n D 1; : : : ; N. Equivalently, we have that

Cov u01 .Qxm /; rQn Cov rQ wO ; rQn
D ; for all n D 1; : : : ; N: (11.15)
Cov u01 .Qxm /; rQ wO Var rQwO
Relation (5.3) applied to the portfolio wO gives

ı Cov u01 .Qxm /; rQ wO
EŒQrwO rf D rf
u00 .xm
0/
from which (5.8) follows by relying on (5.3) and making use of (11.15).
Solution of Exercise 5.3 For simplicity of notation, we shall omit the superscript

for denoting the optimal solution. In correspondence of an equilibrium allocation,
the optimal consumption W e i of every agent i has to satisfy the first order optimality
condition (3.4), so that
0 i
e / EŒQrn rf D Cov ui0 .W
E u i .W e i /; rQn ; for all n D 1; : : : ; N:
Due to the assumption of normally distributed returns, the couple .W e i ; rQn / is

distributed according to a bivariate normal law, for all n D 1; : : : ; N. Hence, we
can apply Stein’s lemma (see Lemma 3.9):
0 i
e / EŒQrn rf D E ui00 .W
E u i .W e i / wi0 Cov rQwi ; rQn ; for all n D 1; : : : ; N;
i
where rQ w denotes the return on the optimal portfolio of the i-th agent and wi0 denotes
the total wealth allocated by agent i at date t D 0 in the N C 1 available assets (recall
i
also that rQw D We i =wi0 ). Dividing by EŒui00 .W
e i /, summing over all i D 1; : : : ; I and
rearranging terms, we get
X
I 1
i 1
EŒQrn rf D wm
0 . / Cov.Qrm ; rQn /; for all n D 1; : : : ; N; (11.16)
iD1
PI P
the economy’s aggregate wealth and rQ m D IiD1 W e i =wm .
where wm0 WD
i
iD1 w0 is P 0
I i 1 1
Note that the quantity . iD1 . / / represents the harmonic average of the
individual global absolute risk aversion coefficients. Moreover, relation (11.16) also
holds for arbitrary portfolios of the traded assets and, hence, in particular for the
market portfolio itself, thus leading to
X
I 1
EŒQrm rf D wm
0 . i /1 Var.Qrm /: (11.17)
iD1
We have thus shown that the risk premium on the market portfolio can be expressed
in terms of the harmonic mean of the individualPI global absolute risk aversion
i 1 1
coefficients. Note also that the quantity wm
0 . iD1 . / / can be thought of as an
aggregate relative risk aversion coefficient of the economy. Moreover, by combining
equations (11.16) and (11.17) we immediately recover the CAPM relation (5.9).
Solution of Exercise 5.4 Assuming a quadratic utility function, the global absolute
risk aversion coefficient of agent i is given by
1
ai e i
i D EŒW ; for all i D 1; : : : ; I:
bi
Then, the same arguments employed in the solution of Exercise 5.3 allow to show
that
I i
!1
X a
EŒQr rf D
m
wm
0
e m
EŒW Var.Qrm /:
iD1
bi
Solution of Exercise 5.6 Recall that, in the representative agent economy, the
equilibrium allocation is characterized in terms of the representative agent’s optimal
consumption problem. In particular, the market portfolio represents the optimal
portfolio for the representative agent in a single agent economy. Hence, we can
apply the results of Sect. 2.2 on the optimal portfolio problem of a single agent. In
particular, the result follows directly from Propositions 2.5 and 2.6.
(i): As in the proof of Proposition 5.6, note that EŒÒ rQn D 1, for all n D 1; : : : ; N.
Hence, letting V.`/O D p> z` be the value at time t D 0 of the portfolio z` , it
holds that, for an arbitrary portfolio z 2 RN such that z> 1 D 1,
X
N X
N
1D zn D zn EŒÒ rQn D EŒÒ rQ z D V.`/EŒQ
O r` rQ z :
nD1 nD1

In particular, taking z D z` , this implies that E rQ z rQ ` D E .Qr ` /2 .
(ii): We first prove that the return of the portfolio z` minimizes the second moment
among all possible returns. Indeed, in view of part (i), for any portfolio z, it
holds that

E .Qr z /2 D E .Qr z Qr` C rQ ` /2 D E .Qr z Qr` /2 C EŒ.Qr` /2 C 2E rQ ` .Qr z Qr` /

D E .Qr z Qr` /2 C EŒ.Qr` /2 E .Qr` /2 ;
where the third equality follows from the first part of the exercise. The claim
then follows by noting that Var.Qr z / D EŒ.Qr z /2 .EŒQr z /2 , so that the portfolio
z` minimizes the variance among all portfolios having expected return equal to
EŒQr` .
Q f,
Q D `=r
Solution of Exercise 5.8 Defining the stochastic discount factor as m
equation (4.35) implies that
1 D EŒm
Q rQn ; for all 1; : : : ; N:
Recall that m Q > 0, due to the absence of arbitrage opportunities (see Proposi-
tion 4.22). Hence, if rQn can take non-positive values with strictly positive probability,
then there is nothing to prove, since EŒlog.Qrn / D 1. So, suppose that rQn > 0, for
all n D 1; : : : ; N. Then, applying Jensen’s inequality, we obtain

0 D log EŒm Q rQn E log.mQ rQn / ;
so that

E log.Qrn / E log.m/
Q ;
Q rn is constant.
with the equality holding only in the case where mQ
Solution of Exercise 5.9 We restrict our attention to the case where A=C > rb > r` ,
using the notation introduced in section “The Case of N Risky Assets and a Risk
Free Asset”. As shown in Fig. 3.5, in this case there are two tangent portfolios,
denoted by we;b and we;` , corresponding to the two risk free rates rb and r` , respec-
tively. As made clear by Fig. 3.5, the portfolio frontier consists of four regions:
(A) positive investment in the risk free asset with return r` and short sale of the
tangent portfolio we;` ;
(B) positive investment in both the risk free asset with return r` and in the tangent
portfolio we;` ;
(C) all the wealth invested in the risky assets;
(D) borrowing at the rate rb and positive investment in the tangent portfolio we;b .
If agents choose to hold efficient portfolios, then they will only choose portfolios
belonging to the regions B, C and D. In any of those regions, the risky component of
a frontier portfolio will be represented by a linear combination of the two portfolios
we;b and we;` (due to the properties of the portfolio frontier discussed in section “The
Case of N Risky Assets”). Hence, if the aggregate supply of the risk free asset is
zero, then the market portfolio will be only composed by the risky assets and given
as a convex linear combination of the two tangents portfolios we;b and we;` . Hence,
the market portfolio belongs to the portfolio frontier and relation (5.37) follows
by the same arguments used in the proof of Proposition 5.7. The efficiency of the
market portfolio (i.e., the fact that EŒQrm > EŒQrzc.m/ ) follows from the fact that
the market portfolio is given by a convex linear combination of the two tangent
portfolios we;b and we;` and the latter are efficient since A=C > rb > r` . Finally, it
remains to show that
rb EŒQrzc.m/ r` :
To this effect, recall that, due to equation (3.30), the portfolio wzc.m/ satisfies
A D=C2
EŒQrzc.m/ D :
C EŒQr A=C
m
e;`
Moreover, it also holds that EŒQrm EŒQrw , so that, applying equation (3.39)
(replacing rf with r` ), we get
A D=C2 A D=C2
EŒQrzc.m/ D D r` :
C EŒQrw A=C
e;`
C .Ar` B/=.Cr` A/ A=C
e;b
Noting that EŒQrm EŒQrw , an analogous computation shows that EŒQrzc.m/ rb .
Solution of Exercise 5.10 We restrict our attention to the case where A=C > rf ,
using the notation introduced in section “The Case of N Risky Assets and a Risk
Free Asset”. As shown in Fig. 3.4, denoting by we the tangent portfolio, borrowing
would occur only at the right of the point ..Qrwe /; EŒQrwe /, while, on the left of that
point, an agent would invest a positive amount of wealth both in the risk free asset
and in the tangent portfolio. Since borrowing is not allowed and all the agents choose
to hold efficient portfolios, this implies that the market portfolio is a convex linear
combination of the risk free asset and the tangent portfolio we . In particular, the mar-
ket portfolio is efficient (i.e., EŒQrm > EŒQrzc.m/ ) and relation (5.38) can be obtained
similarly as in the proof of Proposition 5.7. It remains to show that EŒQrzc.m/ rf .
To this effect, recall that, due to equation (3.30), the portfolio wzc.m/ satisfies
A D=C2
EŒQrzc.m/ D :
C EŒQrm A=C
Moreover, it also holds that EŒQrm EŒQrwe , so that, applying equation (3.39), we get
A D=C2 A D=C2
EŒQrzc.m/ D D rf :
C EŒQrwe A=C C .Arf B/=.Crf A/ A=C
Solution of Exercise 5.11 Under the present assumptions, it holds that
rd D rf C ˇd .rm rf /;
run D rf C ˇun .rm rf /;
rlv D rf C ˇlv .rm rf /:
On the other hand, relation (4.63) gives that
B0
rlv D run C .run rd / :
S0lv
By rewriting the last identity using the CAPM relations for run and rd , we get that
B0
rlv D rf C ˇun .rm rf / C .ˇun ˇdm /.rm rf / ;
S0lv
thus implying that

B0 B0
ˇlv D 1 C lv ˇun C lv ˇd :
S0 S0
Solution of Exercise 5.16 Proposition 5.16 can be proved by relying on arguments

analogous to those used in the proof of Proposition 5.15. Suppose that, for all N 2 N,
in the N-th economy it is possible to trade a risk free asset with rate of return rfN > 0.
For each economy N 2 N, consider then the expected excess returns EŒQrnN rfN of
the N risky assets with respect to the risk free rate rfN . Take the orthogonal projection
of the expected excess returns onto the linear space spanned by the columns of the
matrix BN , so that
EŒQrN rfN D BN N C cN ;
for some N 2 RK and cN 2 RN with .cN /> BN D 0. At this point, one can follow
exactly the proof of Proposition 5.15 and the result is obtained by replacing N0
with rfN .
Solution of Exercise 5.17 By equation (3.9), note that, for all i D 1; : : : ; I,
Vwi D i .i rf 1/;
with i D .i1 ; : : : ; iN /> 2 RN denoting the vector of expected returns with

respect to the beliefs of agent i. Hence, using the notation introduced before
Proposition 5.20, it holds that, for all n D 1; : : : ; N,
X
N X
N X
I
wik 1 X
I X
N
Cov .Qrm ; rQn / D wm
k Vnk D PN PI V D PN
j nk PI j
wik Vnk
kD1 kD1 iD1 lD1 jD1 wl lD1 jD1 wl iD1 kD1
1 X
I
D PN PI j
i .in rf /;
lD1 jD1 wl iD1
and, moreover,
X
N
1 X
N X
I
Var.Qrm / D wm rm ; rQn / D PN
n Cov.Q PI j
wm
n i .in rf /
nD1 lD1 jD1 wl nD1 iD1
1 X
I X
N
1 X
I
D PN PI j
i n .n rf / D PN
wm i
PI j
i .i;M rf /:
lD1 jD1 wl iD1 nD1 lD1 jD1 wl iD1
PI
Dividing the last two expressions by jD1 j , we get
1 X
I
i 1
Cov .Qr ; rQn / D PN
m
PI j PI .in rf / D PN PI j
.N n rf /
lD1 jD1 wl iD1 jD1 j lD1 jD1 wl
and
1 X
I
i 1
Var.Qr / D PN
m
PI j PI .i;M rf / D PN PI j
.N M rf /:
lD1 jD1 wl iD1 jD1 j lD1 jD1 wl
Relation (5.36) follows directly by combining the last two equations.
Solution of Exercise 6.2 By definition, it holds that W N t ./ D t> .SN t C D

N t /,
for all t D 0; 1; : : : ; T. Moreover, it is immediate to see that the self-financing
conditions (6.1)–(6.2) hold as well with respect to discounted quantities (it suffices
to divide both sides of the equality by rft ). By combining these two observations, it
follows that
N tC1 ./ WD W
W N tC1 ./ W
N t ./ D tC1
> N N tC1 SN t / cN t
.StC1 C D
>
D tC1 SN tC1 C tC1
> N
DtC1 cN t ;

Solution of Exercise 6.3 Let x0 > x > 0. For every " > 0, there exists a
consumption plan .c"s /sDt;:::;T 2 CtC .x/ such that
X
T
V.x; t/ u.c"t / C ı st EŒu.c"s /jFt C ":
sDtC1
Then, since the function u is strictly increasing, it holds that
X
T
V.x; t/ < u.c"t C x0 x/ C ı st EŒu.c"s /jFt C "
sDtC1
0
V.x ; t/ C "
By the arbitrariness of ", it holds that V.x; t/ < V.x0 ; t/. In order to prove the
concavity of the value function, let x; x0 > 0, .cs /sDt;:::;T 2 CtC .x/, .c0s /sDt;:::;T 2
CtC .x0 / and 2 .0; 1/ and define x WD x C .1 /x0 . It is straightforward to
check that the consumption process .cs /sDt;:::;T defined by cs WD cs C .1 /c0s ,
for all s D t; : : : ; T, belongs to CtC .x /. Due to the concavity of the utility function
u, it holds that
u.cs / C .1 /u.c0s / u.cs /;
for all s D t; : : : ; T, so that

! !
X
T X
T
u.ct / C ı st EŒu.cs /jFt C .1 / u.c0t / C ı st EŒu.c0s /jFt
sDtC1 sDtC1
X
T
u.ct / C ı st EŒu.cs /jFt :
sDtC1
By taking suprema, it then follows that
V.x; t/ C .1 /V.x0 ; t/ V.x ; t/;
thus proving the concavity of V.; t/.

Solution of Exercise 6.4 The proposition can be proved by relying on the backward
induction scheme explained after Proposition 6.3. Let us start at date T 1, with one
period remaining until the terminal date T. Then, the optimality condition (6.20)
implies that
" !ˇ #
1 X n n ˇ
N
1
D ıE w .r rf / C rf ˇˇFT1
cT1 WT nD1 T T

ˇ
1 ˇ ı
D ıE ˇFT1 D ;
WT1 cT1 WT1 cT1
where the second equality follows from the self-financing condition (6.19) and the
third equality from the adaptedness of the processes W and c . In turn, this implies
that

WT1
cT1 D :
1Cı
Having computed the optimal consumption at date T 1, we can now compute the

value function V.WT1 ; T 1/ at date T 1. By the Bellman equation (6.11), it
follows that, using the self-financing condition (6.19) and recalling that WT D cT ,
we get

V.WT1 ; T 1/ D log.cT1 / C ıE log.cT /jFT1

WT1
D log
1Cı
" X !! ˇ #
1
N
ˇ

C ıE log WT1 1 n n
wT .rT rf / C rf ˇFT1
1Cı ˇ
nD1

D .1 C ı/ log.WT1 /C T1 ;
with

" !ˇ #
ı XN
ˇ
T1 WD log.1Cı/Cı log CıE log wT .rT rf / C rf ˇˇFT1 :
n n
1Cı nD1
We have thus shown that (6.22) and (6.23) hold for t D T 1. Now we use backward
induction. Hence, let us assume that the proposition holds at a generic date t C 1,
for t D 0; 1; : : : ; T 1. Then, equation (6.20) together with (6.22) implies that
" !ˇ #
1 XN
ˇ

0 0
D u .ct / D ıE V .WtC1 ; t C 1/ wtC1 .rtC1 rf / C rf ˇˇFt
n n
ct nD1
" !ˇ #
f .t C 1/ X n n ˇ
N
D ıE wtC1 .rtC1 rf / C rf ˇˇFt
WtC1 nD1

ˇ
1 ˇ
D ıf .t C 1/E ˇFt D ı f .t C 1/ ;
Wt ct ˇ
Wt ct
so that
Wt
ct D :
1 C ıf .t C 1/
The function f W f0; 1; : : : ; Tg ! R satisfies f .T/ D 1 and f .t/ D 1 C ıf .t C 1/,

for all t D 0; 1; : : : ; T 1, so that it is explicitly given by f .t/ D 1ı1ı , for all
TtC1
t D 0; 1; : : : ; T. In turn, due to condition (6.11), it holds that
V.Wt ; t/ D u.c
t / C ıEŒV.WtC1 ; t C 1/jFt

D u.c
t / C ıE f .t C 1/ log.WtC1 /jFt C ıEŒ tC1 jFt

Wt
D log C ıEŒ tC1 jFt
1 C ıf .t C 1/
" !! ˇ #
XN
ˇ
C ıf .t C 1/E log Wt ct n
wtC1 .rtC1 rf / C rf
n ˇFt
ˇ
nD1

D 1 C ıf .t C 1/ log.Wt / log 1 C ıf .t C 1/ C ıEŒ tC1 jFt
" X !! ˇ #
1
N
ˇ
C ıf .t C 1/E log 1 wn ˇFt
tC1 .rtC1 rf / C rf
n
1 C ıf .t C 1/ ˇ
nD1
D f .t/ log.Wt / C t;
thus showing that equation (6.22) holds, with

t D log 1 C ıf .t C 1/ C ıEŒ tC1 jFt
" X !! ˇ #
1
N
ˇ
C ıf .t C 1/E log 1 n
wtC1 .rtC1 rf / C rf
n ˇFt
1 C ıf .t C 1/ ˇ
nD1

f .t/ 1
D ıf .t C 1/ log
f .t/
" ! ˇ #!
log f .t/ XN
ˇ
C E log wtC1 .rtC1 rf / C rf ˇˇFt
n n
ıf .t C 1/ nD1
C ıEŒ tC1 jFt :
Finally, the optimal portfolio process .wt /tD1;:::;T is characterized by equa-

tion (6.21), namely:

ˇ
f .t C 1/ n ˇ
0 D .Wt ct /E .r tC1 rf / ˇFt
WtC1
2 !1 3
X ˇ
N
ˇ
D f .t C 1/E 4 wn
tC1 .rtC1 rf / C rf
n
.rtC1
n
rf /ˇˇFt 5 ;
nD1
for all t D 1; : : : ; T 1 and n D 1; : : : ; N, thus proving equation (6.24), since

f .t/ > 0 for all t D 0; 1; : : : ; T.
Solution of Exercise 6.5 Recall first that any self-financing trading-consumption

strategy .; c/ satisfying P.cT > 0/ D 1 necessarily satisfies Wt > ct for all t D
0; 1; : : : ; T 1. Indeed, by Lemma 6.19 (applied with s D T), if cT .AT / > 0 for
all events AT 2 FT , then it holds that ct .At / > 0 for all events At 2 Ft and all
dates t D 0; 1; : : : ; T 1. Therefore, we can use the parametrization in terms of the
portfolio process .wt /tD1;:::;T defined in (6.18). The corollary can then be proved by
means of the following computations:

WtC1 ˇˇ 1
E ˇFt D E WtC1 V 0 .WtC1

; t C 1/jFt
WtC1 f .t C 1/
" ! ˇ #!
Wt ct XN
ˇ
D E V 0 .WtC1

; t C 1/ wntC1 .rtC1
n
rf / C rf ˇˇFt
f .t C 1/ nD1
Wt ct 1 Wt ct
D
D
f .t C 1/ ıct . f .t/ 1/ct
Wt ct
D ;
Wt ct
for all t D 0; 1; : : : ; T 1, where the first equality follows from equation (6.22), the
second equality from (6.19), the third equality from the optimality condition (6.20),
the fourth equality from the fact that f .t/ D 1 C ıf .t C 1/ (see Exercise 6.4) and,
finally, the last equality from (6.23).
Solution of Exercise 6.6 The proposition can be proved by relying on the backward
induction scheme explained after Proposition 6.3. Let us start at date T 1, with one
period remaining until the terminal date T. Then, the optimality condition (6.20)
implies that
" !ˇ #
XN
ˇ

.cT1 / 1 1
D ıE .WT / n n ˇ
nD1
" ! ˇ #
X
N
ˇ

D ıE .WT1 cT1 / 1 wn ˇFT1 ;
T .rT rf / C rf
n
ˇ
nD1
where the second equality follows from the self-financing condition (6.19). Hence,

using the FT1 -measurability of WT1 and cT1 , it follows that
" ! ˇ #! 1
1
1 X
N
ˇ
cT1 D ı
.WT1 cT1 / E wn ˇFT1
T .rT rf / C rf ;
1 n
ˇ
nD1
which leads to
cT1 D aT1 WT1

;
with
hP ˇ i 1
1
1
N n n ˇ
ı 1 E w
nD1 T .rT rf / C rf ˇF T1
aT1 D hP ˇ i 1
1
1
N n n ˇ
1 C ı 1 E w
nD1 T .rT r f / C rf ˇFT1
0 11
" ! ˇ #! 1
1
B X
N
ˇ C
D @1 C ıE wn ˇFT1
T .rT rf / C rf A :
n
ˇ
nD1
Having computed the optimal consumption at date T 1, let us now compute the

value function V.WT1 ; T 1/ at date T 1. By the Bellman equation (6.11), it
follows that, using the self-financing condition (6.19) and recalling that WT D cT ,

ˇ
.cT1 / .cT / ˇ
V.WT1 ; T 1/ D C ıE ˇFT1

" ! ˇ #

aT1 WT1
WT1 .1 aT1 / XN
ˇ
D C ıE n n ˇ
nD1

1 .WT1 /
D aT1 ;

where the third equality follows from the fact that, as can be checked by means of
straightforward computations,
" ! ˇ #
ı.1 aT1 / X
n
ˇ
aT1 C E wn ˇ
T .rT rf / C rf ˇFT1 D 1:
n
1
aT1 nD1
We have thus shown that (6.25) and (6.26) hold for t D T 1. Now we use backward
induction. Hence, let us assume that the proposition holds at a generic date t C 1,
for t D 0; 1; : : : ; T 1. Then, equation (6.20) together with (6.25) implies that
" !ˇ #
X
N
ˇ
.ct / 1 D u0 .ct / D ıE V 0 .WtC1

; t C 1/ wn ˇ
tC1 .rtC1 rf / C rf ˇFt
n
nD1
" ! ˇ #
1 X
N
1 ˇ
D ıE Wt ct wn
tC1 .rtC1
n
rf / C rf atC1 ˇˇFt
nD1
" ! ˇ #
1 X
N
1 ˇ
D Wt ct ıE wn
tC1 .rtC1
n
rf / C rf atC1 ˇˇFt ;
nD1
so that
ct D at Wt ;
with
0 11
" ! ˇ #! 1
1
B X
N
ˇ C
atC1 ˇˇFt
1
at D @1 C ıE wn
tC1 .rtC1 rf / C rf
n
A :
nD1
In turn, due to condition (6.11), it holds that
V.Wt ; t/ D u.c
t / C ıEŒV.WtC1 ; t C 1/jFt
" #
ˇ
.at Wt / 1 .WtC1 / ˇˇ
D C ıE atC1 ˇFt

" ! ˇ #
.at Wt / Wt .1 at / X N
1 ˇˇ
n
D C ıE wtC1 .rtC1 rf / C rf
n
atC1 ˇFt
nD1
" ! ˇ #!

1 .Wt / .1 at / X
N
1 ˇ
D at at C ı 1
E wn
tC1 .rtC1
n
rf / C rf atC1 ˇˇFt
at nD1

1 .Wt /
D at ;

thus showing that equation (6.25) holds. Finally, the optimal portfolio process
.wt /tD1;:::;T is characterized by equation (6.21), namely:
2 ! 1 3
ˇ
X
N
ˇ
rf /ˇˇFt 5
1
0 D E 4atC1 .Wt / 1 .1 at / 1 wn
tC1 .rtC1 rf / C rf
n
.rtC1
n
nD1
for all t D 1; : : : ; T 1 and n D 1; : : : ; N, from which equation (6.27)

follows, noting that the random variable .Wt / 1 .1 at / 1 is strictly positive and
Ft -measurable.
Solution of Exercise 6.8 Starting from the terminal date T, if the FT1 -conditional
distribution of the asset returns rTn , n D 1; : : : ; N, is equal to the unconditional
distribution of r1n , n D 1; : : : ; N, then the optimal portfolio wT which solves (6.27)
is deterministic and can be characterized as the solution w to
2 ! 1 3
XN
E4 wn .r1n rf / C rf .r1n rf /5 D 0; for all n D 1; : : : ; N:
nD1
(11.18)
In turn, this implies that aT1 defined in Proposition 6.8 is also a deterministic
quantity. By induction, for any t D 1; : : : ; T 1, if the random variable atC1 is
deterministic, then the optimal portfolio wtC1 which solves (6.27) will not depend
1
on the term atC1 and, hence, is characterized by the deterministic vector w which
solves (11.18). In particular, this implies that at is itself a deterministic quantity.
This shows that the process .at /tD0;1;:::;T reduces to a deterministic function of
time. As a consequence, the result of Corollary 6.9 holds true, in the more specific
form (11.18). Note also that, due to Proposition 6.8, defining
" ! #
X
N
n
b WD E w .r1n rf / C rf ;
nD1
the process .at /tD0;1;:::;T satisfies aT D 1 and
1 1 1
D 1 C .ıb/ 1 ;
at atC1
for all t D 0; 1; : : : ; T 1, so that
X Tt TtC1
1 s 1 .ıb/ 1
D .ıb/ 1 D 1
;
at sD0 1 .ıb/ 1
for all t D 0; 1; : : : ; T.
Solution of Exercise 6.10 Suppose that qAsjAt < qAs =qAt and consider the following
strategy at the initial date t D 0: buy qAs =qAt units of the Arrow security paying
in correspondence of At and sell short one unit of the Arrow security paying in
correspondence of As . The initial wealth needed to finance such a strategy is null,
since .qAs =qAt /qAt qAs D 0. At time t, if event At is realized, this strategy generates
the payoff qAs =qAt and we use part of it to buy one unit of the Arrow security paying
in correspondence of As , at a price of qAs jAt . This trade still leaves the strictly positive
amount qAs =qAt qAs jAt . At the later date s, if event As is realized, then the cashflows
related to the long and short positions in the Arrow security paying in As offset
exactly, while if event As does not occur then nothing happens. Also, if at date t
event At does not occur (and, hence, As does not occur as well), the strategy expires
worthless. We have thus shown that this strategy has zero cost and generates a non-
negative and non-null consumption plan, i.e., an arbitrage opportunity. If such a
strategy would be possible, then every agent would invest in it in an unlimited way,
thus contradicting the existence of an optimal portfolio. The case qAs jAt > qAs =qAt
can be dealt with in an analogous way.
Solution of Exercise 6.11 The strict monotonicity of the utility functions P implies
that in equilibrium the available resources are fully allocated, i.e., IiD1 ti;n D
PI
iD1 0 D 1, for all n D 1; : : : ; N and t D 1; : : : ; T. Moreover, for every
i;n
i D 1; : : : ; I, the couple . i ; ci / is a self-financing trading-consumption strategy,

in the sense of Definition 6.1, so that condition (6.1) holds. Hence, summing
condition (6.1) over i, we get
X
I X
I X
N
i;n n
X
N
t D
ci ti;n .snt C dtn / tC1 st D dtn ;
iD1 iD1 nD1 nD1
for all t D 0; 1; : : : ; T, thus proving the claim.

Solution of Exercise 6.12 As a preliminary, let us introduce some notation. For
a given consumption process c D fcs .As /I As 2 Fs ; s D 0; 1; : : : ; Tg and any
date t D 1; : : : ; T and event At1 2 Ft1 , we denote by Ct .At1 / the t1 .At1 /-
dimensional vector of all possible values of the consumption plan c at date t
.A /
in correspondence of the events .A1t ; : : : ; At t1 t1 / that are contained in At1 .
.i/ ) .iii/: in view of Definition 6.14, this implication is trivial, since the dividend
process of an Arrow security paying a unitary dividend at date t in correspondence of
the event At 2 Ft and zero otherwise can be represented as a consumption process
c D fcs .As /I As 2 Fs ; s D 0; 1; : : : ; Tg such that cs .As / D 1At DAs if s D t and zero
otherwise.
.iii/ ) .ii/: for any t D 1; : : : ; T and At 2 Ft , consider the Arrow security
paying a unitary dividend in correspondence of event At at date t and zero otherwise.
The dividend process of this Arrow security can be represented by means of a
consumption process c D fcs .As /I As 2 Fs ; s D 0; 1; : : : ; Tg as above. By
assumption, there exists a trading strategy D .s /sD0;1;:::;T such that .; c/ 2
A .x0 /, for some x0 2 R. Since the process is predictable, it can be represented as
D f0 ; s .As1 /I As1 2 Fs1 ; s D 1; : : : ; Tg, with s .As1 / 2 RN denoting the
position in the N securities over the time interval Œs 1; s given the realization of
the event As1 at date s 1. Moreover, since the process c is null after date t, we
can assume without loss of generality that is null as well after date t. Therefore,
in correspondence of date t 1 and any event At1 2 Ft1 such that At At1 ,
condition (6.1) implies that
Rt1 .At1 / 3 .0 0 1 0 0/> D Ct .At1 / D Pt .At1 /t .At1 /;
with the element 1 being in correspondence of the event At . Equivalently, it holds

that .0 010 0/> 2 I.Pt .At1 // and, since the event At is arbitrary, this implies
that I.Pt .At1 // D Rt1 .At1 / , i.e., rank.Pt .At1 // D t1 .At1 /. Since this holds
for any t D 1; : : : ; T, the implication .iii/ ) .ii/ is proved.
.ii/ ) .i/: let c D fct .At /I At 2 Ft ; t D 0; 1; : : : ; Tg be an arbitrary consumption
plan. At the terminal date T and for any event AT1 2 FT1 , the assumption that
rank.PT .AT1 // D T1 .AT1 / implies that there exists a T1 .AT1 /-dimensional
vector T .AT1 / such that
CT .AT1 / D PT .AT1 /T .AT1 /:
Since the event AT1 is arbitrary, we can then define an FT1 -measurable N-
dimensional random vector T such that cT D T> DT holds with probability one
(recall that ST D 0). By backward induction, at any date t D 1; : : : ; T 1 and for
any event At1 2 Ft1 , the assumption that rank.Pt .At1 // D t1 .At1 / implies
that there exists t .At1 / 2 RN such that
>

ct .At / C tC1 .At /St .At / D t .At1 /> St .At / C Dt .At / (11.19)
holds true for all events At 2 Ft such that At At1 . In other words, letting
Wt .At1 / be the Rt1 .At1 / -valued vector whose components are given by the left-
hand side of (11.19), for all At 2 Ft with At At1 , it holds that
Wt .At1 / D Pt .At1 /t .At1 /:
Again, since the event At1 is arbitrary, we can define an Ft1 -measurable N-
>
dimensional random vector t such that Wt D ct C tC1 St D t> .St C Dt / holds
with probability one. In view of Definition 6.1, we have thus constructed a trading
strategy which finances the consumption plan c, thus proving the claim. In view
of (6.1), the initial wealth x needed to finance c is simply given by x D c0 C 1> S0 .
Solution of Exercise 6.13 The claim can be proved by relying on arguments
analogous to those used in the proof of Proposition 4.4. For every i D 1; : : : ; I,
let the weight ai be equal to the reciprocal of the Lagrange multiplier i of
agent i’s optimal investment-consumption Problem (6.36) in correspondence of
the optimal consumption plan ci and define the representative agent’s utility
function as in (6.48)–(6.49). The allocation fci I i D 1; : : : ; Ig corresponds to an
equilibrium of the I agents economy and, hence, it is feasible. Moreover, in view of
conditions (6.44)–(6.45), it holds that
0 0
u1 c10 uI cI
0
D ::: D D q0 ;
1 I
0 0
u1 c1 .At / uI cI .At / qA
1
D ::: D D t t ;
I ı At
for all At 2 Ft and t D 1; : : : ; T. In turn, in view of equation (6.49), the above

conditions imply that the allocation fci I i D 1; : : : ; Ig defines the representative
agent’s utility function in correspondence of the endowment process e and weights
f1=i I i D 1; : : : ; Ig. This implies that the allocation fci I i D 1; : : : ; Ig and
the prices fq0 ; qAt I At 2 Ft ; t D 1; : : : ; Tg define a no-trade equilibrium in the
representative agent economy.
Solution of Exercise
P 6.14 In correspondence of a no-trade equilibrium, the prices of
the whole set of TtD1 t Arrow securities are described by equation (6.50). Hence,
since the endowment process can be expressed in terms of the Arrow securities
(together with units of the consumption good at the initial date t D 0, the price of
which is supposed to be normalized to q0 D 1), the value at t D 0 of the aggregate

endowment process e is given by
X
T X X
T X u0 .e.At // XT X
e0 C qAt e.At / D e0 C ıt At 0
e.A t / D e 0 C ıt At e0
tD1 At 2Ft tD1 At 2Ft
u .e0 / tD1 A 2F t t
X
T
1 ı TC1
D e0 ı t D e0 ;
tD0
1ı
P
since At 2Ft At D 1 for all t D 1; : : : ; T.
Solution of Exercise 6.15 Let c D .ct /tD0;1;:::;T 2 C0C .x/ and cQ D .Qct /tD0;1;:::;T 2
C0C .Qx/ be two consumption processes satisfying the assumptions of the proposition
Q Arguing by contradiction, suppose that xQ > x
and let x WD W0 ./ and xQ WD W0 ./.
and define the trading strategy D .Ot /tD0;1;:::;T and the consumption process cO D
O
.Oct /tD0;1;:::;T by
( (
0; for t D 0; xQ x; for t D 0;
Ot WD and cO t WD
t Qt ; for t D 1; : : : ; TI ct cQ t ; for t D1; : : : ; T:
By the self-financing condition (6.1), since .; c/ 2 A .x/ and .; Q cQ / 2 A .Qx/, it
O C
is straightforward to check that .; cO / 2 A .0/, so that cO 2 C0 .0/. Moreover, the
process cO is identically null at all dates t D 1; : : : ; T and satisfies cO 0 D xQ x > 0 at
the initial date t D 0. We have thus shown that cO is an arbitrage opportunity of the
second kind. The case xQ < x can be treated in an analogous way.
Consider now the second part of Proposition 6.18. Let .S; D/ be a couple of
price-dividend processes not admitting arbitrage opportunities of the first kind and
c D .ct /tD0;1;:::;T and cQ D .Qct /tD0;1;:::;T two consumption processes financed by
the strategies D .t /tD0;1;:::;T and Q D .Qt /tD0;1;:::;T , respectively, and such that
P.ct D cQ t / D 1 for all t D 0; 1; : : : ; T. Arguing by contradiction, suppose that
there exist a date t 2 f1; : : : ; Tg and an event At 2 Ft such that Wt ./ < Wt ./ Q
O O
holds on At . Similarly as above, define the trading strategy D .s /sD0;1;:::;T and the
consumption process cO D .Ocs /sD0;1;:::;T by
(
0; for s D 0; : : : ; t;
Os WD
.s Qs /1At ; for s D t C 1; : : : ; TI
8
ˆ
<0;
ˆ

for s D 0; : : : ; t 1;
Q
cO s WD Wt ./ Wt ./ 1At ; for s D t;
ˆ
:̂.c cQ /1 ; for s D t C 1; : : : ; T:
s s At
By the self-financing condition (6.1), since .; c/ 2 A .x0 / and .Q ; cQ / 2 A .Qx0 /, for
some x0 > 0 and xQ 0 > 0, it is straightforward to check that .; O cO / 2 A .0/, so that
C
cO 2 C0 .0/. Moreover, the consumption process cO is identically null except at date
t, when it takes the value .Wt ./Q Wt .//1A 0, with strict inequality holding in
t
correspondence of At . We have thus shown that cO is an arbitrage opportunity of the
O cO / is identically null if the event At does not
first kind. Note also that the strategy .;
occur at date t. The case Wt ./Q < Wt ./ can be treated in an analogous way.
Solution of Exercise 6.16 For t D T, the equality WT ./ D cT holds by definition.

By induction, for any t 2 f0; 1; : : : ; T 1g, suppose that P.Ws ./ cs / D 1 holds
for all s D t C 1; : : : ; T and that Wt ./ < ct holds in correspondence of some event
At 2 Ft . Define the consumption process cO D .Ocs /sD0;1;:::;T and the trading strategy
O D .Os /sD0;1;:::;T by
8
ˆ (
< ct Wt ./ 1At ;
ˆ for s D t;
tC1 1At ; for s D t C 1;
cO s WD W ./1At ; for s D t C 1; Os WD
ˆ tC1 0; otherwise:
:̂0; otherwiseI
Note that the process cO is non-negative, since P.WtC1 ./ ctC1 0/ D 1, and cO t is
strictly positive on At . Moreover, it can be easily checked that .; O cO / 2 A .0/. In fact,
the self-financing condition (6.1) is satisfied for all s 2 f0; : : : ; t 1; t C 1; : : : ; Tg,
while at date t it holds that

OtC1
> >
St D 1At tC1 O cO t :
St D 1At Wt ./ ct D Oct D Wt ./
Note that the strategy .;O cO / is identically null if the event At does not occur at date
O
t. Since W0 . / D 0, we have shown that cO 2 C0C .0/ is an arbitrage opportunity, thus
proving the claim.
Consider now the second part of Lemma 6.19. Suppose that the price-dividend
couple .S; D/ does not admit arbitrage opportunities and let .; c/ 2 A .x0 /. Let
t 2 f0; : : : ; T 1g and suppose that there exists a date s 2 ft C 1; : : : ; Tg and an
event As such that cs .As / > 0. Arguing as in the proof of Lemma 6.19, suppose that
there exists an event At 2 Ft such that As At and Wt ./.At / D ct .At / and define
the consumption process cO D .Ocs /sD0;1;:::;T and the trading strategy O D .Os /sD0;1;:::;T
by
( (
0; for s D 0; : : : ; t; 0; for s D 0; : : : ; t;
cO s WD Os WD
cs 1 A t ; for s D t C 1; : : : ; TI s 1 A t ; for s D t C 1; : : : ; T:
Observe that on the event At it holds that
OtC1
> >
St D tC1 O cO t ;
St D Wt ./ ct D 0 D Wt ./
O cO / 2 A .x0 /, with
so that the self-financing condition (6.1) is satisfied and .;
O
W0 ./ D 0. We have thus constructed an arbitrage opportunity (of the first kind),
thus yielding a contradiction.
Solution of Exercise 6.17 In view of Theorem 6.23, the absence of arbitrage
opportunities is equivalent to the existence of a risk neutral probability measure
P . Equation (6.54) of Proposition 6.22 implies that, for each s 2 f0; 1; : : : ; Tg,
" T ˇ # " T ˇ #
X cr ˇ X cr ˇ
Ws ./ D E ˇ ˇ
rs ˇFs D cs C E rs ˇFs cs ;
rDs
rf rDsC1
rf
thus proving the first statement of Lemma 6.19. The second statement immediately
follows by noting that the last conditional expectation is strictly positive if there
exist a date r 2 fs C 1; : : : ; Tg and an event Ar 2 Fr such that cr > 0 holds on Ar .
Solution of Exercise 6.18 .i/ ) .ii/: let c 2 C0C .0/ be an arbitrage opportunity for
.S; D/, in the sense of Definition 6.17. In view of Lemma 6.19 and of the following
discussion, if there exist a date t 2 f0; 1; : : : ; Tg and an event At 2 Ft such that
Wt ./ < ct holds on At , then there exists an arbitrage opportunity (in the sense of
Definition 4.17) in the trading period Œt; t C 1 and, hence, there is nothing to prove.
Therefore, suppose that P.Wt ./ ct 0/ D 1 for all t D 0; 1; : : : ; T and let
˚
WD min t 2 f0; 1; : : : ; Tg W P Wt ./ > 0 > 0 :
Since c 2 C0C .0/ and c is non-null, is well defined and takes values in f1; : : : ; Tg.
We now claim that the strategy N WD realizes an arbitrage opportunity in the
trading period Œ 1; with respect to the couple .S 1 ; P /. Indeed, it holds that
N > S 1 D > S 1 D W 1 ./ c 1 D 0 and N > P D W ./ 0 with strict
inequality holding on some event with strictly positive probability.
.ii/ ) .i/: suppose that there exists a date t D f0; 1; : : : ; T 1g such that the
couple .St ; PtC1 / admits an arbitrage opportunity, in the sense of Definition 4.17.
This means that there exists a vector N 2 RNC1 such that N > St 0 and N > PtC1 0,
with at least one of the two inequalities being strict on some event with strictly
positive probability. Define then the trading strategy D .s /sD0;1;:::;T by tC1 WD N
and s WD 0 for all s ¤ t C 1. Define also the consumption process c D .cs /tD0;1;:::;T
by
8
ˆ N>
< St ;
ˆ if s D t;
cs WD N > PtC1 ; if s D t C 1;
ˆ
:̂0; otherwise:
It can be easily checked that the self-financing conditions (6.1)–(6.2) are satisfied, so
that .; c/ 2 A .0/. Since c 2 C0C .0/ and the consumption process c is non-negative
and non-null, we have thus constructed an arbitrage opportunity in the sense of
Definition 6.17.
Solution of Exercise 6.19 Equation (6.53) easily follows from condition (6.52) in
Definition 6.21. Indeed, it holds that

ˇ ˇ
1 n 1 1 h n n ˇ
i
n ˇ
ˇ
st D E stC1 C dtC1 jFt D E
n n
E stC2 C dtC2 ˇFtC1 C dtC1 ˇFt
rf rf rf
" ˇ # " T ˇ #
snT XT
dsn ˇˇ X dn ˇ
s ˇ
D ::: D E C Ft D E Ft ;
rfTt sDtC1 rfst ˇ rst ˇ
sDtC1 f
where we have used the tower property of the conditional expectation (under the risk

Pt P ) and
neutral probability measure the assumption that snT D 0. In order to prove
that the process .st =rf C rD0 .dr =rf //tD0;1;:::;T is a martingale under the probability
n t n r
measure P , let s; t 2 f0; 1; : : : ; Tg with s < t. Then, in view of equation (6.52) in

Definition 6.21, it holds that
" ˇ # " ˇ #
snt Xt
drn ˇˇ sns Xs
drn Xt n
snr snr1 ˇˇ
dr
E C Fs D s C C E C r r1 ˇFs
rft rr ˇ
rD0 f
rf rr
rD0 f
rfr rf rf
rDsC1
" " ˇ #ˇ #
sns Xs
drn Xt n
snr snr1 ˇˇ ˇ
dr
D s C r
C
E E r
C r r1 ˇFr1 ˇˇFs
rf r
rD0 f
rf rf rf
rDsC1
sns X dn
s
D s C r;
r
rf r
rD0 f
thus establishing the martingale property.

Finally, in order to prove equation (6.54), it suffices to prove that, for all t D
0; 1; : : : ; T 1, it holds that

E ŒWtC1 ./jFt D rf Wt ./ ct :
The latter condition easily follows from the self-financing condition (6.4) together
with the risk neutral condition (6.52). In fact, recalling that tC1 is Ft -measurable,
for all t D 0; 1; : : : ; T 1,

E ŒWtC1 ./jFt D Wt ./ C tC1
>
E ŒStC1 C DtC1 jFt St ct
>
D Wt ./ C rf 1 tC1 S t ct
>
D rf tC1 St D rf .Wt ./ ct / ;
where the last two equalities follow from the self-financing condition (6.1).
Solution of Exercise 6.20 Using equation (6.61), it suffices to compute

P
P .At \ As / P .At / !2At !
P .At jAs / D D D P
P .As / P .As / !2As !

P
i0 i

!2A u cT .!/ !
D P t i0 i
!2As u cT .!/ !

i0 i
P 0
ui ciT .!/ !
At u ct .At / !2At i0 i

u ct .At / At
D
P 0
ui ciT .!/
0
As u cs .As /
i i
!2As i0 i
!
u cs .As /
A s
i0

At u ci t .At / .ırf /
tT
D
As ui0 cis .As / .ırf /
sT

i0 i

ts u ct .At /
D .ırf / At jAs
ui0 ci
s .As /
where the second last equality follows from the fact that
0 0
ui .ci
t / D ırf EŒu .ctC1 /jFt ;
i i
for all t D 0; 1; : : : ; T 1 (see Corollary 6.5 and the following discussion).

Solution of Exercise 6.21 Let P;1 and P;2 be two risk neutral probability measures
and let c 2 C0C .x/, for some x 2 RC . The claim follows from the simple observation
that
XT
1 ;1 XT
1 ;2
c0 C t E Œc t D x D c 0 C E Œct :
r
tD1 f
rt
tD1 f
Solution of Exercise 6.22 It suffices to compute, using the predictability of the

process .t /tD1;:::;T ,
X
N
XN

E tn gnt jFt1 D tn E gnt jFt1
nD1 nD1
X
ˇ
N
1 1 n ˇ
D tn t1
E .dt C snt / snt1 ˇˇFt1 D 0;
nD1
rf rf
where the last equality follows from Definition 6.21. The remaining part of the
exercise simply follows by summing over all dates and taking the risk neutral
(unconditional) expectation.
Solution of Exercise 6.23 Suppose first that c 2 C0C .x/, so that there exists a
trading strategy D .t /tD0;1;:::;T with W0 ./ D x such that .; c/ 2 A .x0 /.
In particular, recall that the process is predictable. Recalling that WT ./ D cT ,
Exercise 6.2 implies that
X
T X
T1
N T ./ D x C
cN T D W t> . SN t C D
N t/ cN t :
tD1 tD0
Moreover, in view of equation (6.54) of Proposition 6.22, by taking the expectation

with respect to the risk neutral probability measure P we get
X
T
x D W0 ./ D E ŒNct :
tD0
This shows that any process belonging to C0C .x/, for some x 2 RC , satisfies the
requirements .i/ and .ii/. Let us now turn to the converse implication and suppose
that there exists a predictable process .t /tD0;1;:::;T satisfying condition .i/. As can be
easily verified, it holds that Ns0t C dN0t D 0, for all t D 1; : : : ; T. In turn, this means
that the first component .t0 /tD0;1;:::;T (representing the investment in the risk free
security) does not play a role in requirement .i/. Therefore, we can freely modify
the process .t0 /tD0;1;:::;T in order to construct a self-financing trading-consumption
strategy which finances the consumption plan c. To this effect, let us define the
process Q D .Qt /tD0;1;:::;T by Qtn WD tn for all n D 1; : : : ; N and t D 0; 1; : : : ; T,
while the process .Qt0 /tD0;1;:::;T will be chosen in order to satisfy the self-financing
condition (expressed in terms of discounted quantities, as in Exercise 6.2):
X
N X
N
QtC1
0
C QtC1
n
sNnt D Qt0 C Qtn sNnt cN t ; for all t D 0; 1; : : : ; T 1:
nD1 nD1
This implies that
X
N X
N
Q00 WD x 0n sn0 and Qt0 WD Qt1
0
.tn t1
n
/Nsnt1 cN t1 ;
nD1 nD1
for all t D 1; : : : ; T. By construction, the process .Qt0 /tD0;1;:::;T is predictable and

satisfies the self-financing condition. Since the self-financing condition is invariant
with respect to discounting (see Exercise 6.2), this shows that .; Q c/ 2 A .x0 /.
Q C
Moreover, since W0 ./ D x, it follows that c 2 C0 .x/, thus completing the
argument.
Solution of Exercise 6.25 Due to Theorem 6.23, the absence of arbitrage opportuni-
ties is equivalent to the existence of a risk neutral probability measure P (note that,
for the purposes of this exercise, the choice of the specific risk neutral probability
measure is irrelevant and, actually, we can allow agents to exhibit heterogeneous
beliefs as long as they all belong to the family of risk neutral probability measures).
For each i D 1; : : : ; I, consider a risk neutral agent i with probability beliefs
represented by the probability measure P and discount factor ı D 1=rf . For such
an agent, the optimal investment-consumption Problem (6.6) can be represented as
!
XT
1 i
max ci0 C E Œct ;
rt
tD1 f
where the maximization is with respect to all . i ; ci / 2 A such that 0i D N0i or,
equivalently, with respect to all consumption processes ci 2 C0C .xi /, with xi D
.N0i /> P0 . However, for any ci 2 C0C .xi /, Proposition 6.22 implies that
XT
1 i
ci0 C t E Œct D x :
i
tD1
rf
In particular, this implies that an optimal solution is simply given by consuming the
dividends generated by the initial endowment of securities. By Definition 6.13, such
a solution corresponds to a Radner equilibrium of the economy.
Solution of Exercise 6.27 Observe first that, due to Theorem 6.23, the absence of
arbitrage opportunities is equivalent to the existence of a risk neutral probability
measure P . Observe then that, in view of Exercise 6.23, the maximization in
Problem (6.6) can be equivalently formulated with respect to all processes belonging
to K C (recall that K C is the set of all non-negative stochastic processes adapted
to the information flow F) which satisfy requirements .i/ .ii/ of Exercise 6.23. In
particular, since the set
( )
X
T X
T1
c2K C
W cN T D x C t> . SN t N t/
CD cN t ; for some predictable .t /tD0;1;:::;T
tD1 tD0
is closed (due to the standing assumption of a finite probability space), the set of
all processes in K C satisfying requirements
P .i/ .ii/ of Exercise 6.23 is also
closed. Moreover, the condition TtD0 E ŒNct D x (together with the finiteness
of the probability space) implies that the same set is also bounded and, hence,
compact. The claim now follows from the fact that any continuous function admits
a maximum over a compact set.
Solution of Exercise 6.28 It suffices to combine equations (6.69) and (6.71).
w 6.29 Let t 2 f1; : : : ; Tg and consider the problem of
minimizing Corr rtC1 ; u0 .etC1 /jFt over all Ft -measurable vectors wtC1 2 RNC1
PN
satisfying nD0 wntC1 D 1. Observe that
>
Corr rtC1 wtC1 ; u0 .etC1 /jFt
hP i
E N
nD1 wntC1 .rtC1
n
EŒrtC1
n
jFt / . u0 .etC1 / EŒu0 .etC1 /jFt / jFt
D r P p ;
N 0
nD1 wtC1 rtC1 jFt Var . u .etC1 /jFt /
n n
Var
so that the minimization problem only depends on the last N components of the
vector w, since security 0 is risk free. The first order conditions of the minimization
problem give
n q >
Cov rtC1 ; u0 .etC1 /jFt Var rtC1 wO tC1 jFt
> > 12 n
Cov rtC1 wO tC1 ; u0 .etC1 /jFt Var rtC1 wO tC1 jFt Cov rtC1 >
; rtC1 wO tC1 jFt D 0;
for all n D 1; : : : ; N, thus proving relation (6.76).

Solution of Exercise 6.30 Since the couple .log mt ; log rtn / conditionally on Ft1
is distributed according to a bivariate normal distribution with mean n 2 R2 and
covariance ˙ n 2 R22 , it holds that
n
˙ n C 2˙12
n
C ˙22
n
log EŒmtC1 rtC1
n
jFt D log EŒelog mtC1 Clog rtC1 jFt D n1 Cn2 C 11 :
2
Since EŒmtC1 rtC1

n
jFt D 1, in view of equation (6.70), this implies that
˙11
n
C 2˙12
n
C ˙22
n
EŒlog mtC1 jFt C EŒlog rtC1
n
jFt D n1 C n2 D ;
2
for all n D 0; 1; : : : ; N. In particular, taking the risk free security (i.e., n D 0), this
gives
1 ˙n
log rf D EŒlog mtC1 jFt Var .log mtC1 jFt / D n1 11 :
2 2
Formula (6.78) follows by combining the last two relations.
Solution of Exercise 6.31 Consider the following two strategies:
(A) at date t do nothing, at date t C 1 buy one unit of the zero-coupon bond with
maturity t C 2, paying the price B.t C 1; t C 2/;
(B) at date t, buy one unit of the zero-coupon bond with maturity t C 2, paying the
price B.t; t C 2/ and sell short B.t; t C 2/=B.t; t C 1/ units of the zero-coupon
bond with maturity t C 1.
Both strategies yield the risk free payoff 1 at date t C 2, with a zero net investment
at the initial date t. By the Law of One Price and comparing the payoff of strategies
A and B at the intermediate date t C 1, it follows that
B.t; t C 2/
D E ŒB.t C 1; t C 2/jFt ;
B.t; t C 1/

Solution of Exercise 6.32 Equation (6.80) implies that

ˇ
1 etC ˇˇ
log 1 C i.t; t C / D log ı log E ˇFt
et
1 ˇ
D log ı log E g.t; / ˇFt

log ı C EŒlog g.t; /jFt ;
where we have used the first order approximation

ˇ
log E g.t; / ˇFt EŒlog g.t; /jFt :
Solution of Exercise 6.34 Equation (6.79) implies that, for any t 2 N,

et ˇˇ 1 ˇˇ ı 1
B.t; t C 1/ D E ı ˇFt D E ı ˇFt D E ;
etC1 utC1 u1
where we have used the assumption that the random variables .ut /t2N are i.i.d.
To prove the second equality, note that, due to the independence of the random
variables composing the sequence .ut /t2N , the covariance term in equation (6.83) is
null. Hence:

2
ı2 1
B.t; t C 2/ D B.t; t C 1/EŒB.t C 1; t C 2/jFt D E :
2 u1
Solution of Exercise 6.35 In view of equation (6.71), the equilibrium risk free
interest rate rf .l/ corresponding to the event fxt D lg, for t 2 N, is given by
1 1
rf .l/ D D
ıEŒx˛
tC1 jxt D l ı.lh h˛ C ll l˛ /
and, similarly, in the alternative state fxt D hg, for t 2 N,
1 1
rf .h/ D D :
ıEŒx˛
tC1 jxt D h ı.hl l˛ C hh h˛ /
To answer the second question, equation (6.84) with D 1 ˛ gives

.l/ D ı lh h1˛ .1 C
.h// C ll l1˛ .1 C
.l// ;

.h/ D ı hh h1˛ .1 C
.h// C hl l1˛ .1 C
.l// ;
where we recall that

.k/ denotes the value of the ratio st =et in correspondence of
the event fxt D kg, with k 2 fl; hg and for every t 2 N. The above linear system
admits a unique solution .
.l/;
.h// which can be easily computed. The price at
any date t 2 N of the security delivering the aggregate endowment can be expressed
as
st D et .
.l/1fst =et Dlg C
.h/1fst =et Dhg /:
Finally, to answer the last question, the expected return of the security can be
computed as, for each t 2 N,

stC1 C etC1 ˇˇ
EŒrtC1 jFt D E ˇFt
st
1
D lh h.1 C
.h// C ll l.1 C
.l// 1fxt Dlg

.l/
1
C hl l.1 C
.l// C hh h.1 C
.h// 1fxt Dhg :

.h/

(i): It suffices to use formula (6.86), noting that the risk neutrality of the represen-
tative
P1 agent implies that his marginal utility is constant and using the fact that
sD1 ı s
D ı=.1 ı/.
(ii): The second claim can be shown by an analogous argument.
(i) Note first that EŒdtC1 jFt D dN C %.dt d/
N and, hence,

EŒdtC2 jFt D dN C % dN C %.dt d/N dN D .1 %2 /dN C %2 dt :
Iterating the same argument and applying formula (6.86), we obtain that, for
all t 2 N,
X1 X1
ı ı% ı%
st D ı s .1 %s /dN C ı s %s dt D dN C dt :
sD1 sD1
1 ı 1 ı% 1 ı%
(ii): This can be shown by similar arguments, noting that

1
X
ı s ˇ s D ıˇ=.1 ıˇ/:
sD1
(iii): Note first that, by the tower property of the conditional expectation,

2 2
C 2
EŒdtCs jFt D E EŒdtCs jFtCs1 jFt D E eClog dtCs1 C 2 jFt D : : : D dt e
s
;
for all t 2 N and s 2 N. The claim then follows by a simple application of

formula (6.86).
6.38 The first claim follows from the martingale property
of the process ı t u0 .et /ˇt t2N together with the fact that a bubble is always non-
negative. Indeed, if ˇt D 0 for some t 2 N, then
EŒu0 .etCs /ˇtCs jFt D ı s u0 .et /ˇt D 0;
for all s 2 N. In turn, since the utility function u is assumed to be strictly increasing,
this implies that ˇtCs D 0, for all s 2 N.
To prove the second claim, note that equation (6.87) implies that, for each t 2 N,
ıT 0 ıT
ˇt D lim EŒu .e tCT /stCT jFt K lim EŒu0 .etCT /jFt
T!1 u0 .et / T!1 u0 .et /
D K lim B.t; t C T/ D 0;
T!1
where we have used equation (6.79). Since a rational bubble is always non-negative,
this shows that ˇt D 0, thus proving the claim.
Solution of Exercise 6.39 In view of Proposition 6.40, it suffices to verify that the
process .ı t u0 .et /ˇt /t2N is a martingale. This can be shown as follows:
ıEŒu0 .etC1 /ˇtC1 jFt D ıcEŒdtC1

jFt
2 2
D ıcEŒe.Clog dt C"tC1 / jFt D ıce.Clog dt /C 2 D u0 .et /ˇt ;
for all t 2 N.
Solution of Exercise 7.1 Let us define sFt WD EŒset jFt and sG t WD EŒst jGt , for
e
all t 2 N. Observe also that, since Ft Gt and due to the tower property of the
conditional expectation, it holds that sFt D EŒsG
t jFt . This implies that
G G F
Cov sG F F F F F
t st ; st D E .st st /st E st st E st D 0;
which in turn implies that

Var sG
t D Var sG F F
t st C st
F F
D Var sG F
t st C Var st Var st ;
thus showing inequality (7.8). The second inequality follows from (7.8) together
with equations (7.6) and (7.7).
Solution of Exercise 7.2 Note first that set st D set st C .st st / and, since
f f
set st D ut (see equation (7.3)) with EŒut jFt D 0, for all t 2 N, it holds that
EŒ.set st /.st st /jFt D 0, for all t 2 N. In turn, this implies that
f
EŒ.set st /2 jFt D EŒ.set st /2 jFt C EŒ.st st /2 jFt ;

f f
thus proving both inequalities in (7.11).

Solution of Exercise 7.3 Note that, as in decomposition (7.3), it holds that
X1 X1
1 F 1 G
sFt D set s "tCs and sG
t D st
e
s "tCs ; for all s 2 N:
r
sD1 f
r
sD1 f
Hence,
2 !2 3
X1
1 F
E4 s "tCs
5 D E .sFt set /2 D E .sFt C sG G e 2
t st st /
r
sD1 f
2 !2 3
X1
1 G
D E 4 sFt sG
t " 5
rs tCs
sD1 f
2 !2 3
X1
F 1
D E .st sG 2
t / CE
4 G 5
s "tCs
sD1
rf
2 !2 3
X1
1 G
E4 s "tCs
5;
r
sD1 f
P1 1 G
where we have used the fact that the quantities sFt sG t and sD1 rfs "tCs are
P1 1 G
uncorrelated, since EŒ sD1 rs "tCs jGt D 0, for all t 2 N. Then, noting that
f
"Ft D sFt C dt rf sFt1 D EŒsFt C dt jFt EŒsFt C dt jFt1

D EŒset C dt jFt EŒset C dt jFt1 ;
and similarly for "G F G

t , it follows that the sequences ."tCs /s2N and ."tCs /s2N are
composed of uncorrelated random variables, so that
2 !2 3
X1
1 F 2 1
E ."t / D E 4 F
s "tCs
5
rf2 1 r
sD1 f
2 !2 3
X1
1 G 1
E4 " 5D E ."G 2
t / ;
rs tCs
sD1 f
rf2 1
thus proving inequality (7.12), under the standing assumption that the dividend
process is adapted to both information flows F and G.
Solution of Exercise 7.5 As a preliminary, equation (7.17) can be rewritten in terms
of de-trended quantities as
X
N
dNtC1 D dNt C k .EN tkC1 EN tk /;
kD0
where k WD k =.1 C g/kC1 . Then, the assumption that Et D rf Vt , for all t 2 N,

implies that
X
N
dNtC1 D dNt C rf k .VN tkC1 VN tk /
kD0
X
N
D dNt C rf ı k .VN tkC1 VN tk /;
kD0
where we define VN t WD Vt =.1 C g/t , for all t 2 N, and
X
N
k
ı WD k and k WD ; for all k D 0; 1; : : : ; N:
kD0
ı
By summing over t in the last relation and using the assumption that d0 D 0, we
obtain
X
N
dNt D rf ı k VN tk :
kD0
Then, the assumption that stocks are priced rationally (i.e., st D Vt ) implies that
X
N
dNt D rf ı k sNtk ;
kD0
where sNt WD st =.1 C g/t denotes the de-trended price, for all
Pt 2 N. Hence, in view
of equation (7.10) together with the assumption that sNeT WD T1 Nt =T, it holds that
tD0 s
(in de-trended terms)
X
Tt1
1 N 1 X
T1
sOet D d tCs C sNt
sD1
rfs TrfTt tD0
X
Tt1
1 X
N
1 X
T1
D rf ı N
s
k tCsk C sNt
sD1
rfs kD0 TrfTt tD0
X
T1
DW wtk sNk ;
kDN
with wtk D 0 for k < t N.

Solution of Exercise 7.6 In order to prove part (i), it suffices to note that (assuming
without loss of generality that k < t)
ˇ
Cov.rtCs ; rt jKk / D Cov.ztCs ; zt jKk / D EŒztCs zt jKk D E EŒztCs jKt zt ˇKk D 0;
where the last equality follows from equation (7.15), since

ˇ
EŒztCs jKt D E EŒztCs jFt ˇKt D 0:
To prove part (ii), note that the value based on the information set Kt of a Kt -
measurable trading strategy tC1 is given by (compare with equation (6.68) and
recall the assumption of risk neutrality):
1 XN
tC1
n ˇ
>
EŒtC1 .StC1 C DtC1 /jKt D E EŒsntC1 C dtC1
n
jFt ˇKt
rf nD0
rf
X
N X
N
D tC1
n
EŒsnt jKt D tC1
n
snt ;
nD0 nD0
for all t 2 N, where we have used the fact that snt is Kt -measurable, for all n D
>
0; 1; : : : ; N and t 2 N. Provided that tC1 St > 0, the corresponding return of the
> >
tC1
strategy is rtC1 WD tC1 .StC1 C DtC1 /=.tC1 St /. Therefore, in view of the previous
arguments, the Kt -conditional expected return is equal to

ˇ
tC1
>
tC1 E EŒStC1 C DtC1 jFt ˇKt >
tC1 EŒSt jKt
EŒrtC1 jKt D >
D rf >
D rf ;
tC1 St tC1 St
where we have used the assumption that St D .s0t ; s1t ; : : : ; sNt /> is Kt -measurable.
Solution of Exercise 7.7 In equilibrium, the aggregate consumption coincides with
the aggregate dividend, so that the key asset pricing relation (6.68) implies that the
price process .st /t2N satisfies
C1
st et D ıEŒstC1 etC1 jFt C ıEŒetC1 jFt ; for all t 2 N:
Let us now replace a candidate solution of the form st D %.Xt /et into the above
relation:
C1 C1
%.Xt /et D ıE %.XtC1 / C 1 etC1 jFt :
Replacing into the last expression the explicit form of the endowment process, we
get
2 2 =2C˛ .1C /X
%.Xt / D ıe˛0 .1C /C.1C / 1 t
EŒ%.XtC1 /jFt C 1 :
Now, since Xt can only assume the two possible values f0; 1g, the last equation
yields a system of two linear equations with unknowns %.0/ and %.1/, which can be
explicitly computed as stated in the exercise.
Solution of Exercise 7.8 Note first that, due to the time additivity of the utility
function U, it holds that
@U
D ı t u0t .ct /; for every t D 0; 1; : : : ; T:
@ct
Moreover, by computing the total derivative,
1 1
d log.ct =cs / D dct dcs ; for every s; t D 0; 1; : : : ; T;
ct cs
and similarly

@U @U u00s .cs / u00t .ct /
d log = D dc s dct ; for every s; t D 0; 1; : : : ; T:
@cs @ct u0s .cs / u0t .ct /
Hence, by the definition of U .s; t/:
1 @ct 1 @cs
ct @U cs @U
U .s; t/ D u00 u00
s .cs / @cs t .ct / @ct
u0s .cs / @U
u0t .ct / @U
1 1
ct ı t u0t .ct /
cs ı s u0s .cs /
D u00 u00
s .cs / t .ct /
ı s .u0s .cs //2
ı t .u0t .ct //2
ct ı ts u0t .ct /

1 cs u0s .cs /
D ;
ct ı ts u0t .ct / r
rur t .ct / cs u0s .cs / rus .cs /
thus proving that (7.19) holds. The special case of CRRA utility functions immedi-
ately follows.
Solution of Exercise 7.9 As a preliminary, observe that the representative agent’s
stochastic discount factor process .Mt /t2N satisfies

(
MtC1 ı etC1 eR ; if there is no disaster at t C 1;
D mtC1 D e D
Mt et eR BtC1 ; otherwise;
where mtC1 is the one-period stochastic discount factor between t and t C 1.

Moreover, in view of the hypotheses of the model, we obtain

dtC1 ˇˇ
E mtC1 ˇFt
dt

RCgd
De EŒ.1 C "tC1 /1fno disaster at tC1g jFt

C EŒ.1 C "tC1 /BtC1 FtC1 1fdisaster at tC1g jFt

D eRCgd .1 pt / C pt EŒBtC1 FtC1 jfthere is a disaster at t C 1g _ Ft
D eRCgd .1 C Ht /;
where Ht is the asset resilience introduced above. Note first that, due to the key asset
pricing relation (6.68), the cum-dividend equilibrium price pt of the security paying
the dividend process .dt /t2N satisfies
pt D dt C E ŒmtC1 ptC1 jFt ; for all t 2 N:


Let us conjecture a solution structure of the form pt D dt a C b.Ht H / , for all
t 2 N. By substitution into the last relation and using the hypotheses of the model,
we can compute:

dtC1 ˇˇ
a C b.Ht H / D 1 C E mtC1 a C b.HtC1 H / ˇFt
dt

RCgd 1 C H
D1Ce .1 C Ht / a C be .Ht H /
1 C Ht

D 1 C eRCgd a 1 C H C .Ht H / C be .1 C H /.Ht H / ;
for all t 2 N. Since the above equation has to hold for all t 2 N and for all possible
values of Ht , we can use a separation of variables argument and solve for a and b,
thus yielding
1 eRCgd a
aD and bD ;
1 er 1 er
thus proving the representation (7.20).
Solution of Exercise 8.1 The claim simply follows from the fact P that U ./ is
defined as the upper envelope of the family of linear functions 7! SsD1 s u.xs /,
with .x1 ; : : : ; xS / 2 B.
Solution of Exercise 8.2 Part (i) directly follows from the first order condition of the
optimal saving problem (8.42), since it must hold that u0 .w0 ˛0 / D u0 .˛0 ˛1 /,
which in turn implies that ˛1 D 2˛0 w0 because the function u0 ./ is strictly
decreasing. Let us then denote
H.˛0 / WD 2u.w0 ˛0 / C EŒu.2˛0 w0 C xQ /; (11.20)
which corresponds to (8.42) when consumption is smoothed over the first two dates
and xQ is observed only at t D 2. In order to prove part (ii), observe first that, by a
similar reasoning, it is optimal for the agent to smooth consumption between date
t D 1 and date t D 2, so that the optimal saving problem (8.42) when xQ is observed
at t D 1 can be rewritten as

˛0 C xQ
max u.w0 ˛0 / C 2E u :
˛0 2R 2
In this case, the first order optimality condition is given by (8.43). Since the function
H introduced in (11.20) is strictly concave, in order to prove part (iii) it suffices to
show that H 0 .˛0i / 0, i.e.,
EŒu0 .2˛0i w0 C xQ / u0 .w0 ˛0i /: (11.21)
Define
3˛0i xQ
z WD w0 ˛0i and yQ WD w0 C :
2 2
With this notation, condition (11.21) holds if and only if
EŒu0 .z C 2Qy/ u0 .z/ 0: (11.22)
In turn, condition (8.43) can be equivalently rewritten as EŒu0 .z C yQ / u0 .z/ D 0.

Since the map k 7! EŒu0 .z C kQy/ is convex, as a consequence of the assumption that
u000 > 0, it follows that condition (11.22) necessarily holds, thus proving the claim.
Solution of Exercise 8.3 By definition, in order to show that the information
structure . a ; a / is finer than the information structure . b ; b /, we need to show
that condition (8.5) holds, for any utility function u W RC ! R. Observe that
X
J
XJ X
S
b .yj /U v b .yj / D b .yj / max vsb .yj /u.xs /
.x1 ;:::;xS /2B
jD1 jD1 sD1
X
J X
S X
J
D b .yj / max vsa .yk /Kkj u.xs /
.x1 ;:::;xS /2B
jD1 sD1 kD1
X
J X
J X
S
b .yj / Kkj max vsa .yk /u.xs /
.x1 ;:::;xS /2B
jD1 kD1 sD1
X
J X
S
D a .yk / max vsa .yk /u.xs /
.x1 ;:::;xS /2B
kD1 sD1
X
J

D a .yk /U v a .yk / ;
kD1
where the first and the last equalities follow from the definition of U .v ` .yj //,
for ` D a; b, the second and the third equalities follow from condition (8.6) and
the inequality in the middle uses the fact that the map 7! U ./ from S
onto R is convex in the vector of probabilities (see Exercise 8.1). This shows that
condition (8.5) holds, thus proving the claim.
Solution of Exercise 8.4 Let fxi 2 RSC I i D 1; : : : ; Ig be an ex-ante Pareto optimal

0
allocation and suppose that there exists a feasible allocation fxi 2 RSC I i D 1; : : : ; Ig
such that
X
S
0
X
S
vsi .yij /ui .xis / vsi .yij /ui .xis /; for all i D 1; : : : ; I and j D 1; : : : ; J;
sD1 sD1
with strict inequality holding for some i 2 f1; : : : ; Ig and j 2 f1; : : : ; Jg. Noting that
P J
jD1 .yj /vs .yj / D s , for every i D 1; : : : ; I and s D 1; : : : ; S, it holds that
i i i
X
S
0
X
J X
S
0
X
J X
S X
S
s u i
.xis / D .yij /vsi .yij /ui .xis / .yij /vsi .yij /ui .xis / D s ui .xis /
sD1 jD1 sD1 jD1 sD1 sD1
and, assuming that .yij / > 0, for all i D 1; : : : ; I and j D 1; : : : ; J, a strict inequality
holds for some i 2 f1; : : : ; Ig, thus contradicting the ex-ante Pareto optimality of the
0
allocation fxi 2 RSC I i D 1; : : : ; Ig. Hence, the allocation fxi 2 RSC I i D 1; : : : ; Ig
cannot be an interim Pareto improvement of fxi 2 RSC I i D 1; : : : ; Ig. Assuming that
all the elements of the matrix v of conditional probabilities are strictly positive, a
similar argument allows to show that every interim Pareto optimal allocation is also
ex-post Pareto optimal.
Solution of Exercise 8.5 In correspondence of a Green-Lucas equilibrium (see
Definition 8.6), every agent maximizes

max E dQ .Qy/ zi j .y/; yi ; for each i D 1; : : : ; I:
zi 2R
As a consequence, in correspondence of the optimal choice zi , for each i D

1; : : : ; I, it holds that

E dQ .Qy/ zi j .y/; yi 0:
By the law of iterated expectations together with the market clearing condition, it
holds that
X
I

E dQ .Qy/ zi j .y/ D 0:
iD1
In turn, this implies that EŒ.dQ .Qy//zi j .y/; yi D 0, for all i D 1; : : : ; I, i.e.,
the expected gain from trading is null. Clearly, letting zOi D 0, for all i D 1; : : : ; I,
satisfies the market clearing condition appearing in Definition 8.6 and, hence,
corresponds to another Green-Lucas equilibrium (without trade).
Solution of Exercise 8.6 Suppose first that zQ is a sufficient statistic for the
conditional density f .y; zjx/. Then, the Bayes rule implies that
g.x/f .y; zjx/

f .xjy; z/ D R 1
1 g.x/f .y; zjx/dx
g.x/g1 .y; z/g2 .z; x/

D R1
1 g.x/g1 .y; z/g2 .z; x/dx
g.x/g2 .z; x/
D R1 ;
1 g.x/g2 .z; x/dx
where g./ denotes the density function of the random variable xQ . This shows that
the conditional distribution of xQ given fQy D y; zQ D zg does not depend on y.
Conversely, if the conditional distribution of xQ given fQy D y; zQ D zg does not
depend on y, then it holds that
f .y; z; x/ f .xjy; z/h.y; z/ f .xjz/h.y; z/

f .y; zjx/ D D D D g1 .y; z/g2 .z; x/;
g.x/ g.x/ g.x/
with g1 .y; z/ WD h.y; z/ and g2 .z; x/ WD f .xjz/=g.x/ and where h.; / denotes the
density function of the pair of random variables .Qy; zQ/. We have thus shown that zQ is
a sufficient statistic for the conditional density f .y; zjx/.
Solution of Exercise 8.8 Let pa W Y ! RN be the equilibrium price functional of
the artificial economy and suppose that it is a sufficient statistic for the conditional
distribution of the random dividends of the N risky assets given the signal random
vector yQ . We want to prove that letting ./ WD pa ./ defines an equilibrium
price functional for the original economy (in the sense of Definition 8.6) which
is a sufficient statistic. In the setting of this exercise, the optimal choice problem
of agent i in the original economy where he observes his private signal together
with the vector of equilibrium prices can be formulated as follows (compare with
Section (3.1)):
" ˇ #
X
N
ˇ
max E ui wi0 rf C win .Qrn rf / ˇˇ yQ i ; .Qy/ ;
wi 2RN
nD1
where wi0 represents the initial wealth of agent i and wi 2 RN denotes the amounts of
wealth invested in the N risky assets, for i D 1; : : : ; I, and rQn represents the random
return of asset n, for n D 1; : : : ; N. Since .Qy/ D pa .Qy/ and using the tower
property of the conditional expectation, this problem can be rewritten as
"

X ˇ ˇˇ #
N
ˇ ˇ
max E E ui wi0 rf C win .Qrn rf / ˇˇ yQ i ; yQ ; pa .Qy/ ˇ yQ i ; pa .Qy/ :
wi 2RN ˇ
nD1
Since yQ i is an element of yQ and pa is a sufficient statistic for the conditional

distribution of the dividends on the N risky assets (and, hence, for the conditional
distribution of the returns as well), it holds that
" ˇ #
X
N
ˇ i
E u w0 rf C
i i
wn .Qrn rf / ˇˇ yQ ; yQ ; p .Qy/
i a
nD1
" ˇ #
X
N
ˇ a
D E u w0 rf C
i i i ˇ
wn .Qrn rf / ˇ p .Qy/ ;
nD1
so that
"

X ˇ ˇˇ #
N
ˇ i ˇ i a
E E u w0 rf C
i i i ˇ
wn .Qrn rf / ˇ yQ ; yQ ; p .Qy/ ˇ yQ ; p .Qy/
a
ˇ
nD1
" ˇ #
X
N
ˇ
D E ui wi0 rf C win .Qrn rf / ˇˇ pa .Qy/
nD1
" ˇ #
X
N
ˇ
D E u w0 rf C
i i
wn .Qrn rf / ˇˇ yQ ;
i
nD1
where the last equality uses again the fact that pa is a sufficient statistic. We have
thus shown that
" ˇ #
XN
ˇ i
max E u w0 rf C
i i
wn .Qrn rf / ˇˇ yQ ; .Qy/
i
wi 2RN
nD1
" ˇ #
X
N
ˇ
D max E u w0 rf C
i i
wn .Qrn rf / ˇˇ yQ :
i
wi 2RN
nD1
Hence, for every agent i D 1; : : : ; I, the optimal choice problem in the artifi-
cial economy corresponds exactly to the optimal choice problem in the original
economy. Since market clearing holds, WD pa defines an equilibrium price
functional which is a sufficient statistics. The converse implication can be proved
by an analogous argument.
Solution of Exercise 8.10 For every i D 1; : : : ; I, the optimal choice problem of
agent i can be formulated as
1X
S
max s .xis /˛ DW U.xi
1 ; : : : ; xS I /;
i
.xi1 ;:::;xiS /2RSC ˛ sD1
P P
subject to the budget constraint SsD1 ps eis D SsD1 ps xis . For this problem, the first
order optimality conditions imply
˛1
ps s xi
D s
; for all r; s D 1; : : : ; S:
pr r xi
r
Since markets are complete, the equilibrium allocation is ex-ante Pareto optimal. In
view of Proposition 4.5, there exists a sharing rule fyi W RC ! RC I i D 1; : : : ; Ig
s D Ki es , for some Ki > 0, for all i D 1; : : : ; I. Using the budget
such that xi
constraint together with the above optimality condition in correspondence of the
aggregate endowment, it follows that
X
S X
S
Ki s e˛s D s eis e˛1
s ; for all i D 1; : : : ; I;
sD1 sD1
so that
PS
s eis e˛1
Ki D sD1
PS
s
; for all i D 1; : : : ; I:
˛
sD1 s es
In turn, this implies that, for every i D 1; : : : ; I, the maximal expected utility of
agent i can be written as
PS !˛
1X
S i ˛1
rD1 r er er
1 ; : : : ; xS I /
U.xi D U.Ki e1 ; : : : ; Ki eS I / D s
i
PS es
˛ sD1 rD1 r er
˛
!˛ !1˛
1 X
S X
S
D s eis e˛1 s e˛s :
˛ sD1
s
sD1
Moreover, the function 7! U.xi 1 ; : : : ; xS I / is concave in the vector of

i
probabilities D .1 ; : : : ; S / (compare also with Exercise 8.1) and, clearly, is non-
linear with respect to if and only if the individual endowment eis is not proportional
to the aggregate endowment, for every s D 1; : : : ; S. The claim then follows by the
arguments analogous to those used in Exercise 8.3.
Solution of Exercise 8.11 The claim simply follows by inserting the equilibrium
price (8.20) into the optimal demand of an agent, thus yielding
1
wi D Q i ; .y1 ; : : : ; yI / .y1 ; : : : ; yI /rf
EŒdjy
Q i ; .y1 ; : : : ; yI //
ai 2 .djy
!1
1 X 1
I
D i ; for every i D 1; : : : ; I:
a iD1 ai
Solution of Exercise 8.12 Suppose that W f1; 2g ! R is a fully revealing

equilibrium price functional. Then, in correspondence of the equilibrium price
q D .y/, the optimal demand of the informed agent is given by
h ˇ i y .y/
Q ˇ
w1;y D arg max E ea.w.d .Qy/// ˇQy D y D ; for y D 1; 2:
w2R a 2
Since is a fully revealing equilibrium price functional, it perfectly reveals to

the uninformed agent the private information of the informed agent. Therefore, the
optimal demand in equilibrium of the uninformed agent is given by
h ˇ i
Q ˇ
w2;y D arg max E ea.w.d .Qy//CkQy p/ ˇQy D y
w2R
h ˇ i y .y/
Q ˇ
D arg max E ea.w.d .Qy//CkQy p/ ˇ .Qy/ D .y/ D ky ;
w2R a 2
for y D 1; 2. In correspondence of the equilibrium, the future market must clear, so

that
y .y/ y .y/
w1;y C w2;y D C ky D 0; for y D 1; 2:
a 2 a 2
The market clearing condition implies that the equilibrium price functional is
given by
2 ky
.y/ D y ; for y D 1; 2:
2
It is easy to see that, for the values chosen, it holds that .1/ D .2/ D 3, thus
contradicting the assumption that the equilibrium price functional is fully revealing.
Solution of Exercise 8.13 Similarly as in the proof of Proposition 8.11, we

conjecture that the equilibrium price functional PI isi linear with respect to the
Q
aggregate supply uQ and the average signal y WD iD1 yQ =I (and recall that yQ was the
sufficient statistic in the Grossman [839] model presented at the end of Sect. 8.2). In
other words, we start from the conjecture that admits the representation (8.23),
for some coefficients ˛, and to be determined. By exploiting the basic properties
of the normal multivariate distribution, it can be easily seen that the conditional
distribution of the random variable dQ given fQyi D yi ; eQ i D ei ; .y;
Q uQ / D g is again
normal with conditional mean and variance that can be explicitly computed. The
optimal demand of each agent can then be determined in terms of the conditional
mean and variance of d, Q similarly as in the proof of Proposition 8.11. Imposing
the market clearing condition then leads to a system of three non-linear equations
in three unknowns which can be explicitly solved, leading to the values reported
after (8.23) for the coefficients ˛, and (compare also with Huang & Litzenberger
[971, Chapter 9]).
Solution of Exercise 8.14 For brevity, we denote by p WD .d; u/, for d 2 R

and u 2 R, a generic realization of the random variable .d; Q uQ / representing
the candidate price functional, conjectured to be linear with respect to dQ and uQ .
Recall that, by the properties of the normal multivariate distribution, the conditional
Q i ; p and EŒdjp
expectations EŒdjy Q are linear functions of .yi ; p/ and p, respectively,
and the conditional variances Var.djy Q i ; p/ and Var.djp/
Q do not depend on the
realizations y and p. Note that EŒdjy ; p denotes the conditional expectation of dQ
i Q i
Q uQ / D pg (and similarly for the
given the observation of the event fQyi D yi ; .d;
conditional variance). Since the two classes of informed and uninformed agents are
composed of homogeneous agents (i.e., with the same preferences and with the same
type of information) and since the utility functions are negative exponential, the
optimal demands of the informed and of the uninformed agents in correspondence
of a price p and of a private signal yi are of the form
vspace*-3pt
winf .yi ; p/ D yi cinf p C bO inf ;

(11.23)
wun . p/ D cun p C bO un ;
for suitable coefficients , cinf , cRun , bO inf and bO un . The market clearing condition,

together with the assumption that 0 "Qi di D 0 almost surely, implies that
. d cinf p C bO inf / C .1 /.cun p C bO un / C u D 0;
for every realization d 2 R and u 2 R. Hence, the market clearing condition implies
that the equilibrium price functional necessarily satisfies

.d; u/ D d C u C bO ; for all d 2 R and u 2 R;
where bO WD bO inf C .1 /bO un and WD 1=.cinf C .1 /cun /, provided that

cinf C .1 /cun ¤ 0. We now conjecture that the optimal demands are of the
form, for p D .d; u/,
N
winf .yi ; p/ D .yi p/ binf . p d/;
(11.24)
N
wun . p/ D bun . p d/:
By comparing (11.23) with (11.24), it must hold that
cinf D C binf ; bO inf D binf d;

N cun D bun and bO un D bun d:
N
Q uQ / D EŒd
With this parametrization, it is easy to check that EŒ .d; Q D d,
N so
Q Q
that the conditional expectation EŒdjp and the conditional variance Var.djp/ are
respectively given by
1 2 u2
Q D dN C
EŒdjp p dN and Q
Var.djp/ D 2 :
2 2 2 2
C u 2 C
2 2 u2
(11.25)
Considering first the case of uninformed agents, due to the assumption of negative
exponential utility functions, their optimal demand necessarily satisfies
Q p
EŒdjp
wun . p/ D N
D bun . p d/: (11.26)
Q
aun Var.djp/
By substituting (11.25) into (11.26), we obtain that the parameter bun satisfies

1 1 2 2
b un
D un C :
a 2 u2 u2
Considering now the class of informed agents, we can perform a similar procedure.
First, by relying on the properties of the multivariate normal distribution, we can
Q
compute the conditional expectation and the conditional variance of d:

Q i ; p D 1 yi 1 2 2 u2 N u2
EŒdjy 2 2
C C dC p ;
1
C 1
C "2 2 u2
"2 2 u2
1
Q i ; p/ D 1 1 2 2
Var.djy C C :
"2 2 u2
The optimal demand of each informed agent necessarily satisfies
Q i ; p p
EŒdjy
winf . p/ D N
D .yi p/ binf . p d/:
un
a Var.djy Q i ; p/
By substitution, we obtain that the parameters and binf satisfy

1 1 1 2 2
D inf 2 ; b inf
D inf C
a " a 2 u2 u2

1 1 1 2 2
C binf D 2
C 2C 2
:
ainf " u u2
We can then solve for , obtaining

1 C C 1
ainf aun
1
u2
D > 0;
2 2
ainf C 1
aun
1
2 C 2
u
which in turn allows to explicitly compute the coefficients binf and bun as given in
Proposition 8.12.
Solution of Exercise 8.15 As follows from Exercise 8.14, the optimal demands of
the informed and of the uninformed agents in correspondence of the equilibrium
price p WD .d; u/ are respectively given by

11 1 2 2 Q i
winf; .yi ; p/ D 2
C 2
C 2
EŒdjy ; p p ;
"
ainf u

1 1 2 2 Q
wun; D un 2
C 2
EŒdjp p ;
a u
Q uQ / D pg. The market

for all i 2 Œ0; and for every realization fQyi D yi ; .d;
clearing condition (8.24) implies that
R
1 1 1 2 2 Q i ; pdi C 1 1 2 2 Q Cu
ainf "2
C 2
C u2
EŒdjy
0 a un 2 C u2 EŒdjp
.d; u/ D p D 2 2
;
2 2
ainf
1
"2
C 12 C 2 C 1aun
1
2 C 2
u u
thus proving the claim. In turn, the last expression can be easily seen to imply that
Q
if aun ! 0 then .d; u/ converges to EŒdjp.
Solution of Exercise 8.16 Following Grossman & Stiglitz [849, Appendix B], we
start by computing the optimal expected utility of an informed agent who buys the
information dQ o by paying the fixed cost c and can invest in the two traded securities,
with the price of the risky security being .dQ o ; uQ /:
h i h i i
E eae
inf; Q o Q o Q Qo
W inf;i;
1; D E ea .W0 c/rf Cw .d ; .d ;Qu//.d .d ;Qu/rf /
h inf; Q o Q o
i
Q Qo
D ea.w0 c/rf E eaw .d ; .d ;Qu//.d .d ;Qu/rf /
i
h h inf; Q o Q o ˇ i
Q Qo
D ea.w0 c/rf E E eaw .d ; .d ;Qu//.d .d ;Qu/rf / ˇ dQ o ; uQ ;
i
where the last equality follows by iterated conditioning. Using the conditional
moment generating function of a normal random variable, it follows that
h ˇ i
inf; Q o Q o Q Qo ˇ
E eaw .d ; .d ;Qu//.d .d ;Qu/rf / ˇ dQ o ; uQ
2
Q o ; .dQ o ;Qu//.dQ o .dQ o ;Qu/rf /C a .winf; .dQ o ; .dQ o ;Qu///2 "2
D eaw
inf; .d
2
2 dQo .dQo ;Qu/rf 2

1 h
dQ o .dQ o ;Qu/rf
p
De 2"2 De 2"2 h
;
where the second equality uses the explicit form of the optimal demand of an
informed agent (see Proposition 8.13) and where for brevity we let

h WD 2 dQ o j .dQ o ; uQ / ;
(recall also that .dQ o ; uQ / and .dQ o ; uQ / are informationally equivalent), so that
h i
h dQo .dQo ;Qu/rf 2
ae p
inf;i;
a.wi0 c/rf
E e W 1;
De E e 2"2 h
:
Recall now that, if a pair of random variables .Qz; yQ / follows a bivariate normal
distribution with 2 .QzjQy/ D 1, then
2 1 t 2
E etQz j yQ D p e 1C2t .EŒQzjQy/ :
1 C 2t
By applying this result to the previous relation, we get

h dQo .dQo ;Qu/rf 2 ˇ p 2
2 p ˇ Qo " 1
2.h C"2 /
EŒdQ j .dQ o ;Qu/ .dQ o ;Qu/rf
E e 2" h
ˇ .d ; uQ / D p e
h C "2
s 2
Q dQ o /
2 .dj 2 Q 1 Qo EŒdQ j .dQ o ;Qu/ .dQ o ;Qu/rf
D e 2 .dj .d ;Qu//
;
Q .dQ o ; uQ //
2 .dj
thus proving that

h ˇ i
E eae 1; ˇ .d
W inf;i;

Q o ; uQ /
s 2 (11.27)
a.wi0 c/rf
Q dQ o /
2 .dj 1
EŒdQ j .dQ o ;Qu/ .dQ o ;Qu/rf
De e 2 2 .dQj .dQo ;Qu// :
Q .dQ o ; uQ //
2 .dj
By performing analogous computations, it can be shown that the optimal expected

utility of an uninformed agent, conditionally on .dQ o ; uQ / is given by
h i 2
ˇ 1 Q Qo Qo
E eae 1; ˇ .d
W un;i;

Q o
; Q
u / D e awi0 rf 2 2 .dQ j .dQ o ;Qu// EŒdj .d ;Qu/ .d ;Qu/rf
e : (11.28)
Hence, by combining (11.27) and (11.28) and using iterated expectations, it follows
that
h i h ˇ i s
E u.We inf;i; / E E u. e inf;i; /ˇ .dQ o ; uQ /
W Q dQ o /
1; 1; 2 .dj
./ D h i D h ˇ i D e acrf ;
E u.We un;i; / E E u.W e un;i; /ˇ .dQ o ; uQ / Q .dQ o ; uQ //
2 .dj
1; 1;
thus proving the first representation in (8.34). The second representation given
in (8.34) simply follows by using relations (8.33). It remains to show that the func-
Q .dQ o ; uQ //
tion 7! ./ is strictly increasing. This follows since the quantity 2 .dj
is strictly decreasing with respect to , as shown in (8.33).
Solution of Exercise 8.17 Noting that

r
1Cm nm 1=2
./ D e acrf
D eacrf 1C
1 C m C mn 1Cm
and recalling that the function 7! ./ is increasing, it follows that increases
if and only if the ratio nm=.1 C m/ increases. Suppose that 2 .dQ o / increases while
the variance 2 remains constant (so that "2 D 2 2 .dQ o / decreases, i.e.,
the information accessible to the informed agents becomes more precise). Then,
differentiating the ratio nm=.1 C m/ with respect to "2 and keeping constant 2
(meaning that @ 2 .dQ o / D @"2 ) gives that
@ nm 1 m n m n m 2n m
D 2 C 2 :
@ 2 .dQ o / 1Cm " 1 C m .1 C m/2 2 .dQ 0 / " 1 C m .1 C m/2 "2
Simplifying this expression, it can be shown that

@ nm 1
sign D sign m 1 C 1
@ 2 .dQ o / 1 C m n
2acrf
e 1 nC1
D sign 1 ;
n e2acrf C 1 n
where the second equality follows from (8.35). Part (iv) of Proposition 8.15 then
follows by letting nN be the value satisfying the equation
e2acrf 1 nN
D :
nN e2acrf C 1 nN C 1
Solution of Exercise 8.18 Suppose that u ! 0. Then, for any fixed > 0, the
quantity m introduced in Proposition 8.14 converges to zero and, hence, ./ !
eacrf > 1 for u2 ! 0. Hence, any value > 0 cannot be an equilibrium if u2 ! 0
and it must hold that ! 0 as well. The limit of the quantity in the right hand
2 1=2
side of (8.34) for ! 0 and p u ! 0 is easily seen to be e .1 C n/
acrf
. Hence,
it follows that, if eacrf > 1 C n then D 0 defines an overall equilibrium.
Furthermore, it follows from (8.35) that the informativeness
p of the equilibrium price
remains unchanged. On the contrary, if eacrf < 1 C n, then there does not exist
an overall equilibrium, since .0/ < 1 and the map 7! ./ is discontinuous at
D 0.
It remains to prove part (ix) of the proposition. If "2 D 0, then nm D 0, so that
./ D eacrf > 1 for every > 0. Moreover, if "2 D 0 then n D 1, so that
.0/ D eacrf .1 C n/1=2 D 0. It then follows that no overall equilibrium can exist.

(i): Note first that the terminal wealth of an early informed trader is given by

W early WD x2 .; p2 /. fN C C "/ x1 .; p1 /p1 x2 .; p2 / x1 .; p1 / p2 C e0
while the terminal wealth of a late informed trader is given by

W late WD y2 .; p2 /. fN C C "/ y1 . p1 /p1 y2 .; p2 / y1 . p1 / p2 C e0 ;
for every realization fQ D g and f"Q D "g. Since W e late , viewed as
e early and W
random variables, are normally distributed conditionally on the information set
G2 and given the above assumptions of independence of the random variable
"Q and of exponential utility functions, it can be easily shown that the optimal
demand of early and late informed agents at date t D 2 (and recall that at t D 2
their information sets coincide) is given by
fN C p2
x2 .; p2 / D y2 .; p2 / D ;
a"2
conditionally on a realization fQ D g, thus proving (8.47).

Let us now turn to the proof of (8.48), which requires more computations.
We start by replacing equation (8.47) into the expression of the terminal wealth
of an early informed agent, thus yielding
fN C p2 N fN C p2
W early D 2
. f C C "/ p2 x1 .; p1 /. p1 p2 / C e0
a" a"2
. fN C p2 /2 fN C p2
D 2
C " x1 .; p1 /. p1 p2 / C e0 ;
a" a"2
for every realizations fQ D g and f"Q D "g. By iterated conditioning, we have
that
W early ˇ h ˇ ˇˇ i
E eae ˇG1 D E E eae W early ˇ
G2 ˇG1
and, noting that, conditionally on the information set G2 , the only unknown
random variable is "Q, which is independent of the conditioning information, it
holds that

. Nf C p2 /2
W early ˇ
E eae ˇG2 D ea e0 x1 .;p1 /p1 Cx1 .;p2 /p2 C
2a"2
;
so that
"
Q 2 /2 ˇ
#
ae ˇ
W early ˇ
Q 1 /p1 Cx1 .Q;p2 /p2 C . Nf Cp
a e0 x1 .;p ˇ
Ee G1 D E e
2
2a" ˇG 1 ;
ˇ
from which (8.48) follows by explicitly computing the last conditional expec-
tation and optimizing with respect to x1 .; p1 /.
Finally, referring to Hirshleifer et al. [947, Appendix] for full details,
y1 . p1 / can be derived, similarly as above, by computing iterated conditional
expectations and showing that
p2 p1 fN C p2 S2 S1
y1 . p 1 / D C ; (11.29)
aS1 a"2 S1
for suitable coefficients S1 and S2 and where and p2 denote the means
of Q and p2 conditionally on the observation of the equilibrium price p1 .
However, equation (8.46) (by risk neutrality and perfect competition of the
market makers) implies that in correspondence of the equilibrium it holds that
ˇ
p2 D fN C E EŒQ jD1 ./; D2 ./ˇD1 ./ D fN C EŒQ jD1 ./ D p1 ;
so that p2 p1 D 0. Moreover, it holds that
fN C D fN C EŒQ jD1 ./ D p1 D p2
and, hence, fN C p2 D 0. Equation (8.49) then follows from (11.29), thus

showing that at equilibrium the late informed agents do not trade at date t D 1.
(ii): The two inequalities (8.50) follow from the linear representation given in part
(i) of the equilibrium price functionals 1 and 2 together with the assumption
that the random variables zQ1 and zQ2 are independent from . Q
Solution of Exercise 8.20 Let us conjecture that the equilibrium price of the risky
asset is indeed given by (8.51). Then, the excess return per share, denoted by qtC1 WD
ptC1 C dtC1 rf pt , for t 2 N, can be written as
1 rf
qtC1 D .rf 1/0 C z .ztC1 rf zt / C yQ tC1 C "QtC1 :
rf ˛d rf ˛d
For each t 2 N, the distribution of the random variable qtC1 conditioned on the
observation of the current price and dividend and of the signal yQ t is normal with
conditional mean and variances given by
EŒqtC1 jpt ; dt ; yt D .rf 1/.0 C z zN/ C z .˛z rf /Qzt ;

Var.qtC1 jpt ; dt ; yt / D f2 C z2 z2 DW q2 :
Since every agent is supposed to maximize a negative exponential utility function,

the optimal demand of class A agents is given by
EŒqtC1 jpt ; dt ; yt 1
wtA; D D 2
.rf 1/.0 C z zN/ C .˛z rf /z zQt ;
a Var.qtC1 jpt ; dt ; yt / aq
while the optimal demand of class B agents is given by
EŒqtC1 jpt ; dt ; yt 1
wB; D D .rf 1/. 0 C z N
z / C .˛z rf / z Q
z t :
t
bt Var.qtC1 jpt ; dt ; yt / bt q2
Market clearing requires that wtA; C .1 /wB;

t D 1, so that

1
C .rf 1/.0 C z zN/ C .˛z rf /z zQt D q2 :
a bt
Since =a C .1 /=bt D 1=zt , it holds that
.˛z rf /z D q2 and .rf 1/.0 C z zN/ D q2 zN:
Under the condition z2 2; WD .rf ˛z /2 =.4f2 /, there are two real solutions z
to the second order equation
.˛z rf /z D f2 C z2 z2 ;
given by
q
rf ˛z
z D 1 ˙ 1 .z2 = 2; / :
2z2
Among the two solutions, we select the one associated with the positive sign, since it
satisfies z ! 0 for f2 ! 0. The term 0 is then given by 0 D .1˛z /z zN=.rf 1/.
Solution of Exercise 8.21 By the same arguments used in the first part of the proof
of Proposition 8.17, the optimal demand wi of each agent belonging to the first
group, in correspondence of a price p at date t D 0, is given by
N
di p
wi D max I0 ; for all dNi 2 ŒdN k; dN C k:
a
Concerning the second group of agents, their aggregate optimal demand is simply
given by .dN p/=aarb, since they are allowed to short sell the risky asset. The market
clearing condition then requires
Z N
1 dCk
dNi p N dN p
dd i C D u:
2k N
max. p;dk/ a aarb
Representation (8.38) can then be obtained by solving the last equation with respect
to p.
Solution of Exercise 8.22 Suppose first that condition (8.40) holds. Then, for every
a 2 A, it holds that
pt ıEa ŒdtC1 C ptC1 jFt ; for all t 2 N:
By iterating the above relation (and using the tower property of conditional
expectation), it follows that
" #
X
T ˇ
ˇ
pt Ea ı kt dk C ı Tt pT ˇFt ;
kDtC1
for every a 2 A and T t. By maximizing with respect to all a 2 f1; : : : ; Ag and

T t, this implies that condition (8.39) holds.
Conversely, suppose that condition (8.39) holds. Then, it holds that (choosing T D
t C 1)
pt max ıEa ŒdtC1 C ptC1 jFt ; for all t 2 N: (11.30)

a2f1;:::;Ag
Arguing by contradiction, suppose that there exists an event Ft 2 Ft such that a

strict inequality holds in (11.30). Then, applying equation (8.39) at t C 1 and using
iterated conditioning, it follows that on the event Ft
pt > max Ea ŒˇdtC1 C ˇptC1 jFt

a2f1;:::;Ag
" #
X
T ˇ
ˇ
max sup E a
ˇ kt
dk C ˇ Tt
pT ˇFt ;
a2f1;:::;Ag Tt
kDtC1
thus leading to a contradiction with (8.39) and proving that (11.30) must hold as an
equality.
Solution of Exercise 9.1 Consider two gambles xQ 1 and xQ 2 taking values in Œ1; 1
and denote by F1 and F2 their distribution functions, respectively. Note that, in
the context of prospect theory, F1 and F2 are defined with respect to some given
weighting function w./ (the result of this exercise does not depend on the choice of
the weighting function). For simplicity, we suppose that F1 .1/ D F2 .1/ D 0 and
that F1 and F2 are continuous. Let u./ W Œ1; 1 ! R be an arbitrary continuous
and increasing value function, such that u00 .x/ 0 for all x 2 Œ1; 0 and u00 .x/ 0
for all x 2 Œ0; 1. Define then
Z x
G.x; y/ WD F1 .z/ F2 .z/ dz; for all 1 y 0 x 1:
y
Note that
@G.x; y/ @G.x; y/
D F1 .x/ F2 .x/ and D F1 .y/ F2 .y/ :
@x @y
Similarly as in (2.13), using twice the integration by parts formula, we can compute
Z 1 Z 1
U.Qx1 / U.Qx2 / D u.z/dF1 .z/ u.z/dF2 .z/
1 1
Z Z
0 1
D u.z/ dF1 .z/ dF2 .z/ C u.z/ dF1 .z/ dF2 .z/
1 0
Z 0 Z 1

D F1 .z/ F2 .z/ u0 .z/dz F1 .z/ F2 .z/ u0 .z/dz
1 0
Z 0 Z 1
D G.x; z/u00 .z/dz C G.z; y/u00 .z/dz
1 0
for arbitrary y 2 Œ1; 0 and x 2 Œ0; 1. Recalling that u00 .x/ 0 for x 0 and that
u00 .x/ 0 for x 0, it follows that G.x; y/ 0 for all 1 y 0 x 1 implies
that U.Qx1 / U.Qx2 /. The converse implication follows by the same calculation by
considering arbitrary value functions u./.
Solution of Exercise 9.2 Note first that, by Proposition 2.8, the fact that xQ 1
FSD xQ 2
implies
X
s X
s
s s0 ; for every s D 1; : : : ; S:
kD1 kD1
By (9.4) and using the above property, we can compute

!
X
S X
s X
s1
U.Qx1 / D u.xs / w k w k
sD1 kD1 kD1
X
S X
s XS1 X
s
D u.xs /w k u.xsC1 /w k
sD1 kD1 sD1 kD1
X
S1
X
s
D u.xs / u.xsC1 / w k C u.xS /w.1/
sD1 kD1
X
S1
X
s
u.xs / u.xsC1 / w k0 C u.xS /w.1/ D U.Qx2 /;
sD1 kD1
where we have used the fact that xs < xsC1 , for every s D 1; : : : ; S 1, and that the
functions u./ and w./ are increasing.
(i): It suffices to compute, using the definition of the preference functional (9.7)
together with the linearity of the function v with respect to its second argument
and the definition of the function vO given in Proposition 9.5,
"
1
#
X
E ı t u.ct C ˛ct / C bt ı tC1 v.XtC1 ; wtC1 C ˛wtC1 ; zt /
tD0
1
" #
X

E ı u.ct / C bt ı v.XtC1 ; wtC1 ; zt /
t tC1
tD0
" 1
#
X

DE ı u.ct C ˛ct / u.ct / C bt ˛wtC1 ı v.r
t tC1
O tC1 ; zt /
tD0
1
" #
X
t 0
E ı u .ct /˛ct C bt ˛wtC1 ı v.r
tC1
O tC1 ; zt / ;
tD0
where, in the last step, we have used the concavity of the function x1 =.1 /.
(ii): Let .Wt /t2N be the self-financing wealth process associated to the strategy
.w ; c / and .Wt /t2N the self-financing wealth process associated to the
strategy .w; c/. Note that, by the self-financing condition, it holds that
;˛
˛WtC1 D WtC1 WtC1 D .˛Wt ˛ct /rf C ˛wtC1 .rtC1 rf /;
where .Wt;˛ /t2N is the wealth process associated to the self-financing strategy
.w C ˛w; c C ˛c/. Making use of this last relation together with the Euler
conditions (9.81)–(9.82), it then suffices to compute
0 0
ıE u0 .c
tC1 /˛WtC1 jFt D ı˛Wt rf E u .ctC1 /jFt ı˛ct rf E u .ctC1 /jFt

C ı˛wtC1 E u0 .ctC1 /.rtC1 rf /jFt
D ˛Wt u0 .c 0
t / ˛ct u .ct / ˛wtC1 ıbt E Œv.r
O tC1 ; zt /jFt ;
from which (9.83) immediately follows.

(iii): It suffices to sum relation (9.83) over all t 2 N. For the proof of the remaining
part of the claim, we refer the reader to the Appendix of Barberis et al. [150].
(iv): The Euler condition (9.81) together with the dynamics (9.6) satisfied by the
aggregate consumption process .Nct /t2N implies that

0 ˇ
ˇ
1 u .NctC1 / ˇˇ cN tC1 ˇˇ gc c tC1
D ıE F D ıE ˇFt D ıE Œe jFt
u0 .Nct / ˇ
t
rf cN t
2 c2
D ıe gc C 2 ;

(v): Note first that, making use of relation (9.5),
p
p
tC1 C dtC1
1C tC1
dtC1 dtC1 1 C f .ztC1 / dtC1 1 C f .ztC1 / gd Cd "tC1
rtC1 D D p
D D e ;
p
t
t dt f .zt / dt f .zt /
dt
for all t 2 N. To prove the claim, it then suffices to compute, using

basic properties of the bivariate normal distribution and noting that ztC1 is
determined only by zt and "tC1 ,

ˇ
cN tC1 ˇˇ
E rtC1 ˇFt
cN t

ˇ
1 C f .ztC1 / gd Cd "tC1 .gc Cc tC1 / ˇˇ
DE e ˇFt
f .zt /

ˇ
gd gc c tC1 1 C f .ztC1 / d "tC1 ˇˇ
De E E Œe jFt _ ."tC1 / e ˇFt
f .zt /

ˇ
2 c2 .12 / 1 C f .ztC1 / .d c /"tC1 ˇˇ
D egd gc C 2 E e ˇFt :
f .zt /
(vi): The Euler conditions (9.81)–(9.82) are necessary and sufficient for the
optimality of a trading-consumption strategy ..wt ; ct //t2N . Moreover, parts
(iv) and (v) of the exercise show that the strategy consisting in consuming the
aggregate consumption and holding the whole supply of the risky asset and the
zero net supply of the risk free asset satisfies the Euler optimality conditions,
from which the optimality of this strategy follows, thus completing the proof
of Proposition 9.5.
(i) Denote by .Wt /tD0;1;:::;T and .ct /tD0;1;:::;T the optimal wealth and consump-
tion processes, respectively, associated to the maximization of the recursive
preferences (9.19). In view of Proposition 9.7, the value function V satisfies
(9.21)–(9.22). In particular, similarly as in the proof of Proposition 6.4, the
first order condition of the maximization problem implies that, at each date
t 2 f0; 1; : : : ; T 1g,
@
@

˛
v ct ; t V.WtC1 ; t C 1/ D v ct ; t V.WtC1 ; t C 1/ t V.WtC1 ; t C 1/
@c @y

ˇ
˛ 0 n ˇ
ˇ
E V.WtC1 ; t C 1/ V .WtC1 ; t C 1/rtC1 ˇFt
@

˛
Dv ct ; t V.WtC1 ; t C 1/ t V.WtC1 ; t C 1/
@y

ˇ
@ n ˇ
ˇ
E V.WtC1 ; t C 1/˛ v c ; tC1 V.W
; t C 2/ rtC1 ˇFt ;
@c tC1 tC2
where the second equality follows from the envelope condition given in
@
Proposition 9.7 and where @y v denotes the first derivative of the function v
with respect to its second argument. Note that (9.19) implies that
@ @
v.c; y/ D v.c; y/% .1 ı/c% and v.c; y/ D v.c; y/% ıy% :
@c @y
By substituting into the above relation, we obtain

%˛ % ˇ

V.WtC1 ; t C 1/ ctC1 n ˇ
ˇ
1 D ıE rtC1 ˇFt ;
t .V.WtC1 ; t C 1// ct
for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N, where we have also

used relation (9.22). This proves the validity of the Euler condition (9.23).
Furthermore, if there exists a risk free asset with a constant rate of return
rf > 0, then the Euler condition (9.24) follows from (9.23) similarly as in the
proof of Corollary 6.5.
(ii): Note first that, as a consequence of the Euler condition (9.23) together with the
self-financing condition (6.19), the optimal wealth process .Wt /tD0;1;:::;T sat-
isfies the given backward stochastic difference equation. Let then .Xt /tD0;1;:::;T
be an arbitrary solution to the backward stochastic difference equation. Note
first that
V.WT ; T/ u.cT /
XT D D D cT D WT :
u0 .cT / u0 .cT /

By induction, suppose that XtC1 D WtC1 , for some t D 0; 1; : : : ; T 1. Then,
the backward stochastic difference equation implies that

ˇ
ˇ

tC1 ˇ
tC1 ˇˇ
Xt D ct CE ˇ
XtC1 ˇFt D ct C E WtC1 ˇFt D Wt ;

t
t
thus showing the uniqueness of the solution to the backward difference

equation.
(iii): Define the process .Xt /tD0;1;:::;T by
V.W ; t/ .c /%
Xt WD @
t D V.Wt ; t/1% t ; for all t D 0; 1; : : : ; T:
1ı
@c v ct ; t V.WtC1 ; t C 1/
It can be checked that the process .Xt /tD0;1;:::;T is a solution to the backward
stochastic difference equation considered in part (ii). Indeed, for all t D
0; 1; : : : ; T 1, it holds that

ˇ h i

tC1 ˇ ı ˛% %
E XtC1 ˇˇFt D
E V.WtC1 ; t C 1/1˛ t V.WtC1

; t C 1/ .ct / jFt

t 1ı
ı ˛% 1˛
D .c /% t V.WtC1

; t C 1/
t V.WtC1 ; t C 1/
1ı t
.ct /%
D V.Wt ; t/1% ct D Xt ct ;
1ı

where we have used the Ft -measurability of t .V.WtC1 ; t C 1// together with
its definition. In part (ii), we have shown that the above backward stochastic
difference equation admits a unique solution, thus proving the validity of
representation (9.25). Relation (9.26) then follows by means of elementary
computations, using the fact that

ˇ

tC1 ˇˇ ı 1%
Wt ct D E WtC1 ˇFt D .ct /% t V.WtC1

; t C 1/ ;

t 1ı
for all t D 0; 1; : : : ; T 1.
(iv): This follows by simply substituting (9.26) in the Euler condition (9.24).
(i) Using (9.29) and (9.30), relation (9.23) can be rewritten as follows:
h ˇ i
1 D E e log ıC. 1/ log rtC1 % log.ctC1 =ct /Clog rtC1 ˇFt ;
n

for all t D 0; 1; : : : ; T 1 and n D 1; : : : ; N. Since rtC1 can be written as a
linear combination of the returns on the single assets, we can rewrite the above
n
relation replacing rtC1 with rtC1 . Making use of the distributional assumptions
of Proposition 9.9, the conditional expectation can then be explicitly computed
as follows:
2 r2 2 %2 c2
% 2 cr
1 D e log ıCr %c C 2 C 2 ;
from which part (i) of the exercise follows.

(ii) Making use of relation (9.23), written in terms of the optimal return process
.rt /tD1;:::;T and with respect to the risk free asset with return rf , we can
compute, similarly as above,
. 1/2 r2 2 %2 c2

1 D e log ıC. 1/r %c C 2 C 2 . 1/%cr Clog rf
:
Relation (9.33) follows directly by substituting r with its expression com-

puted in part (i) of the exercise.
(iii) Similarly as in part (ii) of the exercise, we can compute
h ˇ i
1 D E e log ıC. 1/ log rtC1 % log.ctC1 =ct /Clog rtC1 ˇFt
n
. 1/2 r2 2 %2 c2 Var.log rn jFt /

tC1
D e log ıC. 1/r C %c C CEŒlog rtC1 jFt C
n
2 2 2
n /;log r n jF /
e. 1/%cr C. 1/ Cov.log rtC1 ;log rtC1 jFt /% Cov.log.ctC1 =ct tC1 t
:
Relation (9.34) then follows directly by relying on parts (i) and (ii) of the
exercise.

(i) As a first step, the relation WtC1 D .Wt ct /rtC1

can be rewritten in
logarithmic terms as

log WtC1 D log rtC1 C log 1 elog ct log Wt :
Letting xN WD EŒlog ct log Wt , the non-linear function log.1 ex / admits
the following Taylor expansion up to the first order:
exN
log.1 ex / log.1 exN / C .x xN /:
exN 1
By combining the last two equations, we can then write
1
log WtC1 log rtC1 C log .log ct log Wt log.1 //;

thus proving part (i) of the exercise.
(ii): Making use of part (i) of the exercise together with the approximate rela-
tion (9.85), it holds that

tC1
log WtC1 tC1 .log rtC1 log rf / C log rf C .1 tC1 / Var.log rtC1 jFt /
2

1
C 1 .log c
t log Wt / C K:

Let us then write the elementary identity
log ctC1 D .log ctC1 log WtC1

/ .log ct log Wt / C log WtC1

;
for all t D 0; 1; : : : ; T 1. It holds that
Cov.log rtC1 ; log ctC1 jFt / D Cov.log rtC1 ; log ctC1 log WtC1

jFt /

C Cov.log rtC1 ; log WtC1 jFt /:
Making use of the above relations, we get that

Cov.log rtC1 ; log rtC1 jFt / tC1 Var.log rtC1 jFt /
and

Cov.log rtC1 ; log WtC1 jFt / tC1 Var.log rtC1 jFt /;
so that
Cov.log rtC1 ; log ctC1 / Cov.log rtC1 ; log ctC1 log WtC1

jFt /

C tC1 Var.log rtC1 jFt /:
Making use of (9.85), relation (9.86) then follows directly.

(iii): To solve the final part of the exercise, it suffices to solve relation (9.86) with

respect to tC1 and note that
1 1 % % ˛1
D and D :
1 C % ˛ 1 C % %1 ˛

(i) For a given consumption process c D .ct /tD0;1;:::;T , let us denote
X
T
U.c/ D ı t u.ct ; zt /:
tD0
Similarly as in the proof of Proposition 6.4, at optimality it must hold that

" @ ˇ #
@ctC1
EŒU.c /jFtC1 ˇ
E @
.rtC1
n
rf /ˇˇFt D 0;
EŒU.c /jF
@ct t
where c D .ct /tD0;1;:::;T denotes the optimal consumption process. Condi-

tion (9.39) then follows by computing
@ XT
U.c / D ı t @c u.ct ; zt / C a ı s @z u.cs ; zs /bst
@ct sDtC1
!
X
Tt
Dı t
@c u.ct ; zt / Ca ı s
@z u.ctCs ; ztCs /bs ;
sD1
for all t D 0; 1; : : : ; T, where we have used relation (9.36) and where we have
denoted by @c u the partial first derivative of the function u with respect to
the first argument and, analogously, by @z u the partial first derivative of the
function u with respect to the second argument.
(ii): The second part of the exercise follows from a straightforward computation.
(i) It suffices to compute
@ .1˛/ 1˛ ˇ.1˛/1 .1ˇ/.1˛/

U.c/ D ı t c˛
t zt ı tC1 ˇ ctC1 ct cN t1 ;
@ct
from which (9.87) follows by elementary computations.

(ii): Relation (9.88) follows directly from (9.87) together with the assumption of
market equilibrium, using the notation introduced in the exercise. The Euler
condition can be obtained as usual, making use of relation (9.88).
(iii): It suffices to observe that

dtC1 C ptC1 dtC1 ptC1 ptC1
D 1C D xtC1 1 C
dt dt dtC1 dtC1
and make use of the Euler condition obtained in part (ii) of the exercise.
Solution of Exercise 9.10 As a first step, note that the intertemporal marginal rate
of substitution is given by
@ ˛
@ctC1 u.ctC1 ; ztC1 / ctC1 stC1
mtC1 D ı @
Dı ;
@ct
u.ct ; zt / ct st
for all t D 0; 1; : : : ; T 1, from which (9.48) follows in view of (9.46)–(9.47). In

order to prove (9.49), it suffices to compute

EŒmtC1 jFt D ıe˛.gC.'1/.log st Ns// E e˛.1C.st //vtC1 jFt
2 .1C.s 2 2
D ıe˛.gC.'1/.log st Ns//C˛ t // 2
and
Var.mtC1 jFt / D ı 2 e2˛.gC.'1/.log st Ns// Var.e˛.1C.st //vtC1 jFt /

2 2 2
2 2 2

D ı 2 e2˛.gC.'1/.log st Ns// e˛ .1C.st // e˛ .1C.st // 1 ;
from which (9.49) directly follows. Finally, in order to determine the equilibrium
risk free rate rf , it suffices to compute
1 1 ˛.gC.'1/.log st Ns//˛2 .1C.st //2 2

rf D D e 2 :
EŒmtC1 jFt ı
Solution of Exercise 9.11 Suppose that we are at date t, for some arbitrary t 2 N,
and denote the initial wealth at date t of an agent (rational or noise trader) by Wt . At
date t, both the rational and the noise trader have to choose how much to invest in
the risky asset in order to maximize the expected utility of wealth at the next period.
Let us first consider the case of a rational trader. Denoting by tC1 r
the quantity of
the risky asset held in the portfolio by a rational trader in the period Œt; t C1, it holds
that
WtC1 D .Wt tC1

r
pt /.1 C rf / C tC1
r
.rf C ptC1 /

D Wt .1 C rf / C tC1
r
rf C ptC1 pt .1 C rf / :
The same relation also holds for a noise trader, replacing tC1r
with tC1
n
, with the
latter denoting the quantity of the risky asset held in the portfolio by a noise trader
in the period Œt; t C 1. Due to the assumption of an exponential utility function, the
rational trader’s optimal portfolio problem consists in choosing the optimal quantity
r;
tC1 which maximizes the expression
r 2
Wt .1 C rf / C tC1
r
rf C EŒ ptC1 jFt pt .1 C rf / tC1 Var. ptC1 jFt /
over all tC1

r
2 R. As can be easily verified, the optimal tC1
r;
is given by
rf C EŒ ptC1 jFt .1 C rf /pt

tC1
r;
D :
2 Var. ptC1 jFt /
We now perform a similar analysis in the case of a noise trader. Similarly as above,
n;
the optimal portfolio problem consists in determining the quantity tC1 2 R which
maximizes the expression
n 2
Wt .1 C rf / C tC1
n
rf C EŒ ptC1 jFt pt .1 C rf / tC1 Var. ptC1 jFt / C tC1
n
%t ;
where %t represents the noise trader’s misperception of the next period’s expected
n;
price of the risky asset. As can be easily verified, the optimal tC1 is given by
n; rf C EŒ ptC1 jFt .1 C rf /pt %t

tC1 D C :
2 Var. ptC1 jFt / 2 Var. ptC1 jFt /
In equilibrium, the aggregate demand of the risky asset by rational and noise traders
must be equal to the aggregate supply, which is fixed at one. Hence, it must hold
that
tC1
n;
C .1 /tC1
r;
D 1;
for all t 2 N, leading to the condition

%t
tC1
r;
C D 1:
2 Var. ptC1 jFt /
Solving for pt , this implies that the equilibrium price process . pt /t2N must satisfy
1
pt D rf C EŒ ptC1 jFt 2 Var. ptC1 jFt / C %t :
1 C rf
Note that this formula relates the equilibrium price pt of the risky asset to the noise
traders’ misperception %t , to the dividend rf on the risk free asset, to the coefficient
of absolute risk aversion 2 as well as to the Ft -conditional distribution of the next
period’s equilibrium price ptC1 . Considering a steady-state equilibrium where the
unconditional distribution of pt is constant over time and following the arguments
given in De Long et al. [549, Section I.B], the last relation can be solved recursively
yielding
.%t %/
N %N 2
pt D 1 C C Var. ptC1 jFt /:
1 C rf rf rf
Observe that, in the right-hand side of the last relation, only the second term is
variable, since , %N and rf are constant and the conditional variance Var. ptC1 jFt / is
a function of the constant variance of noise traders’ misperception %t :
2 %2
Var. ptC1 jFt / D Var. ptC1 / D :
.1 C rf /2
Formula (9.51) then follows directly from the last two relations.
Let us now prove part (ii) of the proposition, following the arguments given in
De Long et al. [549, Section II.A]. In equilibrium, the difference in the returns on
the optimal portfolios of noise and rational traders is given by
r;
n;
RtC1 WD .tC1 tC1 / rf C ptC1 pt .1 C rf / :
By relying on the results established in the first part of the exercise, it holds that
n; r; %t .1 C rf /2 %t
tC1 tC1 D D ;
2 Var. ptC1 jFt / 22 %2
for all t 2 N. On the other hand, the Ft -conditional expected value of the excess
return on the risky asset is given by
22 %2
E rf C ptC1 pt .1 C rf /jFt D 2 Var. ptC1 jFt / %t D %t ;
.1 C rf /2
so that
.1 C rf /2 %2t
EŒ RtC1 jFt D %t :
2%2
By the law of iterated expectation, it then follows that
.1 C rf /2 %N 2 C .1 C rf /2 %2
EŒ RtC1 D E EŒ RtC1 jFt D %N ;
2%2
thus proving part (ii) of the proposition.

Solution of Exercise 9.12 Part (i) of the exercise easily follows from the assumption
of exponential utility functions together with the assumption of a normal distribution
for all the random variables appearing in the model. Indeed, it suffices to observe
that, due to the assumptions on the sequence of random variables .Q"t /t2N , it holds
that
EŒrf C "QtC1 C ptC1 jFt D rf C EŒ ptC1 jFt
and
Var.rf C "QtC1 C ptC1 jFt / D "2 C Var. ptC1 jFt /;
for all t 2 N. The explicit expressions for the optimal demand of the risky asset by
rational and noise investors then follow exactly as in Exercise 9.11.
In correspondence of a steady-state equilibrium of the economy, part (ii) of the
exercise simply follows by noting that
2 %2
Var.Q"tC1 C ptC1 jFt / D "2 C Var. ptC1 jFt / D "2 C Var. ptC1 / D "2 C
.1 C rf /2
and by following the derivation of part (i) of Proposition 9.11 as detailed in

Exercise 9.11, replacing with t in the limit as
converges to zero.
Finally, part (iii) of the exercise can be obtained by following the same steps as
in the last part of Exercise 9.11 and making use of the results established above.
Solution of Exercise 9.13 We first prove that the equilibrium price of the risky
security at dates t 2 f1; 2; 3g is given by (9.55). Since the informed agents are risk
neutral, they set market prices equal to the future expected dividend of the security,
conditionally on the available information at each date t 2 f1; 2; 3g. It is trivial that
Q while at the previous dates it holds that
p3 D d,
Q dQ C "Q;
p1 D EŒdj
Q dQ C "Q; dQ C :
p2 D EŒdj Q
Formula (9.55) then follows by standard properties of the multivariate normal

distribution, making use of the independence among the random variables d, Q "Q and
.
Q
We now compute the covariances stated in (9.56), by making use of for-
mula (9.55) together with the distributional assumptions of the model (compare with
Daniel et al. [516, Appendix B]):
d6 c2 2 ."2 c2 /

Cov. p3 p2 ; p2 p1 / D ;
.d2 C c2 /.d2 .c2 C 2 / C c2 2 /2
which is strictly positive since "2 > c2 , due to the informed agents’ overconfidence.
Let us then compute
d6 c2 ."2 c2 /

Cov. p2 p1 ; p1 p0 / D ;
.d2 C c2 /2 .d2 .c2 C 2 / C c2 2 /
which is strictly negative always due to the assumption that "2 > c2 . Similarly, we
can compute (noting that p0 D dN D 0)
d4 ."2 c2 /

N D
Cov. p3 p1 ; p1 d/ < 0;
.d2 C c2 /2
and
d4 2 ."2 c2 /

N D
Cov. p3 p2 ; p1 d/ < 0:
.d2 C c2 /.c2 .d2 C 2 / C d2 2 /
Solution of Exercise 9.14 Similarly as in Exercise 9.13, it holds that
Q dQ C "Q D d2
p1 D EŒdj .dQ C "/:
Q
d2 C c2
Concerning the equilibrium price at date t D 2, if sign.d C "/ ¤ sign.s2 /, then the
(over)confidence remains constant and, since the public signal sQ2 is uninformative
regarding the fundamental value of the asset, the equilibrium price does not change
at date t D 2, so that p2 D p1 . On the other hand, if sign.d C "/ D sign.s2 /, then
the equilibrium price at date t D 2 must be computed by taking into account the
increased degree of overconfidence of informed investors, so that
Q dQ C "Q D d2
p2 D EŒdj .dQ C "Q/;
d2 C c2 k
thus establishing (9.57). We now compute the covariances between the equilibrium
prices at different dates stated in (9.58) (see also Daniel et al. [516, Appendix D]).
Since the probability p is assumed to be exogenously given and the probability that
the price change at date t D 1 is positive is 1=2, the law of iterated expectations
gives that

Cov. p2 p1 ; p1 p0 / D E EŒ. p2 p1 /. p1 p0 /jQs2
kd4 .d2 C "2 /
D > 0:
2.d2 C c2 /2 .d2 C c2 k/
Similarly, it holds that

Cov. p3 p2 ; p2 p1 / D E EŒ. p3 p2 /. p2 p1 /jQs2
d2 .k2 d2 C k."2 c2 //
D < 0:
2.d2 C c2 /.d2 C c2 k/2
Finally, we have that
d4 ."2 c2 /

Cov. p3 p1 ; p1 p0 / D EŒ.dQ p1 /p1 D < 0;
.d2 C c2 /2
thus completing the proof of the proposition.

Solution of Exercise 9.15 We start from (9.62). Solving recursively for st , we obtain
X1 X1
1 EŒstCT jFt 1
st D s EŒd tCs jFt C lim T
D s EŒdtCs jFt C ˇt ;
r
sD1 f
T!1 rf r
sD1 f
where ˇt WD limT!1 EŒstCT jFt =rfT , t 2 N, represents the bubble component

(compare with the proof of Proposition 6.40). Making use of relation (9.63), solving
recursively for dt and relying on the independence of the random variables .Q"t /t2N ,
it holds that
EŒdtC2 jFt D EŒ˛ C dtC1 C "QtC1 jFt D ˛ C dtC1 ;

where the last equality uses the fact that dtC1 is Ft -measurable. Similarly, for s 3,
it holds that
EŒdtCs jFt D ˛.1 C : : : C s2

/C s1
dtC1 :
Therefore, we get that
X1 1
1 X dtC1 X
s2 s
st D ˛ k
C C ˇt ;
rs
sD2 f kD0
rs
sD1 f
Let us first compute

1
1 X s
1 rf 1
D 1 D :
sD1
rfs rf rf
For the other summation, it holds that
X1 1
1 X X
s2
1 1 s1
k
D
rs
sD2 f kD0
rs 1
sD2 f

1 rf 1 1 rf
D 1 1
1 rf 1 rf rf rf
1
D :
.rf 1/.rf /
Formula (9.64) then follows directly by combining the above results.

Solution of Exercise 9.16 Following the original arguments given in Weil [1650,
Proposition 2], if u000 0
1 > 0, then the function u1 is convex and Jensen’s inequality
implies that

EŒu01 .dQ 1 C eQ i1 / D E EŒu01 .dQ 1 C eQ i1 /jdQ 1

E u01 .dQ 1 C EŒQei1 jdQ 1 /

D E u01 .dQ 1 C eN 1 / ;
where we have used the fact that the random variables dQ 1 and eQ 1 are independent, so
that
EŒQe1 jdQ 1 D EŒQe1 D eN 1 :
In view of equations (9.78) and (9.80), this immediately implies that rOf rf if and
only if u000
1 < 0. Similarly, in the case of the expected equilibrium return, it holds
that

EŒdQ 1 u01 .dQ 1 C eQ i1 / D E EŒdQ 1 u01 .dQ 1 C eQ i1 /jdQ 1

E dQ 1 u01 .dQ 1 C EŒQei1 jdQ 1 /

D E dQ 1 u01 .dQ 1 C eN 1 / :
In view of equations (9.77) and (9.79), this completes the proof of the proposition.
Solution of Exercise 10.1 As a preliminary, note that the precisions inf and un
are both bounded from below by the prior precision d and bounded from above
by the full information precision F . Hence, there exist two constants 'inf 2 Œ0; 1
and 'un 2 Œ0; 1 such that (10.6) holds. In correspondence of a symmetric linear
equilibrium of the form (10.4), the market clearing condition (10.1) implies that
X
N
Ninf C Nˇ dQ C ˇ "Qn N inf pQ C M.un un p / C zQ D 0:
nD1
Solving for pQ yields

!
X
N
pQ D Nˇ dQ C ˇ "Qn C zQ C Ninf C Mun :
nD1
Observe that the random variable pQ is informationally equivalent to the random

variable hQ defined by
1 X
N
zQ
hQ WD dQ C "Qn C ;
N nD1 Nˇ
while, for every n D 1; : : : ; N, the couple . Qp ; yQ n / is informationally equivalent to

the couple .hQ n ; yQ n /, where
1 X
N
zQ
hQ n WD dQ C "Qk C :
N 1 kD1 .N 1/ˇ
k¤n
Relations (10.7) and (10.8) then follow by standard computations, using the basic
properties of the normal multivariate distribution (compare also with Kyle [1148,
Appendix A]) and computing
Q p D EŒdj
EŒdjQ Q h
Q and Q p ; yQ n D EŒdj
EŒdjQ Q hQ n ; yQ n ;
for every n D 1; : : : ; N. The fact that, if inf D un D 0, then EŒdjQQ p D pQ

holds if and only if 'un " D ˇun simply follows by substituting inf D un D 0
into (10.8). Finally, it remains to show that (10.9) holds. This can be shown by
making use of the definition of the parameters 'inf and 'un . Then, in order to show
that
inf un D .1 'un /.1 'inf /" ;
it suffices to substitute N.'inf 'un / D .1 'inf /'un , which follows from the
relations in (10.9), into

inf un D 1 'inf C N.'inf 'un / " ;
which is obtained from (10.6).

(i): Given the conjectured linear structure of the equilibrium, the expected profit
Q if he
of the insider trader, conditionally on the observation of the dividend d,
chooses to trade a quantity x of the asset can be written as

E dQ P.x C zQ/ xjdQ D .dQ x/x;
where we have used the assumption that the random variables dQ and zQ are
independent and that EŒQz D 0. Note also that we have implicitly used the
assumption that the insider trader takes the price setting rule of the market
maker as given and exploits it when deciding the quantity to trade. Since the
insider trader is assumed to maximize his expected profits, he chooses the
quantity x which maximizes the above expression, so that
dQ
x D ;
2
thus showing that ˇ D 1=.2/ and ˛ D ˇ.

(ii): Given the conjectured linear structure of the measurable functions X and P, the
market efficiency condition (10.11) can be rewritten as
C y D EŒdQ j˛ C ˇ dQ C zQ D y D EŒdQ jˇ dQ C zQ D y ˛;
for any y 2 R representing an arbitrary realization of the aggregate demand.

In view of the distributional assumptions on the random variables dQ and zQ, the
Q ˇ dQ C zQ/ is distributed as a bivariate normal random variable with

couple .d;
mean vector and covariance matrix respectively given by

dN d2 ˇd2
and :
ˇ dN ˇd2 ˇ 2 d2 C z2
Hence, the above conditional expectation can be explicitly computed as
ˇd2
EŒdQ j˛ C ˇ dQ C zQ D y D dN C N
.y ˛ ˇ d/:
ˇ 2 d2 C z2
In turn, this implies that
ˇd2 ˇd2
D and dN D N
.˛ C ˇ d/:
ˇ 2 d2
C z2 ˇ 2 d2
C z2
Combining the above results and solving for and ˇ lead directly to the result
of Proposition 10.3.
Solution of Exercise 10.4 We follow the arguments given in the solution of
Exercise 10.3. Observe first that, given the conjectured linear structure of the
equilibrium, the optimal demand by an informed agent observing the realization
y of the signal yQ can be obtained by determining the quantity x which maximizes
with respect to x the expected profits conditionally on the observation of fQy D yg:

E .dQ P.x C zQ//xjQy D y D E .dQ .x C zQ//xjQy D y

d2
D .y N x x:
d/
d2 C "2
It follows that the optimal demand x is explicitly given by
d2 N
2
d C"2
.y d/
x D ;
2
so that ˛ D =.2/ and ˇ D d2 =..d2 C "2 /2/. The equilibrium price is set
by the risk dealer according to the zero expected profit condition (10.11), given
the observation of the total market demand by informed and noise traders. Hence,
in equilibrium the following condition must hold, letting ! WD x C z denote an
arbitrary realization of the random variable xQ C zQ:
Q C ˇQy C zQ D !:
C ! D EŒdj˛
By computing explicitly the conditional expectation appearing on the right-hand

side of the last equation, relying on the assumption of normal distribution, we have
that
ˇd2
Q C ˇQy C zQ D ! D
EŒdj˛ N C d:
.! ˛ ˇ d/ N
ˇ 2 .d2 C "2 / C z2
From the last expression, it follows that
ˇd2
D :
ˇ 2 .d2 C "2 / C z2
Putting together the equations for ˇ and yields the explicit values
s s
z2 1 d4
ˇD 2
and D ;
d C "2 2 .d2 C "2 /z2

(i): The expected utility maximization problem of each informed trader consists
in determining the quantity x which maximizes the conditional expectation
h ˇ i h ˇ i
E ea.dQp/x ˇ"Q C Q D E ea.Q".xC.N1/ˇ.Q"CQ/CQz//x ˇ"Q C Q ;
Q
where we have used the fact that the total market demand, conditionally on
the observation of the signal "Q C Q and using the conjecture that each of the
remaining N 1 informed traders demands a quantity equal to ˇ.Q" C /, Q is
given by
x C .N 1/ˇ.Q" C /
Q C zQ;
so that the price set by the market maker equals

dN C x C .N 1/ˇ.Q" C /
Q C zQ :
By the assumption of a normal distribution, the above maximization problem

can be rewritten as the problem of finding the quantity xQ which maximizes
the mean-variance function
ˇ
Q C zQ/ xˇ"Q C Q
E "Q .x C .N 1/ˇ.Q" C /
a ˇ
Var "Q .x C .N 1/ˇ.Q" C / Q C zQ/ xˇ"Q C Q :
2
By using the basic properties of the normal multivariate distribution, the

conditional expectation and variance appearing above can be computed in
closed form. The optimal demand xQ can then be obtained by differentiating
with respect to x, setting the derivative equal to zero and solving with respect
to x. This yields the expression given in (10.31).
(ii): To prove the claim it suffices to set xQ D ˇ.Q" C / Q in (10.31) and solve the
resulting equation for ˇ.
(iii): Note that the total market demand by noise traders and informed traders is
given by
Nˇ.Q" C /
Q C zQ:
Hence, if the equilibrium price rule set by the market maker is linear with
respect to the market demand, it holds that

dN C Nˇ.Q" C / Q
Q C zQ D E djNˇ.Q
" C /
Q C zQ :
In particular, by the properties of the bivariate normal distribution, it holds

that
Cov.Q"; Nˇ.Q" C zQ/ C /

Q
D :
Var.Nˇ.Q" C /Q C zQ/
The claim then follows due to the normality assumptions on the random
variables zQ, "Q and .
Q
(i): For any k D 1; : : : ; k, let xk denote the demand submitted by the informed
trader k. In correspondence of such a demand and conditionally on the
observation of the signal yQ k D "Q C Q k , the expected profits of the informed
trader are given by
20 0 11 3
ˇ
X
K X
N ˇ
6B B CC ˇ 7
E dN C "Q pQ xk jQyk D E 4@"Q @xk C ˇ.Q" C Q l / C eQ n AA xk ˇyQ k 5 ;
ˇ
lD1 nD1
l¤k
where we have used the conjecture on the market maker’s price setting
rule and on the demand submitted by the other market participants. By the
distributional assumptions on the random variables appearing in the model,
the above conditional expectation can be computed as follows:

20 0 11 3
ˇ
6B B X
K X
N
CC ˇ ˇ 7
E6 B B
4@"Q @xk C ˇ.Q" C Q l / C eQ n C C 7
AA xk ˇˇ"Q C Q k 5
lD1 nD1
l¤k
"2 "2
D .Q
" C
Q k /x k ˇ.K 1/ .Q" C Q k /xk x2k :
"2 C 2 "2 C 2
Maximizing the last quantity with respect to xk yields the optimal value
1 .K 1/ˇ
xQ k D "2 .Q" C Q k /:
2."2 C 2 /
Therefore, the conjecture
xQ k D X.Q" C Q k / D ˇ.Q" C Q k /
can hold in equilibrium if and only if the coefficient ˇ satisfies condi-

tion (10.32).
(ii): Given the conjectured behavior of the other market participants, the profits
of the uninformed agent n 2 f1; : : : ; Ng if he demands a quantity wn of the
security are given by
0 0 11
B B X
K X
N
CC
.dN C "Q/.Qen C wn / B N B N
@d C @Kˇ d C ˇ .Q" C Q k / C wn C eQ m C C
AA wn :
kD1 mD1
m¤n
Due to the assumption of a negative exponential utility function together

with the multivariate normality assumption, the expected utility maximization
problem of the uninformed trader n reduces to determining the quantity wQ n
which maximizes the following mean-variance function with respect to wn :
2 0 1 3
ˇ
X
K X
N ˇ
6 B C ˇ 7
E 4.dN C "Q/Qen C "Qwn @Kˇ dN C ˇ .Q" C Qk / C wn C Qem A wn ˇQen 5
ˇ
kD1 mD1
m¤n
0 0 1 1
ˇ
XK XN ˇ
a B B C ˇ C
Var @.dN C "Q/Qen C "Qwn @Kˇ dN C ˇ .Q" C Qk / C wn C Qem A wn ˇQen A :
2 ˇ
kD1 mD1
m¤n
By relying on the multivariate normality assumption, the conditional expec-

tation and the conditional variance appearing in the last expression can be
explicitly calculated, similarly as above, and the optimal value wQ n can then be
calculated by differentiating the resulting expression with respect to wn and
setting the derivative equal to zero. This yields the optimal value wQ n given
in (10.33). Then, in equilibrium the conjecture
wQ n D W.Qen / D eQ n
is verified if and only if
a"2 .1 Kˇ/Qen
eQ n D ;
2 C a "2 .1 Kˇ/2 C K2 2 ˇ 2 C 2 .N 1/ 2 e2
which is equivalent to the validity of equation (10.34).

(iii): The zero expected profits condition for the risk neutral market maker, together
with the conjecture of a linear price formation rule, implies that a condition
equivalent to the market efficiency condition (10.11) must be satisfied, i.e.,
" ˇ K # !
ˇX X N X
K X
N
Q
E dˇˇ xQ k C N
wQ n D d C
xQ k C
wQ n
kD1 nD1 kD1 nD1
or, equivalently,
" ˇ K # !
ˇX X N X
K X
N
E "Qˇˇ xQ k C wQ n D xQ k C wQ n :
kD1 nD1 kD1 nD1
By relying on the optimal demands of the informed and of the uninformed

traders derived in parts (i) and (ii) of the exercise, the latter condition can be
rewritten as
" ˇ # !
ˇ XK X
N X K X
N
E "QˇˇKˇ "Q C ˇ Q k C eQ n D Kˇ "Q C ˇ Q k C eQ n :
kD1 nD1 kD1 nD1
Due to the distributional assumptions of the model, the last equation reduces
to
Kˇ"2
D :
K 2 ˇ 2 "2 C Kˇ 2 2 C 2 Ne2
Recall that, from step (i) of the exercise, the coefficient ˇ satisfies equa-
tion (10.32). By substitution, we obtain that and must satisfy equa-
tion (10.35). Moreover, condition (10.32) implies that 1 Kˇ > 0, so that
in turn equation (10.34) implies that we must choose the negative solution
to equation (10.35), for a given , thus proving part (iii) of the exercise.
(iv): In order to complete the proof of Proposition 10.5, it suffices to replace

the expression for the parameter obtained in part (iii) of the exercise into
equation (10.34), which then yields a linear equation for with an explicit
solution given as in the statement of Proposition 10.5. Substituting the explicit
value of the parameter into the equation obtained in part (iii) of the exercise
then yields the explicit value of . Finally, the equilibrium value of ˇ can be
deduced making use of relation (10.32).
(i): The expected utility maximization problem of each informed trader i 2 Œ0; 1,
conditionally on the observation of the market price pQ and the private signal yQ i ,
consists in determining the optimal demand xQ i which maximizes with respect
to xi the conditional expectation
h ˇ i
E ea.dQp/xi ˇp; yQ i :
Q
Due to the distributional assumptions of the model and the conjectured

linearity of the equilibrium price, the distribution of the random variable
dQ conditionally on the observation of . Qp; yQ i / is normal. Hence, the above
expected utility maximization problem can be reduced to the maximization
over xi 2 R of the following mean-variance function:

Q p; yQ i pQ xi a Var.djQ
EŒdjQ Q p; yQ i /x2i :
2
The optimal value xQ i is given by (see also Sect. 3.1)
Q yi p
EŒdjp;
xQ i D Xi . p; yi / D ;
Q yi /
a Var.djp;
Q yi WD EŒdjQ
for every realization f pQ D p; yQ i D yi g, denoting EŒdjp; Q p D p; yQ i D yi
(and similarly for the conditional variance).
(ii): Under the conjecture of linear demand schedules by informed traders, it holds
that
L. Qp/ D ˛ dQ C bQp C c C zQ:
The market maker sets an equilibrium price pQ as the solution of the equation
Q
pQ D EŒdjL./:
Note that, from the point of view of the market maker, the informative part of
the aggregate demand schedule L./ is given by ˛ dQ C zQ . Hence, the equilibrium
price pQ is determined by
Q dQ C zQ:
pQ D EŒdj˛
By standard properties of the bivariate normal distribution, it can be easily

computed that
˛d2
Q dQ C zQ D dN C
pQ D EŒdj˛ ˛. Q d/
d N C zQ ;
2
˛ 2 d C z2
from which the claim of part (ii) of the exercise follows by the definition of .
(iii): To complete the proof of Proposition 10.6, it remains to show that the demand
schedule of the informed traders admits the representation (10.18). This easily
follows by relying on the conditional distribution of a normal multivariate
random variable in order to compute the terms EŒdjQ Q p; yQ i and Var.djQ
Q p; yQ i /
appearing in the demand schedule computed in part (i) of the exercise:
Q p; yQ i D " yQ i C .˛ 2 z C d /Qp Q p; yQ i / D 1
EŒdjQ and Var.djQ :
" C ˛ 2 z C d " C ˛2 z C d
It then follows that the coefficients ˛, b and c introduced in the conjectured

linear representation in part (ii) of the exercise are given by
" "
˛D ; bD and c D 0;
a a
thus proving that representation (10.18) holds.
Solution of Exercise 10.9 Note first that, by the linearity of the expectation, it holds
that
e 0 D EŒW
EŒW e xn .EŒdQ n pN n /;
where We denotes the optimal wealth which solves the expected utility maximiza-
tion problem of the dealer at the initial date t D 0, xn is the order arriving at the
dealer and pN n D pask e0
n 1fxn >0g C pn 1fxn <0g . Similarly, the variance of W can be
bid
computed as
e 0 / D Var.W
Var.W e / C x2n n2 2xn Cov.W
e ; dQ n /:
Hence, the mean-variance problem (10.20) can be rewritten as

a 2 2
xn .EŒdQ n pN n / e ; dQ n / D 0:
xn n 2xn Cov.W
2
Solving the last equation with respect to pN n gives

a
pN n D dNn C xn n2 a Cov.W
e ; dQ n /;
2
from which the statement of Proposition 10.7 immediately follows.
(i): It suffices to observe that, since the current ask price is part of the publicly
available information (i.e., it is measurable with respect to the -algebra Dt ^
Ft ), it holds that
pask
t D Et Œ pask
t jCt D EŒEŒvjD
Q t ; Ct jDt ^ Ft ; Ct D Et ŒvjC
Q t :
(ii): Again by the tower property of the conditional expectation, it holds that
h ˇ i
Q t _ .ıt /; Ct jDt ^ Ft _ .ıt /; Ct ˇCt
Q t D Et E EŒvjK
Et ŒvjC
h ˇ i
Q t _ .ıt /jDt ^ Ft _ .ıt /; Ct ˇCt
D Et E EŒvjK
h ˇ i
Q t jDt ^ Ft _ .ıt /; Ct ˇCt ;
D Et E EŒvjK
where the last equality follows from the fact that the knowledge of the random
variable ıt does not bring any useful information in order to forecast the
fundamental value vQ (i.e., condition (10.21) holds).
(iii): Observe first that
EŒX1fX>ag EŒ.X a C a/1fX>ag EŒ.X a/C

EŒXjX > a D D D Ca
P.X > a/ P.X > a/ P.X > a/
EŒ.X a/C C a:
Moreover, the function ./C is a convex function and, hence, Jensen’s inequal-
ity implies that EŒ.X a/C .EŒX a/C . This yields
C
EŒXjX > a a C EŒX a EŒX:
With the same reasoning, an analogous inequality can be obtained in the case
of a conditional expectation of the form EŒXjA ; X > a, for an arbitrary
-algebra A .
(iv): It holds that

E EŒvjK
Q t jDt ^ Ft _ .ıt /; Ct E EŒvjK
Q t jDt ^ Ft _ .ıt / ;
where we have used the result established in step (iii) applied to the random
variable X D EŒvjK Q t , with a D paskt =ıt and A D Dt ^ Ft _ .ıt /. In view
of the result of step (ii), by taking the conditional expectation on both sides of
the last inequality, it then follows that
h ˇ i
Q t jDt ^ Ft _ .ıt /; Ct ˇCt
Q t D Et E EŒvjK
Et ŒvjC
h ˇ i
Q t jDt ^ Ft _ .ıt / ˇCt
Et E EŒvjK
h ˇ i
Q t jDt ^ Ft ˇCt
D Et E EŒvjK
h ˇ i
Q t ^ Ft ˇCt D Et Œv;
D Et E vjD Q
where in the second equality we have used condition (10.21) and again the
tower property of the conditional expectation. This proves the claim.
Solution of Exercise 10.11 The exercise is a simple application of Bayes’ rule.
Indeed, note that
pask D vP.vQ D vjbuy order/ C vP.vQ D vjbuy order/:
Hence, in order to compute pask , it suffices to compute the two conditional

probabilities P.vQ D vjbuy order/ and P.vQ D vjbuy order/. To this end:
P.vQ D v; buy order/ P.vQ D v/P.buy orderjvQ D v/

P.vQ D vjbuy order/ D D
P.buy order/ P.buy order/
P.vQ D v/P.buy orderjvQ D v/
D
P.vQ D v/P.buy orderjvQ D v/ C P.vQ D v/P.buy orderjvQ D v/
1˛
.˛ C 2
/
D 1˛
.˛ C 2
/ C .1 / 1˛
2
.1 C ˛/
D :
1 ˛ C 2 ˛
By means of similar computations, it holds that
.1 /.1 ˛/
P.vQ D vjbuy order/ D :
1 ˛ C 2 ˛
Hence:
.1 C ˛/ .1 /.1 ˛/
pask D v Cv :
1 ˛ C 2 ˛ 1 ˛ C 2 ˛
The computations in the case of the bid price are completely analogous.

(i): Similarly as in Exercise 10.11, the equilibrium ask price pask .1/ can be
computed as
pask .1/ D EŒvjbuy

Q size 1;
where EŒvjbuy
Q size 1 denotes the conditional expectation of the fundamental
value of the asset given that a buy order of size 1 has been submitted. By the
Bayes formula together with the assumptions of the present exercise, it holds
that
EŒvjbuy
Q size 1 D P.vQ D 1jbuy size 1/
P.vQ D 1/P.buy size 1jvQ D 1/
D
P.vQ D 1/P.buy size 1jvQ D 1/ C P.vQ D 0/P.buy size 1jvQ D 0/
˛.1 / C .1˛/.1ˇ/
2
D :
˛.1 / C .1 ˛/.1 ˇ/
The ask price pask .2/ associated to buy orders of size 2 can be computed in an
analogous way.
(ii): By relying on the ask prices computed in the previous step of the exercise and
using condition 2.1 pask .2// D 1 pask .1/, it holds that
!
˛ C .1˛/ˇ
2 ˛.1 / C .1˛/.1ˇ/
2
2 1 D1 :
˛ C .1 ˛/ˇ ˛.1 / C .1 ˛/.1 ˇ/
Solving for then gives the explicit value reported above. The values of
the ask prices pask .1/ and pask .2/ can then be obtained by substituting the
endogenously determined value of into the expressions computed in step (i)
of the exercise.
(iii): By arguing similarly as in Exercise 10.11 and in the first part of this exercise,
it holds that
pask .1/ D P.vQ D 1jbuy size 1/

D
1 .1˛/.1ˇ/
2 2 1
D 1 .1˛/.1ˇ/ 1 .1˛/.1ˇ/
D :
C 2
2 2 2 2
Similarly,
pask .2/ D P.vQ D 1jbuy size 2/

D
1 .1˛/ˇ
2. 2 C ˛/
D .1˛/ˇ
1
2. 2 C ˛/ C 12 .1˛/ˇ
2
.1˛/ˇ
2 C˛
D :
.1 ˛/ˇ C ˛
The necessary condition ˇ ˛=.1 ˛/ follows by substituting the explicit

expression of pask .1/ and pask .2/ into the inequality

2 1 pask .2/ 1 pask .1/:

(i): Due to the assumption of a normal distribution for the liquidation value of the
asset together with the assumption of a negative exponential utility function,
the expected utility maximization problem of each dealer m 2 f1; : : : ; Mg
corresponds to maximizing the following mean-variance function with respect
to xm , representing the quantity demanded of the asset, given a price pQ :

Q m a Var .dQ pQ /xm C dE
E .dQ pQ /xm C dE Q m :
2
For each m D 1; : : : ; M, the optimal solution is given by
dN pQ
Em :
ad2
(ii): The market clearing condition implies that

N
d pQ
M E C z D 0;
ad2
where z denotes an arbitrary realization of the random variable zQ. Solving for
the equilibrium price pQ directly yields the result.
(iii): It suffices to replace the explicit expression for the equilibrium price p into the
optimal demand schedule computed in part (i) of the exercise, thus yielding
z
xm D E Em ;
M
for each m D 1; : : : ; M.
References
1. Abarbanell, J. and Bernard, V. (1992) Tests of analysts’ overreaction/underreaction to

earnings information as an explanation for anomalous stock price behavior. Journal of
Finance, 47:1181–1207.
2. Abdellaoui, M., Bleichrodt, H. and Paraschiv, C. (2007) Loss aversion under prospect theory:
a parameter-free measurement. Management Science, 57:1659–1674.
3. Abel, A. (1989) Asset prices under heterogeneous beliefs: implications for the equity
premium puzzle. Mimeo.
4. Abel, A. (1990) Asset prices under habit formation and catching up with the Joneses. The
American Economic Review, 80:38–42.
5. Abel, A. (1999) Risk premia and term premia in general equilibrium. Journal of Monetary
Economics, 43:3–33.
6. Abel, A. (2001) The effects of investing social security funds in the stock market when fixed
costs prevent some houses from holding stocks. The American Economic Review, 91:128–
148.
7. Abel, A. (2002) An exploration of the effects of pessimism and doubt on asset returns.
Journal of Economic Dynamics and Control, 26:1075–1092.
8. Abel, A., Mankiw, G., Summers, L. and Zeckhauser, R. (1989) Assessing dynamic
efficiency: theory and evidence. Review of Economic Studies, 56:1–20.
9. Abreu, D. and Brunnermeier, M. (2002) Synchronization risk and delayed arbitrage. Journal
of Financial Economics, 66:341–360.
10. Abreu, D. and Brunnermeier, M. (2003) Bubbles and crashes. Econometrica, 71:173–204.
11. Acharya, V. and Pedersen, L.H. (2005) Asset pricing with liquidity risk. Journal of Financial
12. Ackert, R., Church, B. and Jayaraman, N. (2001) An experimental study of circuit breakers:
The effects of mandated market closures and temporary halts on market behavior. Journal
of Financial Markets, 4:185–208.
13. Ackert, R. and Smith, M. (1993) Stock price volatility, ordinary dividends, and other cash
flows to shareholders. Journal of Finance, 48:1147–1165.
14. Adam, K. and Marcet, A. (2011) Internal rationality, imperfect market knowledge and asset
prices. Journal of Economic Theory, 146:1224–1252.
15. Admati, A. (1985) A noisy rational expectations equilibrium for multi-asset securities
markets. Econometrica, 53:629–657.
16. Admati, A. (1989) Information in financial markets: the rational expectations approach. In
Financial Markets and Incomplete Information, S. Bhattacharya and G. Constantinindes
(eds.), 139–152.

Springer Finance, DOI 10.1007/978-1-4471-7322-9
766 References
17. Admati, A. and Pfleiderer, P. (1985) Interpreting the factor risk premia in the arbitrage
pricing theory. Journal of Economic Theory, 35:191–195.
18. Admati, A. and Pfleiderer, P. (1986) A monopolistic market for information. Journal of
Economic Theory, 39:400–438.
19. Admati, A. and Pfleiderer, P. (1988) Selling and trading on information in financial markets.
The American Economic Review, 78:96–103.
20. Admati, A. and Pfleiderer, P. (1988) A theory of intraday patterns. Volume and price
variability. Review of Financial Studies, 1:3–40.
21. Admati, A. and Pfleiderer, P. (1989) Divide and conquer: a theory of intraday and day-of-
the-week mean returns. Review of Financial Studies, 2:189–223.
22. Admati, A. and Pfleiderer, P. (1990) Direct and indirect sale of information. Econometrica,
58:901–928.
23. Admati, A. and Pfleiderer, P. (1991) Sunshine trading and financial market equilibrium.
Review of Financial Studies, 4:443–482.
24. Admati, A. and Pfleiderer, P. (2000) Forcing firms to talk: financial disclosure regulation and
externalities. Review of Financial Studies, 13:479–519.
25. Adrian, T. and Franzoni, F. (2009) Learning about beta: Time-varying factor loadings,
expected returns, and the conditional CAPM. Journal of Empirical Finance, 16:537–556.
26. Affleck-Graves, J., Hegde, S. and Miller, R. (1994) Trading mechanisms and the components
of the bid–ask spread. Journal of Finance, 49:1471–1488.
27. Aggarwal, R. and Wu, G. (2006) Stock market manipulations. Journal of Business, 79:1915–
1953.
28. Ahn, D., Boudoukh, J., Richardson, M. and Whitelaw, R. (2002) Partial adjustment or stale
prices? Implications from stock index and futures autocorrelations. Review of Financial
Studies, 15:655–689.
29. Ahn, D., Conrad, J. and Dittmar, R. (2003) Risk adjustment and trading strategies. Review
of Financial Studies, 16:459–485.
30. Ai, H. (2010) Information quality and long-run risk: asset pricing implications. Journal of
Finance 65:1333–1367.
31. Ait-Sahalia, Y. and Brandt, M. (2001) Variable selection for portfolio choice. Journal of
Finance, 56:1297–1351.
32. Ait-Sahalia, Y., Parker, J. and Yogo, M. (2004) Luxury goods and the equity premium.
Journal of Finance, 59:2959–3004.
33. Aiyagari, R. (1994) Uninsured idiosyncratic risk and aggregate saving. Quarterly Journal of
Economics, 109:659–684.
34. Aiyagari, R. and Gertler, M. (1991) Asset returns with transactions costs and uninsured
individual risk. Journal of Monetary Economics, 27:311–331.
35. Aiyagari, R. and Gertler, M. (1999) Overreaction of asset prices in general equilibrium.
Review of Economic Dynamics, 2:3–35.
36. Akgiray, V. (1989) Conditional heteroscedasticity in time series of stock returns: evidence
and forecasts. Journal of Business, 62:55–80.
37. Aktas, N., de Bodt, E. and Van Oppens, E. (2008) Legal insider trading and market efficiency.
Journal of Banking and Finance, 32:1379–1392.
38. Al-Najjar, N. (1999) On the robustness of factor structures to asset repackaging. Journal of
Mathematical Economics, 31:309–320.
39. Alessie, R. and Lusardi, A. (1997) Consumption, saving and habit formation. Economics
Letters, 55:103–108.
40. Aliber, R., Chowdhry, B. and Yan, S. (2003) Some evidence that a Tobin tax on foreign
exchange transactions may increase volatility. European Finance Review, 7:481–510.
41. Allais, M. (1953) Le comportement de l’homme rationnel devant le risque, critique des
postulates et axioms de l’école americaine. Econometrica, 21:503–546.
42. Allen, B. (1981) Generic existence of completely revealing equilibria for economies with
uncertainty when prices convey information. Econometrica, 49:1173–1199.
References 767
43. Allen, B. (1982) Strict rational expectations equilibria with diffuseness. Journal of Economic
Theory, 27:20–46.
44. Allen, F. and Gale, D. (1992) Stock price manipulation. Review of Financial Studies, 5:503–
529.
45. Allen, F. and Gale, D. (1994) Limited market particitpation and volatility of asset prices. The
46. Allen, F. and Gale, D. (1994) Financial Innovation and Risk Sharing. MIT Press, Cambridge.
47. Allen, F. and Gorton, G. (1993) Stock price manipulation, market microstructure and
asymmetric information. European Economic Review, 36:624–630.
48. Allen, F., Litov, L. and Mei, J. (2006) Large investors, price manipulation, and limits to
arbitrage: an anatomy of market corners. Review of Finance, 10:645–693.
49. Allen, F., Morris, S. and Postlewaite, A. (1993) Finite bubbles with short sale constraints
and asymmetric information. Journal of Economic Theory, 61:206–229.
50. Allen, F., Morris, S. and Shin, S. (2006) Beauty contests and iterated expectations in asset
markets. Review of Financial Studies, 19:719–752.
51. Allingham, M. (1990) Existence theorems in the capital asset pricing model. Econometrica,
59:1169–1174.
52. Alon, S. and Schmeidler, D. (2014) Purely subjective maxmin expected utility. Journal of
53. Alvarez, F. and Jermann, U. (2000) Efficiency, equilibrium and asset pricing with default
risk. Econometrica, 68:775–797.
54. Alvarez, F. and Jermann, U. (2001) Quantitative asset pricing implications of endogenous
solvency constraints. Review of Financial Studies, 14:1117–1151.
55. Ameriks, J. and Zeldes, S. (2004) How do portfolio shares vary with age? Mimeo.
56. Amershi, A. and Stoeckenius, J. (1983) The theory of syndicates and linear sharing rules.
Econometrica, 51:1407–1416.
57. Amihud, Y. (2002) Illiquidity and stock returns: cross–sectional and time series effects.
Journal of Financial Markets, 5:31–56.
58. Amihud, Y. and Mendelson, H. (1980) Dealership market. Journal of Financial Economics,
31:31–53.
59. Amihud, Y. and Mendelson, H. (1986) Asset pricing and the bid–ask spread. Journal of
Financial Economics, 17:223–249.
60. Amihud, Y., Mendelson, H. and Lauterbach, B. (1997) Market microstructure and securities
values: evidence from the Tel Aviv Stock Exchange. Journal of Financial Economics,
45:365–390.
61. Anderson, E., Ghysels, E. and Juergens, J. (2005) Do heterogeneous beliefs matter for asset
pricing? Review of Financial Studies, 18:875–924.
62. Anderson, E., Ghysels, E. and Juergens, J. (2009) The impact of risk and uncertainty on
expected returns. Journal of Financial Economics, 94:233–263.
63. Anderson, E., Hansen, L.P. and Sargent, T. (2003) A quartet of semigroups for model
specification, robustness, prices of risk, and model detection. Journal of the European
Economic Association, 1:68–123.
64. Ang, A., Bekaert, G., and Liu, J. (2005) Why stocks may disappoint. Journal of Financial
65. Ang, A. and Bekaert, G. (2001) Stock return predictability: is it there? Review of Financial
Studies, 20:651–707.
66. Ang, A. and Chen, J. (2007) CAPM over the long run: 1926–2001. Journal of Empirical
Finance, 14:1–40.
67. Ang, A., Chen, J. and Xing, Y. (2006) Downside risk. Review of Financial Studies, 19:1191–
1239.
68. Ang, A., Hodrick, R., Xing, Y. and Zhang, X. (2009) High idiosyncratic volatility and low
returns: International and further U.S. evidence. Journal of Financial Economics, 91:1–23.
69. Ang, A. and Liu J. (2004) How to discount cashflows with time-varying expected returns.
768 References
70. Ang, A., Papanikolau, D. and Westerfield, M. (2014) Portfolio choice with illiquid assets.
Management Science, 60:2737–2761.
71. Angeletos, G., Laibson, D., Repetto, A., Tobacman, J. and Weinberg, S. (2001) The
hyperbolic consumption model: calibration, simulation, and empirical evaluation. The
Journal of Economic Perspectives, 15:47–68.
72. Anscombe, F. and Aumann, R. (1963) A definition of subjective probability. Annals of
Mathematical Statistics, 34:199–205.
73. Antonelli, F., Barucci, E. and Mancino, M.E. (2001) Asset pricing with endogenous
aspirations. Decisions in Economics and Finance, 24:21–41.
74. Antonelli, F., Barucci, E. and Mancino, M.E. (2001) Asset pricing with a forward-backward
stochastic differential utility. Economics Letters, 72:151–157.
75. Araujo, A., Páscoa, M.R. and Torres-Martínez, J. P. (2002) Collateral avoids Ponzi schemes
in incomplete markets. Econometrica, 70:1613–1638.
76. Arrow, K. (1953) The role of securities in the optimal allocation of risk–bearing. Review of
Economic Studies, 30:91–96.
77. Arrow, K. (1968) Economic equilibrium. International Encyclopedia of the Social Sciences.
Crowell Collier & Macmillan, New York.
78. Arrow, K. (1970) Essays, in the Theory of Risk Bearing. North-Holland, Amsterdam.
79. Asness, C., Moskowitz, T. and Pedersen, L. (2013) Value and momentum everywhere.
80. Attanasio, O. (1991) Risk, time–varying second moments and market efficiency. Review of
81. Attanasio, O., Banks, J. and Tanner, S. (2002) Asset holding and consumption volatility.
Journal of Political Economy, 110:771–792.
82. Attanasio, O. and Browning, M. (1995) Consumption over the life cycle and over the
business cycle. The American Economic Review, 85:1118–1137.
83. Attanasio, O. and Davis, S. (1996) Relative wage movements and the distribution of
consumption. Journal of Political Economy, 104:1227–1250.
84. Attanasio, O., Goldberg, P. and Kyriazidou, E. (2008) Credit constraints in the market for
consumer durables: evidence from micro data on car loans. International Economic Review,
49:401–436.
85. Attanasio, O. and Vissing-Jørgensen, A. (2003) Stock-market participation, intertemporal
substitution, and risk-aversion. The American Economic Review, 93:383–391.
86. Attanasio, O. and Weber, G. (1989) Intertemporal substitution, risk aversion and the Euler
equation for consumption. The Economic Journal 99:59–73.
87. Attanasio, O. and Weber, G. (1993) Consumption growth, the interest rate and the aggrega-
tion. Review of Economic Studies, 60:631–649.
88. Attanasio, O. and Weber, G. (1995) Is consumption growth consistent with intertemporal
optimization? Evidence from the consumer expenditure survey. Journal of Political Econ-
omy, 103:1121–1157.
89. Ausubel, L. (1990) Insider trading in a rational expectations economy. The American
Economic Review, 80:1022–1041.
90. Ausubel, L. (1990a) Partially revealing rational expectations equilibrium in a competitive
economy. Journal of Economic Theory, 50:93–126.
91. Avramov, D. (2002) Stock–returns predictability and model uncertainty. Journal of Finan-
cial Economics, 64:423–458.
92. Avramov, D. (2004) Stock return predictability and asset pricing models. Review of
Financial Studies, 17:699–738.
93. Avramov, D. and Chordia, T. (2006) Asset pricing models and financial market anomalies.
Review of Financial Studies 19:1001–1040.
94. Avramov, D., Chordia, T. and Goyal, A. (2006) Liquidity and autocorrelations in individual
stock returns. Journal of Finance, 61:2365–2394.
95. Bacchetta, P. and van Wincoop, E. (2008) Higher order expectations in asset pricing. Journal
of Money, Credit and Banking, 40:837–866.
References 769
96. Bachelier, L. (1900) Theory of speculation. Reprinted in The Random Character of Stock
Market Prices, Cootner ed. (1964). MIT Press, Cambridge.
97. Back, K. (1992) Insider trading in continuous time. Review of Financial Studies, 5:387–409.
98. Back, K. (2010) Asset Pricing and Portfolio Choice Theory. Oxford University Press.
99. Back, K. and Baruch, S. (2004) information in securities markets: Kyle meets Glosten and
Milgrom. Econometrica, 72:433–465.
100. Back, K. and Baruch, S. (2007) Working orders in limit order markets and floor exchanges.
101. Back, K., Cao, H. and Willard, G. (2000) Imperfect competition among informed traders.
102. Back, K. and Pedersen, H. (1998) Long-lived information and intraday patterns. Journal of
Financial Markets, 1:385–402.
103. Backus, D., Gregory, A. and Zin, S. (1989) Risk premium in the term structure. Journal of
Monetary Economics, 23:371–399.
104. Badrinath, S., Kale, J. and Noe, T. (1995) Of shephelders, sheep, and the cross-
autocorrelations in equity returns. Review of Financial Studies, 8:401–430.
105. Bagehot, W. (1971) The only game in town. Financial Analysts Journal, 27:12–14.
106. Bagnoli, M. and Khanna, N. (1992) Insider trading in financial signalling models. Journal
of Finance, 47:1905–1934.
107. Bagnoli, M. and Lipman, B. (1996) Stock price manipulation through takeover bids. Rand
Journal of Economics, 27:124–147.
108. Baker, M. and Stein, J. (2004) Market liquidity as a sentiment indicator. Journal of Financial
Markets, 7:271–299.
109. Baker, M. and Wurgler, J. (2000) The equity share in new issues and aggregate stock returns.
110. Baker, M. and Wurgler, J. (2006) Investor sentiment and the cross-section of stock returns.
111. Baker, M. and Wurgler, J. (2007) Investor sentiment in the stock market. Journal of
Economic Perspectives, 21:129–151.
112. Baker, M., Wurgler, J. and Yuan, Y. (2012) Global, local, and contagious investor sentiment.
Journal of Financial Economics, 104:272–287.
113. Bakshi, G. and Chen, Z. (1996) The spirit of capitalism and stock-market prices. The
114. Balasko, Y. and Cass, D. (1989) The structure of financial equilibrium with exogenous
yields: the case of incomplete markets. Econometrica, 57:135–162.
115. Balduzzi, P., Bertola, G. and Foresi, S. (1995) Asset price dynamics and infrequent feedback
traders. Journal of Finance, 50:1747–1766.
116. Balduzzi, P., Foresi, S. and Hait, D. (1997) Price barriers and the dynamics of asset prices in
equilibrium. Journal of Financial and Quantitative Analysis, 32:137–159.
117. Balduzzi, P. and Kallal, H. (1997) Risk premia and variance bounds. Journal of Finance,
52:1913–1950.
118. Balduzzi, P. and Lynch, A. (1999) Transaction costs and predictability: some utility cost
calculations. Journal of Financial Economics, 52:47–78.
119. Balduzzi, P. and Robotti, C. (2001) Minimum-variance kernels, economic risk premia, and
tests of multi-beta models. Mimeo.
120. Balduzzi, P. and Yao, T. (2001) Does heterogeneity matter for asset pricing? Mimeo.
121. Ball, R. (1978) Anomalies in relationships between securities’ yields and yields-surrogates.
122. Ball, R. (1992) The earnings–price anomaly. Journal of Accounting and Economics, 15:
319–345.
123. Ball, R. and Brown, P. (1968) An empirical evaluation of accounting income numbers.
Journal of Accounting Research, 6:159–177.
124. Ball, R. and Kothari, S. (1989) Nonstationary expected returns: implications for tests
of market efficiency and serial correlation in returns. Journal of Financial Economics,
25:51–74.
770 References
125. Ball, R., Kothari, S. and Shanken, J. (1995) Problems in measuring portfolio performance.
An application to contrarian investment strategies. Journal of Financial Economics, 38:79–
107.
126. Balvers, R., Cosimano, T. and McDonald, B. (1990) Predicting stock returns in an efficient
market. Journal of Finance, 45:1109–1128.
127. Balvers, R., Wu, Y. and Gilliland, E. (2000) Mean reversion across national stock markets
and parametric contrarian investment strategies. Journal of Finance, 55:745–771.
128. Bamber, L., Barron, O. and Stober, T. (1999) Differential interpretations and trading volume.
Journal of Financial and Quantitative Analysis, 34:369–386.
129. Banerjee, S. (2011) Learning from prices and the dispersion in beliefs. Review of Financial
Studies, 24:3025–3068.
130. Banerjee, S. and Kremer, I. (2010) Disagreement and learning: dynamic patterns of trade.
131. Banerjee, S., Kaniel, R. and Kremer, I. (2009) Price drift as an outcome of differences in
higher-order beliefs. Review of Financial Studies, 22:3707–3743.
132. Banks, J., Blundell, R. and Brugiavini, A. (2001) Risk pooling, precautionary saving and
consumption growth. Review of Economic Studies, 68:757–780.
133. Bansal, R. and Coleman II, J. (1996) A monetary explanation of the equity premium, term
premium, and risk-free rate puzzles. Journal of Political Economy, 104:1135–1171.
134. Bansal, R., Dittmar, R. and Lundblad, C. (2005) Consumption, dividends, and the cross-
section of equity returns. Journal of Finance, 60:1639–1672.
135. Bansal, R., Dittmar, R. and Kiku, D. (2009), Cointegration and consumption risk in equity
returns. Review of Financial Studies, 22:1343–1375.
136. Bansal, R. and Lehman, B. (1997) Growth optimal portfolio restrictions on asset pricing
models. Macroeconomic Dynamics, 1:333–354.
137. Bansal, R. and Lundblad, C. (2002) Market efficiency, fundamental values, and asset returns
in global equity markets. Journal of Econometrics, 109:195–237.
138. Bansal, R. and Viswanathan, S. (1993) No-arbitrage and arbitrage pricing: A new approach.
139. Bansal, R. and Yaron, A. (2004) Risks for the long run: a potential resolution of asset pricing
puzzles. Journal of Finance, 59:1481–1509.
140. Banz, R. (1981) The relation between return and market value of common stocks. Journal
141. Barber, B., Lehavy, R., McNichols, M. and Trueman, B. (2001) Can investors profit from the
prophets? Security analyst recommendations and stock returns. Journal of Finance, 56:531–
563.
142. Barber, B. and Lyon, J. (1997) Detecting long run abnormal stock returns: the empirical
power and specification of test statistics. Journal of Financial Economics, 43:341–372.
143. Barber, B. and Odean, T. (2000) Trading is hazardous to your wealth: the common stock
investment performance of individual investors. Journal of Finance, 55:773–805.
144. Barber, B. and Odean, T. (2001) Boys will be boys: gender, overconfidence, and common
stock investment. Quarterly Journal of Economics, 116:261–292.
145. Barber, B. and Odean, T. (2002) Online investors: do the slow die first? Review of Financial
Studies, 15:455–487.
146. Barberis, N. (2000) Investing for the long run when returns are predictable. Journal of
Finance, 55:225–264.
147. Barberis, N., Greenwood, R.M., Jin, L. and Shleifer, A. (2015) X-CAPM: An extrapolative
capital asset pricing model. Journal of Financial Economics, 115:1–24.
148. Barberis, N. and Huang, M. (2001) Mental accounting, loss aversion, and individual stock
returns. Journal of Finance, 56:1247–1282.
149. Barberis, N. and Huang, M. (2008) Stocks as lotteries: the implications of probability
weighting for security prices. The American Economic Review, 98:2066–2100.
150. Barberis, N., Huang, M. and Santos, T. (2001) Prospect theory and asset prices. Quarterly
References 771
151. Barberis, N. and Shleifer, A. (2003) Style investing. Journal of Financial Economics,
68:161–199.
152. Barberis, N., Shleifer, A. and Wurgler, J. (2002) Comovement. Journal of Financial
Economics, 75:-283–317.
153. Barberis, N., Shleifer, A. and Vishny, R. (1998) A model of investor sentiment. Journal of
154. Barberis, N. and Thaler, R. (2003) A survey of behavioral finance. In Handbook of
Economics and Finance, G. Constantinides, M. Harris and R. Stulz (eds.), North-Holland,
Amsterdam, 1051–1121.
155. Barberis, N. and Xiong, W. (2009) What drives the disposition effect? An analysis of a long-
standing preference-based explanation. Journal of Finance, 64:751–784.
156. Barberis, N. and Xiong, W. (2012) Realization utility. Journal of Financial Economics,
104:251–271.
157. Barclay, M. (1997) Bid–ask spreads and the avoidance of odd-eighth quotes on Nasdaq: an
examination of exchange listings. Journal of Financial Economics, 45:35–60.
158. Barclay, M., Christie, W., Harris, J., Kandel, E. and Schultz, P. (1999) Effects of market
reform on the trading costs and depths of Nasdaq stocks. Journal of Finance, 44:1–34.
159. Barclay, M., Kandel, E. and Marx, L. (1999) The effects of transaction costs on stock prices
and trading volume. Journal of Financial Intermediation, 7:130–150.
160. Barclay, M., Litzenberger, R. and Warner, J. (1990) Private information, trading volume, and
stock–return variances. Review of Financial Studies, 3:233–253.
161. Barclay, M. and Warner, J. (1993) Stealth trading and volatility. Journal of Financial
162. Barillas, F., Hansen, L. P. and Sargent, T. (2009) Doubts or variability. Journal of Economic
Theory, 144:2388–2418.
163. Barlevy, G. and Veronesi, P. (2000) Information acquisition in financial markets. Review of
164. Barlevy, G. and Veronesi, P. (2003) Rational panics and stock market crashes. Journal of
165. Barner, M., Feri, F. and Plott, C. (2005) On the microstructure of price determination
and information aggregation with sequential and asymmetric information arrival in an
experimental asset market. Annals of Finance, 1:73–107.
166. Barro, R. (2006) Rare disasters and asset markets in the twentieth century. The Quarterly
167. Barro, R. and Jin, T. (2011) On the size distribution of macroeconomic disasters. Economet-
rica, 79:1567–1589.
168. Barsky, R. and De Long, B. (1993) Why does the stock market fluctuate? Quarterly Journal
of Economics, 108:291–311.
169. Barsky, R., Juster, T., Kimball, M. and Shapiro, M. (1997) Preference parameters and
behavioral heterogeneity: an experimental approach in the health and retirement study.
Quarterly Journal of Economics, 112:536–579.
170. Barucci, E. and Casna, M. (2014) On the market selection hypothesis in a mean reverting
environment. Computational Economics, 44:101–126.
171. Barucci, E. and Landi, L. (1996) Speculative dynamics under bounded rationality. European
Journal of Operational Research, 91:284–300.
172. Barucci, E. and Marazzina, D. (2012) Optimal investment, stochastic labor income and
retirement. Applied Mathematics and Computation, 218:5588–5604.
173. Barucci, E., Monte R. and Renò, R. (2004) Asset price anomalies under bounded rationality.
Computational Economics, 23:255–269.
174. Baruch, S. (2005) Who benefits from an open limit-order book? Journal of Business,
78:1267–1306.
175. Basak, S. (2002) A model of dynamic equilibrium asset pricing with heterogeneous beliefs
and extraneous risk. Journal of Economic Dynamics and Control, 24:63–95.
772 References
176. Basak, S. and Croitoru, B. (2001) Nonlinear taxation, tax arbitrage and equilibrium asset
prices. Journal of Mathematical Economics, 35:347–382.
177. Basak, S. and Croitoru, B. (2000) Equilibrium mispricing in a capital market with portfolio
constraints. Review of Financial Studies, 13:715–748.
178. Basak, S. and Cuoco, D. (1998) An equilibrium model with restricted stock market
particitpation. Review of Financial Studies, 11:309–341.
179. Basu, S. (1977) The investment performance of common stocks in relation to their prices to
earning ratios: a test of the efficient market hypothesis. Journal of Finance, 32:663–682.
180. Bateman, I., Munro, A., Rhodes, B., Starmer, C. and Sugden, R. (1997) A test of the theory
of reference-dependent preferences. Quarterly Journal of Economics, 112:479–505.
181. Bajeux-Besnainou, I., Jordan, J. and Portait, R. (2003) Dynamic asset allocation for stocks,
bonds and cash. The Journal of Business, 76:263–287.
182. Bajeux-Besnainou, I., Jordan, J. and Portait, R. (2000) An asset allocation puzzle: a
comment. The American Economic Review, 91:1170–1179.
183. Bajeux-Besnainou, I. and Rochet, J. (1996) Dynamic spanning: are options an appropriate
instrument? Mathematical Finance, 6:1–16.
184. Bebchuk, L. and Fershtman, C. (1994) Insider trading and the managerial choice among
risky projects. Journal of Financial and Quantitative Analysis, 29:1–14.
185. Bekaert, G., Campbell, R. and Lundblad, C. (2007) Liquidity and expected returns: lessons
from emerging markets. Review of Financial Studies, 20:1783–1831.
186. Bekaert, G. and Hodrick, R. (1992) Characterizing predictable in excess returns on equity
and foreign exchange markets. Journal of Finance, 47:467–507.
187. Bekaert, G., Hodrick, R. and Marshall, D. (1997) The implications of first–order risk
aversion for asset market risk premiums. Journal of Monetary Economics, 40:3–39.
188. Bekaert, G. and Liu, J. (2004) Conditioning information and variance bounds on pricing
kernels. Review of Financial Studies, 17:339–378.
189. Benabou, R. and Laroque, G. (1992) Using privileged information to manipulate market:
insiders, gurus, and credibility. Quarterly Journal of Economics, 107:921–958.
190. Benartzi, S. (2001) Excessive extrapolation and the allocation of 401(k) accounts to
company stock. Journal of Finance, 56:1747–1764.
191. Benartzi, S. and Thaler, R. (1995) Myopic loss aversion and the equity premium puzzle.
192. Benninga, S. and Protopapadakis, A. (1990) Time preference and the equity premium puzzle.
Journal of Monetary Economics, 25:49–58.
193. Benos, A. (1998) Aggressiveness and survival of overconfident traders. Journal of Financial
Markets, 1:353–383.
194. Benveniste, L., Marcus, A. and Wilhelm, W. (1992) What’s special about the specialist?
195. Benzoni, L., Collin-Dufresne, P. and Goldstein, R. (2007) Portfolio choice over the life cycle
when stock and labor markets are cointegrated. Journal of Finance, 62:2123–2167.
196. Bergman, Y. (1985) Time preference and capital asset pricing models. Journal of Financial
197. Berk, J. (1995) A critique of size-related anomalies. Review of Financial Studies, 8:275–286.
198. Berk, J. (1997) The acquisition of information in a dynamic market. Economic Theory,
9:441–451.
199. Berk, J., Green, J. and Naik, V. (1999) Optimal investment, growth options, and security
200. Berk, J. and Uhlig, H. (1993) The timing of information in a general equilibrium framework.
Journal of Economic Theory, 59:275–287.
201. Berkelaar, A., Kouwenberg, R. and Post, T. (2004) Optimal portfolio choice under loss
aversion. Review of Economics and Statistics, 86:973–987.
202. Bernard, V. and Thomas, J. (1990) Evidence that stock prices do not fully reflect the
implications of current earnings for future earnings. Journal of Accounting Economics,
13:305–340.
References 773
203. Bernardo, A. and Judd, K. (2000) Asset market equilibrium with general tastes, returns, and
informational asymmetries. Journal of Financial Markets, 3:17–43.
204. Bernardo, A. and Ledoit, O. (2000) Gain, loss, and asset pricing. Journal of Political
Economy, 108:144–172.
205. Bernhardt, D., Hollifield, B. and Hughson, E. (1995) Investment and insider trading. Review
206. Bernhardt, D. and Hughson, E. (1997) Splitting orders. Review of Financial Studies, 10:69–
101.
207. Berrada, T. (2009) Bounded rationality and asset pricing with intermediate consumption.
Review of Finance, 13:693–672.
208. Berry, T. and Howe, K. (1994) Public information arrival. Journal of Finance, 49:1331–
1346.
209. Bertaut, C. (1998) Stockholding behavior of US households: evidence from the 1983–1989
survey of consumer finances. Review of Economics and Statistics, 80:263–275.
210. Bertaut, C. and Starr-McCluer, M. (2001) Household portfolios in the United States. In
Household Portfolios, L. Guiso, M. Haliassos and T. Jappelli (eds.), MIT Press, Cambridge,
181–217.
211. Bertola, G., Guiso, L. and Pistaferri, L. (2005) Uncertainty and consumer durables adjust-
ment. Review of Economic Studies, 72:973–1007.
212. Bessembinder, H. (1997) The degree of price resolution and equity trading costs. Journal of
213. Bessembinder, H. (1999) Trade execution costs on Nasdaq and the NYSE: a post reform
comparison. Journal of Financial and Quantitative Analysis, 34:387–407.
214. Bessembinder, H. (2003) Quote-based competition and trade execution costs in NYSE-listed
stocks. Journal of Financial Economics, 70:385–422.
215. Bessembinder, H., Chan, K. and Seguin, P. (1996) An empirical examination of information,
differences of opinion, and trading activity. Journal of Financial Economics, 40:105–134.
216. Bessembinder, H. and Kaufman, H. (1997) A comparison of trade execution costs for NYSE
and NASDAQ–listed stocks. Journal of Financial and Quantitative Analysis, 32:287–310.
217. Bessembinder, H. and Venkataraman, K. (2004) Does an electronic stock exchange need an
upstairs market? Journal of Financial Economics, 73:3–36.
218. Bettis, J., Coles, J., Lemmon, M. (2000) Corporate policies restricting trading by insiders.
219. Bewley, T. (1980) The optimum quantity of money. In Models of Monetary Economies, J.
Kareken and N. Wallace (eds.), Federal Reserve Bank of Minneapolis.
220. Bhamra, H. and Uppal, U. (2014) Asset prices with heterogeneity in preferences and beliefs.
221. Bhandari, L. (1988) Debt/equity ratio and expected common stock returns: empirical
evidence. Journal of Finance, 43:507–528.
222. Bhattacharya, S. and Nicodano, G. (2001) Insider trading, investment, and liquidity: a
welfare analysis. Journal of Finance, 56:1141–1156.
223. Bhattacharya, U. and Daouk, H. (2002) The world price of insider trading. Journal of
Finance, 57:75–108.
224. Bhattacharya, U. and Daouk, H. (2009) When no law is better than a good law. Review of
Finance, 13:577–627.
225. Bhattacharya, U., Daouk, H., Jorgenson, B. and Kehr, C. (2001) When an event is not an
event: the curious case of an emerging market. Journal of Financial Economics, 55:69–102.
226. Bhattacharya, U. and Lipman, B. (1995) Ex ante versus interim rationality and the existence
of bubbles. Economic Theory, 6:469–494.
227. Bhattacharya, U. and Spiegel, M. (1991) Insiders, outsiders, and market breakdowns. Review
228. Bhattacharya, U. and Spiegel, M. (1998) Anatomy of a market failure: NYSE trading
suspensions (1974–1988) Journal of Business and Economic Statistics, 16:216–226.
774 References
229. Bhojraj, S. and Swaminathan, B. (2001) Macromomentum: evidence of predictability in

international equity markets. Mimeo.
230. Bhushan, R., Brown, D. and Mello, S. (1997) Do noise traders “create their own space?”
231. Biais, B. (1993) Price formation and equilibrium liquidity in fragmented and centralized
markets. Journal of Finance, 48:157–184.
232. Biais, B. and Bossaerts, P. (1998) Asset prices and trading volume in a beauty contest.
Review of Economic Studies, 65:307–340.
233. Biais, B., Bossaerts, P. and Spatt, C. (2010) Equilibrium asset pricing and portfolio choice
under asymmetric information. Review of Financial Studies, 23:1503–1543.
234. Biais, B., Foucault, T. and Salanié, F. (1998) Floors, dealer markets and limit order markets.
235. Biais, B. and Germain, L. (2002) Incentive compatible contracts for the sale of information.
236. Biais, B., Glosten, L. and Spatt, C. (1995) Market microstructure: a survey of microfounda-
tions, empirical results, and policy implications. Journal of Financial Markets, 8:217–264.
237. Biais, B. and Hillion, P. (1994) Insider and liquidity trading in stock and option markets.
238. Biais, B., Hillion, P., Mazurier, K. and Pouget, S. (2005) Judgemental overconfidence,
self-monitoring, and trading performance in an experimental financial market. Review of
239. Biais, B., Hillion, P. and Spatt, C. (1995) An empirical analysis of the limit order book and
the order flow in the Paris bourse. Journal of Finance, 50:1655–1689.
240. Biais, B., Martimort, D. and Rochet, J. (2000) Competing mechanisms in a common value
environment. Econometrica, 68:799–837.
241. Biais, B and Shadur, R. (2000) Darwinian selection does not eliminate irrational traders.
European Economic Review, 44:469–490.
242. Black, F. (1972) Capital market equilibrium with restricted borrowing. Journal of Business,
45:444–454.
243. Black, F. (1986) Noise. Journal of Finance, 41:529–541.
244. Black, F. (1993) Estimating expected return. Financial Analyst Journal, 49:36–38.
245. Black, F., Jensen, M. and Scholes, M. (1972) The capital asset pricing model: some empirical
tests. In Studies in the Theory of Capital Markets, M.C. Jensen (ed.), Praeger, New York.
246. Blackwell, D. (1951) Comparison of experiments. In Proceedings of the Second Berkeley
Symposium on Mathematical Statistics and Probability, J. Neyman (ed.), University of
California Press, Berkeley, 93–102.
247. Blanchard, O. (1993) Movements in the equity premium. Brooking papers on economic
activity, 2:75–138.
248. Blanchard, O. and Watson, M. (1982) Bubbles, rational expectations and financial markets.
In Crises in the Economic and Financial Structure, P. Wachtel (ed.), Lexington Books.
249. Bloomfield, R. (1996) Quotes, prices, and estimates in a laboratory market. Journal of
Finance, 51:1791–1808.
250. Bloomfield, R., Libby, R. and Nelson, M. (2000) Underreactions, overreactions and moder-
ated confidence. Journal of Financial Markets, 3:113–137.
251. Bloomfield, R. and O’Hara, M. (1998) Does order preferencing matter? Journal of Financial
252. Bloomfield, R. and O’Hara, M. (1999) Market transparency: Who wins and who loses?
253. Bloomfield, R. and O’Hara, M. (2000) Can transparent markets survive? Journal of
254. Bloomfield, R., O’Hara, M. and Saar, G. (2005) The “make or take” decision in an electronic
market: Evidence on the evolution of liquidity. Journal of Financial Economics, 75:165–199.
255. Blume, L. and Easley, D. (1992) Evolution and market behavior. Journal of Economic
Theory, 58:9–40.
References 775
256. Blume, L. and Easley, D. (2006) If you’re so smart, why aren’t you rich? Belief selection in
complete and incomplete markets. Econometrica, 74:929–966.
257. Blume, L., Easley, D. and O’Hara, M. (1994) Market statistics and technical analysis: the
role of volume. Journal of Finance, 49:153–179.
258. Blume, M. and Friend I. (1973) A new look at the capital asset pricing model. Journal of
Finance, 28:19–33.
259. Blundell, R., Pistaferri, L. and Preston, I. (2008) Consumption inequality and partial
insurance. The American Economic Review, 98:1887–1921.
260. Bodie, Z., Merton R. and Samuelson, W. (1992) Labour supply flexibility and portfolio
choice in a life cycle model. Journal of Economic Dynamics and Control, 16:427–449.
261. Bodurtha, J. and Mark, N. (1991) Testing the CAPM with time-varying risks and returns.
262. Boehme, R., Danielsen, B. and Sorescu, S. (2006) Short-sale constraints, differences of
opinion, and overvaluation. Journal of Financial and Quantitative Analysis, 41:455–487.
263. Boldrin, M., Christiano, L. and Fisher, J. (2001) Habit persistence, asset returns, and the
business cycle. The American Economic Review, 91:149–166.
264. Bollerslev, T., Engle, R. and Wooldridge, J. (1988) A capital asset pricing model with time–
varying covariances. Journal of Finance, 96:116–131.
265. Bollerslev, T., Tauchen, G. and Zhou, H. (2009) Expected stock returns and variance risk
premia. Review of Financial Studies, 22:4463–4492.
266. Bonomo, M. and Garcia, R. (1996) Consumption and equilibrium asset pricing: an empirical
assessment. Journal of Empirical Finance, 3:239–265.
267. Bonomo, M., Garcia, R., Meddahi, N. and Teodongap, R: (2011) Generalized disappoint-
ment aversion, long-run volatility risk, and asset prices. Review of Financial Studies,
24:82–122.
268. Boot, A. and Thakor, A. (2001) The many faces of information disclosure. Review of
269. Borch, K. (1962) Equilibrium in a reinsurance market. Econometrica, 30:424–444.
270. Bossaerts, P. and Green, R. (1990) A general equilibrium model of changing risk premia:
theory and tests. Review of Financial Studies, 2:467–493.
271. Bossaerts, P., Frydman, C. and Ledyard, J. (2014) The speed of information revelation and
eventual price quality in markets with insiders: comparing two theories. Review of Finance,
18:1–22.
272. Boswijk, H., Hommes, C. and Mazan, S. (2007), Behavioral heterogeneity in stock prices.
273. Boudoukh, J., Richardson, M. and Whitelaw, R. (1994) A tale of three schools: insights on
autocorrelations of short horizon stock returns. Review of Financial Studies, 7:539–573.
274. Bowman, D., Minehart, D. and Rabin, M. (1999) Loss aversion in a consumption-savings
model. Journal of Economic Behavior & Organization, 38:155–178.
275. Boyer, B., Mitton, T. and Vorkink, K. (2010) Expected idiosyncratic skewness. Review of
276. Boyle, P., Garlappi, L., Uppal, R. and Wang, T. (2012) Keynes meets Markowitz: the trade-
off between familiarity and diversification. Management Science, 58:253–272.
277. Branch, W. and Evans, G. (2010) Asset return dynamics and learning. Review of Financial
Studies, 23:1651–1680.
278. Branch, W. and Evans, G. (2011) Learning about risk and return: a simple model of bubbles
and crashes. American Economic Journal: Macroeconomics, 3:159–191.
279. Brandt, M. (1999) Estimating portfolio and consumption choice: a conditional Euler
equations approach. Journal of Finance, 54:1609–1645.
280. Brandt, M. (2010) Portfolio choice problems. In Handbook of Financial Econometrics,
Volume 1, Y. Ait-Sahalia and L.P. Hansen (eds.), North Holland, Amsterdam, 269–336.
281. Brandt, M., Zeng, Q. and Zhang, L. (2004) Equilibrium stock return dynamics under
alternative rules of learning about hidden states. Journal of Economic Dynamics and Control,
28:1925–1954.
776 References
282. Braun, P., Constantinides, G. and Ferson, W. (1993) Time nonseparability in aggregate
consumption. International evidence. European Economic Review, 37:897–920.
283. Brav, A., Constantinides, G. and Gezcy, C. (2002) Asset pricing with heterogeneous
consumers and limited particitpation: empirical evidence. Journal of Political Economy,
110:793–824.
284. Brav, A. and Heaton, J. (2002) Competing theories of financial anomalies. Review of
285. Breeden, D. (1979). An intertemporal asset pricing model with stochastic consumption and
investment opportunities. Journal of Financial Economics, 7:265–296.
286. Breeden, M., Gibbons, R. and Litzenberger, R. (1989) Empirical tests of the consumption-
oriented CAPM. Journal of Finance, 46:231–262.
287. Breeden, D. and Litzenberger, R. (1978) Prices of state-contingent claims implicit in option
prices. Journal of Business, 51:621–651.
288. Breen, W. and Korajczyk, R. (1995) On selection biases in book-to-market based tests of
asset pricing models. Mimeo.
289. Brennan, M. (1970) Taxes, market valuation and corporate financial policy. National Tax
Journal, 23:417–427.
290. Brennan, M. (1998) The role of learning in dynamic portfolio decisions. European Finance
Review, 1:295–306.
291. Brennan, M. and Cao, H. (1997) International portfolio flows. Journal of Finance, 52:1851–
1880.
292. Brennan, M. and Chordia, T. (1993) Brokerage commission schedules. Journal of Finance,
48:1379–1402.
293. Brennan, M., Chordia, T. and Subrahmanyam, A. (1998) Alternative factor specifications,
security characteristics and the cross expected stock returns. Journal of Financial Eco-
nomics, 49:345–373.
294. Brennan, M., Chordia, T., Subrahmanyam, A. and Tong, Q. (1998) Sell-order liquidity and
the cross-section of expected stock returns. Journal of Financial Economics, 105:523–541.
295. Brennan, M., Jegadeesh, N. and Swaminathan, B. (1993) Investment analysis and the
adjustment of stock prices to common information. Review of Financial Studies, 6:799–824.
296. Brennan, M., Schwartz, E. and Lagnado, R. (1997) Strategic asset allocation. Journal of
Economic Dynamics and Control, 21:1377–1403.
297. Brennan, M. and Subrahmanyam, A. (1996) Market microstructure and asset pricing: on
the compensation for adverse selection in stock returns. Journal of Financial Economics,
41:341–364.
298. Brennan, M. and Subrahmanyam, A. (1998) The determinants of average trade size. Journal
of Business, 71:1–23.
299. Brennan, M., Wang, A. and Xia, Y. (2004) Estimation and test of a simple model of
intertemporal capital asset pricing. Journal of Finance 59:1743–1775.
300. Brennan, M. and Xia, Y. (2000) Stochastic interest rates and the bond–stock mix. European
Finance Review, 4:197–210.
301. Brennan, M. and Xia, Y. (2002) Dynamic asset allocation under inflation. Journal of
Finance, 57: 1201–1238.
302. Brennan, M. and Xia, Y. (2001) Stock price volatility, learning, and the equity premium.
Journal of Monetary Economics, 47: 249–283.
303. Brennan, M. and Xia, Y. (2001) Assessing asset pricing anomalies. Review of Financial
Studies, 14:905–942.
304. Bris, A. (2005) Do insider trading laws work? European Financial Management, 11:267–
312.
305. Brock, W. (1982) Asset prices in a production economy. In The Economics of Information
and Uncertainty, J. McCall (ed.), University of Chicago Press, 1–46.
306. Brock, W. and Hommes, C. (1998) Heterogeneous beliefs and routes to chaos in a simple
asset pricing model. Journal of Economic Dynamics and Control, 22:1235–1274.
References 777
307. Brock, W. and Kleidon, A. (1992) Periodic market closure and trading volume. Journal of
308. Brock, W., Lakonishok, J. and LeBaron, B. (1992) Simple technical trading rules and the
stochastic properties of stock returns. Journal of Finance, 47:1731–1764.
309. Brock, W. and LeBaron, B. (1996) A dynamic structural model for stock return volatility
and trading volume. Review of Economics and Statistics, 78:94–110.
310. Brown, D. and Jennings, R. (1989) On technical analysis. Review of Financial Studies,
2:527–552.
311. Brown, D. and Ross, S. (1991) Spanning, valuation and options. Economic Theory, 1:3–12.
312. Brown, D. and Zhang, Z. (1997) Market orders and market efficiency. Journal of Finance,
52:277–306.
313. Brown, J., Ivkovic, Z., Smith, P. and Weisbenner, S. (2008) Neighbors matter: causal
community effects and stock market participation. Journal of Finance, 63:1509–1531.
314. Brown, S. and Goetzmann, W. (1995) Performance persistence. Journal of Finance, 50:679–
698.
315. Brown, S., Goetzmann, W. and Ross, S. (1995) Survival. Journal of Finance, 50:853–873.
316. Brown, S. and Weinstein, M. (1983) A new approach to testing asset pricing models: the
bilinear paradigm. Journal of Finance, 38:711–743.
317. Browning, M. and Crossley, T. (2001) The life–cycle model of consumption and saving.
Journal of Economic Perspectives, 15:3–22.
318. Brunnermeier, M. (2005) Information leakage and market efficiency. Review of Financial
Studies, 18:417–457.
319. Brunnermeier, M. (2001) Asset Pricing Under Asymmetric Information. Oxford University
Press.
320. Brunnermeier, M. and Nagel, S. (2008) Do wealth fluctuations generate time-varying risk
aversion? Micro-evidence on individuals’ asset allocation. The American Economic Review,
98:713–736.
321. Buckley, G. and Tonks, I. (1989) Are UK stock prices excessively volatile? Trading rules
and variance bounds tests. The Economic Journal, 99:1083–1098.
322. Buckley, G. and Tonks, I. (1992) Trading rules and excess volatility. Journal of Financial
and Quantitative Analysis, 27:365–377.
323. Burmeister, E. and McElroy, M. (1998) Joint estimation of factor sensitivities and risk
premia for the arbitrage pricing theory. Journal of Finance, 43:721–733.
324. Caballé, J. and Krishnam, M. (1994) Imperfect competition in a multi–security market with
risk neutrality. Econometrica, 62:695–704.
325. Caballero, R. (1990) Consumption puzzles and precautionary savings. Journal of Monetary
326. Cabrales, A. and Hoshi, T. (1996) Heterogeneous beliefs, wealth accumulation, and asset
price dynamics. Journal of Economic Dynamics and Control, 20:1073–1100.
327. Camerer, C. (1989) Bubbles and fads in asset pricing. Journal of Economic Surveys, 3:3–41.
328. Camerer, C. (1995) Individual decision making. In The Handbook of Experimental Eco-
nomics, J. Hagel and A. Roth (eds.), Princeton University Press.
329. Camerer, C. and Weber, M. (1992) Recent developments in modeling preferences: uncer-
tainty and ambiguity. Journal of Risk and Uncertainty, 5:325–370.
330. Campbell, J. (1987) Stock returns and the term structure. Journal of Financial Economics,
18:373–399.
331. Campbell, J. (1991) A variance decomposition for stock returns. The Economic Journal,
101:157–179.
332. Campbell, J. (1993) Intertemporal asset pricing without consumption data. The American
333. Campbell, J. (1996) Understanding risk and return. Journal of Political Economy, 104:298–
343.
334. Campbell, J. (1998) Asset prices, consumption, and the business cycle. In Handbook of
Macroeconomics, J.B. Taylor and M. Woodford (eds.), North-Holland, Amsterdam.
778 References
335. Campbell, J. (2000) Asset pricing at the millennium. Journal of Finance, 55:1515–1567.
336. Campbell, J. (2003) Consumption-based asset pricing. In Handbook of Economics and
Finance, G. Constantinides, M. Harris and R. Stulz (eds.), North-Holland, Amsterdam, 803–
887.
337. Campbell, J. (2006) Household finance. Journal of Finance, 61:1553–1603.
338. Campbell, J. and Ammer, J. (1993) What moves the stock and bond markets? A variance
decomposition for long-term asset returns. Journal of Finance, 48:3–36.
339. Campbell, J., Chan, L., Gomes, F. and Maenhout, P. (2001) Investing retirement wealth: a
life cycle model. In Risk Aspects of Investment Based Social Security Reform, J.Y. Campbell
and M. Feldstein (eds.), University of Chicago Press.
340. Campbell, J., Chan, L. and Viceira, L. (2004) A multivariate model of strategic asset
allocation. Journal of Financial Economics, 67:41–80.
341. Campbell, J. and Cocco, J. (2005) How do house prices affect consumption? Evidence from
micro data. Journal of Financial Economics, 54:591–621.
342. Campbell, J., Cocco, J., Gomes, F., Maenhout, P. and Viceira, L. (2001) Stock market
mean reversion and the optimal equity allocation of a long lived investor. European Finance
Review, 5:269–292.
343. Campbell, J. and Cochrane, J. (1999) By force of habit: a consumption-based explanation of
aggregate stock market behavior. Journal of Political Economy, 107:205–251.
344. Campbell, J. and Cochrane, J. (2000) Explaining the poor performance of consumption-
based asset pricing models. Journal of Finance, 55:2863–2878.
345. Campbell, J. and Deaton, A. (1989) Why is consumption so smooth? Review of Economic
Studies, 56:357–373.
346. Campbell, J., Grossman, S. and Wang, J. (1993) Trading volume and serial correlation in
stock returns. Quarterly Journal of Economics, 108:905–939.
347. Campbell, J. and Kyle, A. (1993) Smart money, noise trading and stock price behaviour.
348. Campbell, J., Lo, A. and MacKinlay, C. (1997) The Econometrics of Financial Markets.
Princeton University Press.
349. Campbell, J. and Mankiw, G. (1991) The response of consumption to income: a cross
country investigation. European Economic Review, 35:715–721.
350. Campbell, J., Polk, C. and Vuolteenaho, T. (2009) Growth or glamour? Fundamentals and
systematic risk in stock returns. Review of Financial Studies, 23:305–344.
351. Campbell, J. and Shiller, R. (1988) The dividend–price ratio and expectations of future
dividends and discount factors. Review of Financial Studies, 1:195–228.
352. Campbell, J. and Shiller, R. (1988) Stock prices, earnings, and expected dividends. Journal
of Finance, 43:661–676.
353. Campbell, J. and Shiller, R. (2001) Valuation ratios and the long-run stock market outlook:
an update. NBER working paper, no. 8221.
354. Campbell, J. and Viceira, L. (1999) Consumption and portfolio decisions when expected
returns are time varying. Quarterly Journal of Economics, 114:433–493.
355. Campbell, J. and Viceira, L. (2001) Who should buy long-term bonds? The American
356. Campbell, J. and Viceira, L. (2002) Strategic Asset Allocation: Portfolio Choice for Long-
Term Investors. Oxford University Press.
357. Campbell, J. and Thompson, B. (2008) Predicting excess stock returns out of sample: can
anything beat the historical average? Review of Financial Studies, 21:1509–1531.
358. Campbell, J. and Vuolteenaho, T. (2004) Bad beta, good beta. The American Economic
Review, 94:1249–1275.
359. Campbell, Y. and Yogo, M. (2006) Efficient tests of stock return predictability. Journal of
360. Canner, N., Mankiw, G. and Weil, D. (1997) An asset allocation puzzle. The American
References 779
361. Cao, H. (1999) The effect of derivative assets on information acquisition and price behavior
in a rational expectations equilibrium. Review of Financial Studies, 12:131–163.
362. Cao, H. and Ou-Yang, H. (2009) Differences of opinion of public information and
speculative trading in stocks and options. Review of Financial Studies, 22:299–335.
363. Cao, H., Wang, T., Zhang, H. (2005) Model uncertainty, limited market participation, and
asset prices. Review of Financial Studies, 18:1219–1251.
364. Caplin, A. and Lehay, J. (1994) Business as usual, market crashes, and wisdom after the fact.
365. Caplin, A. and Lehay, J. (2001) Psychological expected utility theory and anticipatory
feelings. Quarterly Journal of Economics, 166:55–79.
366. Carceles-Poveda, E. and Giannitsarou, C. (2008) Asset pricing with adaptive learning.
367. Carhart, M. (1997) On persistence in mutual fund performance. Journal of Finance, 52:57–
82.
368. Carroll, C. (1994) How does future income affect current consumption? Quarterly Journal
369. Carroll, C. (1997) Buffer-stock saving and the life cycle/permanent income hypothesis.
370. Carroll, C. (2001) A theory of the consumption function, with and without liquidity
constraints. Journal of Economic Perspectives, 15:23–45.
371. Carroll, C. (2002) Portfolios of the rich. In Household Portfolios, L. Guiso, M. Haliassos
and T. Jappelli (eds.), MIT Press, Cambridge, 389–429.
372. Carroll, C., Overland, J. and Weil, D. (2000) Saving and growth with habit formation. The
373. Carroll, C. and Samwick, A. (1997) The nature of precautionary wealth. Journal of Monetary
374. Carroll, C. and Samwick, A. (1998) How important is precautionary saving? Review of
Economics and Statistics, 80:410–419.
375. Cass, D. (1984) Competitive equilibrium with incomplete financial markets. CARESS
working paper, no.84–09, University of Pennsylvania.
376. Cass, D. and Stiglitz, J. (1970) The structure of investor preferences and assets returns, and
separability in portfolio allocation: a contribution to the pure theory of mutual funds. Journal
of Economic Theory, 2:122–160.
377. Cassano, M. (1999) Learning and mean reversion in asset returns. Quarterly Review of
Economics and Finance, 39:529–545.
378. Cecchetti, S., Lam, P. and Mark, N. (1990) Mean reversion in equilibrium asset prices. The
379. Cecchetti, S., Lam, P. and Mark, N. (1993) The equity premium and the risk-free rate.
380. Cecchetti, S., Lam, P. and Mark, N. (1994) Testing volatility restrictions on the intertemporal
marginal rates of substitution implied by Euler equations and asset returns. Journal of
Finance, 49:123–151.
381. Cecchetti, S., Lam, P. and Mark, N. (2000) Asset pricing with distorted beliefs: are equity
returns too good to be true? The American Economic Review, 90:787–805.
382. Cecchetti, S. and Mark, N. (1990) Evaluating empirical tests of asset pricing models:
alternative interpretations. The American Economic Review, 80:48–52.
383. Chabakauri, G. (2013) Dynamic equilibrium with two stocks, heterogeneous investors, and
portfolio constraints. Review of Financial Studies, 26:3104–3141.
384. Chacko, G. and Viceira, L. (2005) Dynamic consumption and portfolio choice with
stochastic volatility in incomplete markets. Review of Financial Studies, 18:1369–1402.
385. Chae, J. (2005) Trading volume, information asymmetry, and timing information. Journal
of Finance, 60:413–442.
386. Chai, J., Horneff, W., Maurer, R. and Mitchell, O. (2014) Optimal portfolio choice with
annuities and life insurance for retired couples. Review of Finance, 18:147–188.
780 References
387. Chacko, G. and Viceira, L. (2005) Dynamic consumption and portfolio choice with
stochastic volatility in incomplete markets. Review of Financial Studies, 18:1369–1402.
388. Chakraborty, A. and Yilmaz, B. (2004) Manipulation in market order models. Journal of
Financial Markets, 7:187–206
389. Chakraborty, A. and Yilmaz, B. (2004) Informed manipulation. Journal of Economic Theory,
114:132–152.
390. Chakravarty, S. (2001) Stealth-trading: which traders’ trades move stock prices? Journal of
391. Chakravarty, S., and Holden, C. (1995) An integrated model of market and limit orders.
Journal of Financial Intermediation 4:213–241.
392. Chakravarty, S. and McConnell, J. (1999) Does insider trading really move stock prices?
393. Chalmers, J. and Kadlec, G. (1998) An empirical examination of amortized spreads. Journal
394. Chamberlain, G. (1983) Funds, factors and diversification in the arbitrage pricing theory.
395. Chamberlain, G. (1983) A characterization of the distributions that imply mean-variance
utility functions. Journal of Economic Theory, 29:185–201.
396. Chamberlain, G. (1985) Asset pricing in multiperiod securities markets. Econometrica,
56:1283–1300.
397. Chamberlain, G. and Rothschild, M. (1983) Arbitrage, factor structure, and mean variance
analysis on large asset markets. Econometrica, 51:1281–1304.
398. Chamley, C. (2008) On “acquisition of information in financial markets”. Review of
399. Chan, K. (1988) On the contrarian investment strategy. Journal of Business, 61:147–163.
400. Chan, K. (1993) Imperfect information and cross–autocorrelation among stock prices.
401. Chan, K. and Chen, N. (1991) Structural and return characteristics of small and large firms.
402. Chan, K., Chen, N. and Hsieh, H. (1985) An exploratory investigation of the firm size effect.
403. Chan, K., Christie, W. and Schultz, P. (1995) Market structure and the intraday pattern of
bid-ask spreads for NASADQ securities. Journal of Business, 68:35–60.
404. Chan, K. and Fong, W. (2000) Trade size, order imbalance, and the volatility-volume
relation. Journal of Financial Economics, 57:247–273.
405. Chan, K., Hameed, A. and Tong, W. (2000) Profitability of momentum strategies in the
international equity markets. Journal of Financial and Quantitative Analysis, 35:153–172.
406. Chan, L., Jegadeesh, N. and Lakonishok, J. (1995) Evaluating the performance of value
versus glamour stocks: the impact of selection bias. Journal of Financial Economics,
38:269–296.
407. Chan, L., Jegadeesh, N. and Lakonishok, J. (1996) Momentum strategies. Journal of
Finance, 51:1681–1711.
408. Chan, L., Karceski, J. and Lakonishok, J. (1998) The risk and return from factors. Journal
of Financial and Quantitative Analysis, 33:159–185.
409. Chan, Y. and Kogan, L. (2002) Catching up with the Joneses: heterogeneous preferences and
the dynamics of asset prices. Journal of Political Economy, 110:1235–1285.
410. Chapman D. (2002) Does intrinsic habit formation actually resolve the equity premium
puzzle? Review of Economic Dynamics, 5:618–645.
411. Chateauneuf, A. (1991) On the use of capacities in modeling uncertainty aversion and risk
aversion. Journal of Mathematical Economics, 20: 343–369.
412. Chateauneuf, A., Dana, R. and Tallon, J. (2000) Optimal risk-sharing rules and equilibria
with Choquet-expected-utility. Journal of Mathematical Economics, 34:191–214.
413. Chateauneuf, A., Eichberger, J. and Grant, S. (2007) Choice under uncertainty with the best
and worst in mind: neo-additive capacities. Journal of Economic Theory, 137:538–567.
References 781
414. Chau, M. and Vayanos, D. (2008) Strong-form efficiency with monopolistic insiders. Review
415. Chen, H., Ju, N. and Miao, J. (2014) Dynamic asset allocation with ambiguous return
predictability. Review of Economic Dynamics, 17:799–823.
416. Chen, J. (2001) Intertemporal Capm and the cross-section of stock returns. Mimeo.
417. Chen, J., Hong, H. and Stein, J. (2001) Forecasting crashes: trading volume, past returns and
conditional skewness in stock prices. Journal of Financial Economics, 61:311–344.
418. Chen, J., Hong, H. and Stein, J. (2002) Breadth of ownership and stock returns. Journal of
419. Chen, L. and Zhang, L. (2010) A better three-factor model that explains more anomalies.
420. Chen, L. and Ludvigson, S. (2009) Land of addicts? An empirical investigation of habit-
based asset pricing models. Journal of Applied Econometrics, 24:1057–1093.
421. Chen, N. (1991) Financial investment opportunities and the macroeconomy. Journal of
Finance, 46:529–544.
422. Chen, N. (1983) Some empirical tests of arbitrage pricing theory. Journal of Finance,
38:1393–1414.
423. Chen, N. and Ingersoll, J. (1983) Exact pricing in linear factor models with finitely many
assets. Journal of Finance, 38: 985–988.
424. Chen, N., Roll, R. and Ross, S. (1986) Economic forces and the stock market. Journal of
Business, 59:383–403.
425. Chen, N. and Zhang, F. (1998) Risk and return of value stocks. Journal of Business, 71:501–
535.
426. Chen, Z. and Epstein, L. (2002) Ambiguity, risk and asset returns in continuous time.
427. Cheng, S., Nagar, V. and Rajan, M. (2007) Insider trades and private information: the special
case of delayed-disclosure trades. Review of Financial Studies, 20:1833–1864.
428. Chew, H., Karni, E. and Safra, Z. (1987) Risk aversion in the theory of expected utility with
rank dependent probabilities. Journal of Economic Theory, 42:370–381.
429. Chiarella, C. and He, T. (2001) Asset price and wealth dynamics under heterogeneous
expectations. Quantitative Finance, 1:509–526.
430. Cho, J., Shin, J. and Singh, R. (1999) Endogenous informed trading in the presence of trading
costs: theory and evidence. Journal of Financial Markets, 2:273–305.
431. Chopra, N., Lakonishok, J. and Ritter, J. (1992) Measuring abnormal performance: do stock
market overreact? Journal of Financial Economics, 31:235–268.
432. Chordia, T., Huh, S. and Subrahmanyam, A. (2007) The cross-section of expected trading
activity. Review of Financial Studies, 20:709–740.
433. Chordia, T., Huh, S. and Subrahmanyam, A. (2009) Theory-based illiquidity and asset
pricing. Review of Financial Studies, 22:3629–3668.
434. Chordia, T., Roll, R. and Subrahmanyan, A. (1995) Market liquidity and trading activity.
435. Chordia, T., Roll, R. and Subrahmanyan, A. (2011) Recent trends in trading activity and
market quality. Journal of Financial Economics, 101:243–263.
436. Chordia, T. and Shivakumar, L. (2002) Momentum, business cycle and time-varying
expected returns. Journal of Finance, 57:985–1019.
437. Chordia, T. and Subrahmanyan, A. (1995) Market making, the tick size, and payment for
order flow: theory and evidence. Journal of Business, 68:543–573.
438. Chordia, T. and Subrahmanyan, A. (2004) Order imbalance and individual stock returns:
Theory and evidence. Journal of Financial Economics, 72:485–518.
439. Chordia, T., Subrahmanyam, A. and Anshuman, R. (2001) Trading activity and expected
stock returns. Journal of Financial Economics, 59:3–32.
440. Chordia, T. and Swaminathan, B. (2000) Trading volume and cross-autocorrelations in stock
782 References
441. Chowdhry, B. and Nanda, V. (1991) Multimarket trading and market liquidity. Review of
442. Chowdhry, B. and Nanda, V. (1998) Leverage and market stability: the role of margin rules
and price limits. Journal of Business, 71:179–210.
443. Christelis, D., Jappelli, T. and Padula, M. (2010) Cognitive abilities and portfolio choice.
444. Christiano, L., Eichenbaum, M. and Marshall, D. (1991) The permanent income hypothesis
revisited. Econometrica, 59:371–396.
445. Christie, W., Harris, J. and Schultz, H. (1994) Why did NASDAQ market makers stop
avoiding odd-eight quotes? Journal of Finance, 49:1841–1860.
446. Christie, W. and Schultz, H. (1994) Why do NASDAQ market makers avoid odd-eight
quotes? Journal of Finance, 49:1813–1840.
447. Chuang, W. and Lee, B. (2006) An empirical evaluation of the overconfidence hypothesis.
448. Chung, K. L. (1979) Elementary Probability Theory with Stochastic Processes. Springer,
New York.
449. Chung, K., Chuwonganant, C. and McCormick, T. (2004) Order preferencing and market
quality on NASDAQ before and after decimalization. Journal of Financial Economics,
71:581–612.
450. Chung, K., Van Ness, B. and Van Ness, R. (1999) Limit orders and the bid ask spread.
451. Chung, K., Van Ness, B. and Van Ness, R. (2001) Can the treatment of limit orders reconcile
the differences in trading costs between NYSE and NASDAQ Issues? Journal of Financial
452. Chung, K. and Van Ness, R. (2001) Order handling rules, tick size, and the intraday pattern
of bid-ask spreads for Nasdaq stocks. Journal of Financial Economics, 4:143–161.
453. Citanna, A. and Villanacci, A. (2000) Existence and regularity of partially revealing rational
expectations equilibrium in finite economies. Journal of Mathematical Economics, 34:1–26.
454. Clark, A., Frijters, P. and Shields, M. (2008) Relative income, happiness, and utility: An
explanation for the Easterlin paradox and other puzzles. Journal of Economic Literature,
46:95–144.
455. Cocco, J. (2005) Portfolio choice in the presence of housing. Review of Financial Studies,
18:535–567.
456. Cocco, J., Gomes, F. and Maenhout, P. (2005) Consumption and portfolio choice over the
life cycle. Review of Financial Studies, 18:491–533.
457. Cochrane, J. (1988) How big is the random walk in GNP? Journal of Political Economy,
96:893–920.
458. Cochrane, J. (1991) Volatility tests and efficient markets. Journal of Monetary Economics,
27:463–485.
459. Cochrane, J. (1991) A simple test of consumption insurance. Journal of Political Economy,
99:957–976.
460. Cochrane, J. (1991) Production based asset pricing and the link between stock returns and
economic fluctuations. Journal of Finance, 46:209–237.
461. Cochrane, J. (1992) Explaining the variance of price-dividend ratios. Review of Financial
Studies, 5:243–280.
462. Cochrane, J. (1996) A cross-sectional test of an investment-based asset pricing model.
463. Cochrane, J. (1997) Where is the market going? Uncertain facts and novel theories.
464. Cochrane, J. (1999) New facts in finance. Economic Perspectives, 23:36–58.
465. Cochrane, J. (2001) Asset Pricing. Princeton University Press.
466. Cochrane, J. (2008) The dog that did not bark: a defense of return predictability. Review of
References 783
467. Cochrane, J. and Hansen, L. (1992) Asset pricing explorations for macroeconomics. NBER
Annual Report: 117–182.
468. Cochrane, J. and Saa’-Requejo, J. (2000) Beyod arbitrage: good deal asset price bounds in
incomplete markets. Journal of Political Economy, 108:79–119.
469. Cogley, T. (2002) Idiosyncratic risk and the equity premium: evidence from the consumer
expenditure survey. Journal of Monetary Economics, 49:309–334.
470. Cogley, T. and Sargent T. (2008) The market price of risk and the equity premium: A legacy
of the Great Depression? Journal of Monetary Economics, 55:454–476.
471. Cohen, R., Gompers, P. and Vuolteenaho, T. (2002) Who underreacts to cash flow news? Evi-
dence from trading between individuals and institutions. Journal of Financial Economics,
66:409–462.
472. Cohen, R., Polk, C. and Vuolteenaho, T. (2003) The value spread. Journal of Finance,
58:609–641.
473. Coles, J., Loewenstein, U. and Suay, J. (1995) On equilibrium pricing under parameter
uncertainty. Journal of Financial and Quantitative Analysis, 30:347–364.
474. Conlon, J. R. (2004) Finite horizon bubbles robust to higher order knowledge. Econometrica,
72:927–936.
475. Connor, G. (1984) A unified beta pricing theory. Journal of Economic Theory, 34:13–31.
476. Connor, G. (1989) Notes on the arbitrage pricing theory. In Theory of Valuation, S.
Bhattacharya and G. Constantinides (eds.), Rowman & Littlefield, New Jersey.
477. Connor, G. (1995) The three types of factor models: a comparison of their explanatory
power. Financial Analyst Journal, 51:42–46.
478. Connor, G. and Korajczyk, R. (1988) Risk and return in an equilibrium APT: applications of
a new methodology. Journal of Financial Economics, 21:255–289.
479. Connor, G. and Korajczyk, R. (1990) An intertemporal equilibrium beta pricing model.
480. Connor, G. and Korajczyk, R. (1993) A test for the number of factors in an approximate
factor model. Journal of Finance, 48:1263–1290.
481. Connor, G. and Korajczyk, R. (1995) The arbitrage pricing theory and multifactor models
of asset returns. In Handbooks in Operations Research and Management Science, Vol. 9
(Finance), R.A. Jarrow, V. Maksimovic and W.T. Ziemba (eds.), North Holland, Amsterdam.
482. Conrad, J., Hameed, A. and Niden, C. (1994) Volume and autocovariances in short-horizon
individual security returns. Journal of Finance, 49:1305–1329.
483. Conrad, J. and Kaul, G. (1988) Time-variation in expected returns. Journal of Business,
61:409–425.
484. Conrad, J. and Kaul, G. (1989) Mean reversion in short horizon expected returns. Review of
485. Conrad, J. and Kaul, G. (1993) Long-term market overreaction or biases in computed returns.
486. Conrad, J. and Kaul, G. (1998) An anatomy of trading strategies. Review of Financial
Studies, 11:489–519.
487. Conrad, J., Kaul, G. and Nimalendran, M. (1991) Components of short-horizon individual
security returns. Journal of Financial Economics, 29:365–284.
488. Constantinides, G. (1982) Intertemporal asset pricing with heterogeneous consumers and
without demand aggregation. Journal of Business, 55:253–267.
489. Constantinides, G. (1983) Capital market equilibrium with personal tax. Econometrica,
51:611–636.
490. Constantinides, G. (1984) Optimal stock trading with personal taxes. Journal of Financial
Economics 13:65–89.
491. Constantinides, G. (1986) Capital market equilibrium with transaction costs. Journal of
Political Economy, 94:842–862.
492. Constantinides, G. (1989) Theory of valuation: overview and recent developments. In
Theory of Valuation, S. Bhattacharya and G. Constantinides (eds.), Rowman & Littlefield,
New Jersey.
784 References
493. Constantinides, G. (1990) Habit formation: a resolution of the equity premium puzzle.
494. Constantinides, G. (2002) Rational asset prices. Journal of Finance, 57:1567–1591.
495. Constantinides, G., Donaldson, G. and Mehra, R. (1998) Junior can’t borrow: a new
perspective on the equity premium puzzle. Quarterly Journal of Economics, 117:269–296.
496. Constantinides, G. and Duffie, D. (1996) Asset pricing with heterogeneous consumers.
497. Constantinides, G. and Ghosh, A. (2011) Asset pricing tests with long-run risks in
consumption growth. Review of Asset Pricing Studies, 1:96–136.
498. Copeland, T. and Galai, D. (1983) Information effects on the bid-ask spread. Journal of
Finance, 38:1457–1468.
499. Cornell, B. and Sirri, E. (1992) The reaction of investors and stock prices to insider trading.
500. Corwin, S. and Lipson, M. (2000) Order flow and liquidity around NYSE trading halts.
501. Coval, J. and Moskowitz, T. (1999) Home bias at home: local equity preference in domestic
portfolios. Journal of Finance, 54:145–166.
502. Coval, J. and Moskowitz, T. (2001) The geography of investment: informed trading and asset
prices. Journal of Political Economy, 109:811–841.
503. Coval, J. and Shumway, T. (2001) Do behavioral biases affect asset prices? Journal of
Finance, 60:1–34.
504. Cox, J. and Huang, C. (1989) Optimal consumption and portfolio policies when asset prices
follow a diffusion process. Journal of Economic Theory, 49:33–83.
505. Cox, J. and Huang, C. (1991) A vartiational problem occurring in financial economics.
Journal of Mathematical Economics, 20:465–487.
506. Cox, J., Ingersoll, J. and Ross, S. (1985) An intertemporal general equilibrium model of
asset prices. Econometrica, 53:363–384.
507. Cox, J., Ross, S. and Rubinstein, M. (1979) Option pricing: a simplified approach. Journal
508. Cuoco, D. and Liu, H. (2000) Optimal consumption of a divisible good. Journal of Economic
Dynamics and Control, 24:561–613.
509. Cutler, D., Poterba, J. and Summers, L. (1990) Speculative dynamics and the role of
feedback traders. The American Economic Review, 80:63–68.
510. Cutler, D., Poterba, J. and Summers, L. (1991) Speculative dynamics. Review of Economic
Studies, 58:529–546.
511. Cvitanić, J. and Zapatero, F. (2004) Introduction to the Economics and Mathematics of
Financial Markets. MIT Press, Cambridge.
512. Damodaran, A. and Liu, C. (1993) Insider trading as a signal of private information. Review
513. Dana R. (2004) Ambiguity, uncertainty aversion and equilibrium welfare. Economic Theory,
23:569–587.
514. Dana, R. and Jeanblanc-Picqué, M. (2003) Financial Markets in Continuous Time. Springer,
Berlin/Heidelberg.
515. Daniel, K. (2001) The power and size of mean reversion tests. Journal of Empirical Finance,
8:493–535.
516. Daniel, K., Hirshleifer, D. and Subrahmanyam, A. (1998) Investor psychology and security
market under- and overreactions. Journal of Finance, 53:1839–1883.
517. Daniel, K., Hirshleifer, D. and Subrahmanyam, A. (2001) Overconfidence, arbitrage, and
equilibrium asset pricing. Journal of Finance, 56:921–965.
518. Daniel, K., Hirshleifer, D. and Teoh, S. (2002) Investor psychology in capital markets:
evidence and policy implications. Journal of Monetary Economics, 49:139–209.
519. Daniel, K. and Marshall, D. (1997) The equity premium puzzle and the risk-free rate puzzle
at long horizons. Macroeconomic Dynamics, 1:452–484.
References 785
520. Daniel, K. and Titman, S. (1997) Evidence of the characteristics of cross-sectional variation
in stock returns. Journal of Finance, 52:1–33.
521. Daniel, K. and Titman, S. (2006) Market reactions to tangible and intangible information.
522. Daniel, K. and Titman, S. (2012) Testing factor-models explanations of market anomalies.
Critical Finance Review, 1:103–139.
523. Daniel, K., Titman, S. and Wei, J. (2001) Explaining the cros-section of stock returns in
Japan: factors or characteristics. Journal of Finance, 56:743–766.
524. Danthine, J. and Donaldson, J. (2002) Labour relations and asset returns. Review of
525. Datar, V., Naik, N. and Radcliffe, R. (1998) Liquidity and stock returns: an alternative test.
Journal of Financial Markets 1:203–219.
526. Davis, J. (1994) The cross-section of realized stock returns: the pre-COMPUSTAT evidence.
527. Davis, J., Fama, E. and French, K. (2000) Characteristics, covariances, and average returns:
1929–1997. Journal of Finance, 55:389–406.
528. Davis, S., Kubler, F. and Willen, P. (2006) Borrowing costs and the demand for equity over
the life cycle. Review of Economics and Statistics, 88:348–362.
529. Deaton, A. (1991) Saving and liquidity constraints. Econometrica, 59:1221–1248.
530. Deaton, A. (1992) Understanding Consumption. Clarendon Press, Oxford.
531. Deaves, R., Luders, E. and Luo, G. (2009) An experimental test of the impact of overconfi-
dence and gender on trading activity. Review of Finance, 13:555–575.
532. De Bondt, W. and Thaler, R. (1985) Does the stock market overreact? Journal of Finance,
40:793–805.
533. De Bondt, W. and Thaler, R. (1987) Further evidence on investor overreaction and stock
market seasonality. Journal of Finance, 42:557–582.
534. De Bondt, W. and Thaler, R. (1990) Do security analysts overreact? The American Economic
Review, 80:52–57.
535. Debreu, G. (1959) Theory of Value. Yale University Press, New Haven.
536. Dechow, P. and Sloan, R. (1997) Returns to contrarian investment strategies: tests of naive
expectations hypothesis. Journal of Financial Economics, 43:3–27.
537. De Finetti, B. (1952) Sulla preferibilità. Giornale degli Economisti and Annali di Economia,
11:685–709.
538. De Finetti, B. (1970) Teoria della Probabilità. Einaudi, Torino.
539. De Jong, F., Nijman, T. and Roell, A. (1996) Price effects of trading and components of
bid-ask spreads on the Paris Bourse. Journal of Empirical Finance, 3:193–213.
540. De Jong, F. and Rindi, B. (2009) The Microstructure of Financial Markets. Cambridge
University Press.
541. DeJong, D. and Whiteman, C. (1992) The temporal stability of dividends and stock prices:
evidence from the likelihood function. The American Economic Review, 81:600–617.
542. DeJong, D. and Ripoll, M. (2007) Do self-control preferences help explain the puzzling
behavior of asset prices? Journal of Monetary Economics, 54:1035–1050.
543. Dekel, E. (1989) Asset demand without the independence axiom. Econometrica, 57:163–
169.
544. Delbaen, F. and Schachermayer, W. (1994) A general version of the fundamental theorem of
asset pricing. Mathematische Annalen, 300:463–520.
545. Delbaen, F. and Schachermayer, W. (1998) The fundamental theorem of asset pricing for
unbounded stochastic processes. Mathematische Annalen, 312:215–250.
546. De Long, B. and Magin, K. (2009) The U.S. equity return premium: past, present, and future.
The Journal of Economic Perspectives, 23:193–208.
547. De Long, B., Shleifer, A., Summers, L. and Waldmann, R. (1989) The size and incidence of
the losses from noise trading. Journal of Finance, 44:681–696.
786 References
548. De Long, B., Shleifer, A., Summers, L. and Waldmann, R. (1990) Positive feedback
investment strategies and destabilizing rational speculation. Journal of Finance, 45:379–
395.
549. De Long, B., Shleifer, A., Summers, L. and Waldmann, R. (1990) Noise trader risk in
financial markets. Journal of Political Economy, 98:703–737.
550. De Long, B., Shleifer, A., Summers, L. and Waldmann, R. (1991) The survival of noise
traders in financial markets. Journal of Business, 64:1–19.
551. Demange, G. and Laroque, G. (2006) Finance and the Economics of Uncertainty. Blackwell
Publishing.
552. DeMarzo, P., Fishman, M. and Hagerty, K. (1998) The optimal enforcement of insider
trading regulations. Journal of Political Economy, 106:602–632.
553. DeMarzo, P., Kaniel, R. and Kremer, I. (2004) Diversification as a public good: community
effects in portfolio choice. Journal of Finance, 59:1677–1716.
554. DeMarzo, P., Kaniel, R. and Kremer, I. (2008) Relative wealth concerns and financial
bubbles. Review of Financial Studies, 21:19–50.
555. DeMarzo, P. and Skiadas, C. (1998) Aggregation, determinacy, and informational efficiency
for a class of economies with asymmetric information. Journal of Economic Theory, 80:123–
152.
556. DeMarzo, P. and Skiadas, C. (1999) On the uniqueness of fully informative rational
expectations equilibria. Economic Theory, 13:1–24.
557. De Miguel, V., Nogales, F. and Uppal, R. (2014) Stock return serial dependence and out-of-
sample portfolio performance. Review of Financial Studies, 27:1031–1073.
558. De Nardi, M., French, E. and Jones, J. (2010) Why do the elderly save? The role of medical
expenses. Journal of Political Economy, 118:39–75.
559. Dennert, J. (1993) Price competition between market makers. Review of Economic Studies,
60:735–751.
560. De Santis, G. and Gerard, B. (1997) International asset pricing and portfolio diversification
with time varying risk. Journal of Finance, 52:1881–1912.
561. Detemple, J. (1986) Asset pricing in an economy with incomplete information. Journal of
Finance, 61:383–392.
562. Detemple, J. and Giannikos, C. (1996) Asset and commodity prices with multi-attribute
durable goods. Journal of Economic Dynamics and Control, 20:1451–1504.
563. Detemple, J. and Gottardi, P. (1998) Aggregation, efficiency and mutual fund separation in
incomplete markets, Economic Theory, 11:443–455.
564. Detemple, J. and Murthy, S. (1994) Intertemporal asset pricing with heterogeneous beliefs.
565. Detemple, J. and Murthy, S. (1997) Equilibrium asset prices and no-arbitrage with portfolio
566. Detemple, J. and Serrat, A. (2003) Dynamic equilibrium with liquidity constraints. Review
567. Detemple, J. and Zapatero, F. (1991) Asset prices in an exchange economy with habit
formation. Econometrica, 59:1633–1657.
568. Diamond, D. (1985) Optimal release of information by firms. Journal of Finance, 40:1071–
1094.
569. Diamond, D. and Verrecchia, R. (1981) Information aggregation in a noisy rational
expectations economy. Journal of Financial Economics, 9:221–235.
570. Diamond, D. and Verrecchia, R. (1991) Disclosure, liquidity, and the cost of capital. Journal
of Finance, 46:1325–1360.
571. Diamond, P. (1967) The role of a stock market in a general equilibrium model with
technological uncertainty. The American Economic Review, 57:759–776.
572. Díaz, A., Pijoan-Mas, J. and Ríos-Rull, J. (2003) Precautionary savings and wealth
distribution under habit formation preferences. Journal of Monetary Economics, 50:1257–
1291.
References 787
573. Diba, B. and Grossman, H. (1987) On the inception of rational bubbles. Quarterly Journal
574. Diba, B. and Grossman, H. (1988) Explosive rational bubbles in stock prices? The American
575. Diba, B. and Grossman, H. (1988) The theory of rational bubbles in stock prices. The
Economic Journal, 98:746–754.
576. Diether, K., Lee, K. and Werner, I. (2008) Short-sale strategies and return predictability.
577. Diether, K., Malloy, C. and Scherbina, A. (2002) Differences of opinion and the cross section
of stock returns. Journal of Finance, 57:2113–2141.
578. Dimson, E. and Marsh, P. (1999) Murphy’s law and market anomalies. Journal of Portfolio
Management, 23:53–69.
579. Dittmar, R. (2002) Nonlinear pricing kernels, kurtosis preference, and evidence from the
cross-section of equity returns. Journal of Finance, 57:369–403.
580. Doidge, C., Karolyi, A., Lins, K., Miller, D. and Stulz, R. (2009) Private benefits of control,
ownership, and the cross-listing decision. Journal of Finance, 64: 425–466.
581. Dothan, M. (1990) Prices in Financial Markets. Oxford University Press.
582. Dothan, U. and Feldman, D. (1986) Equilibrium interest rates and multiperiod bonds in a
partially observable economy. Journal of Finance, 61:369–382.
583. Doukas, J., Kim, C. and Pantzalis, C. (2006) Divergence of opinion and equity returns.
584. Dow, J. and Gorton, G. (1994) Arbitrage chains. Journal of Finance, 49:819–849.
585. Dow, J. and Gorton, G. (1995) Profitable informed trading in a simple general equilibrium
model of asset pricing. Journal of Economic Theory, 67:327–369.
586. Dow, J. and Rahi, R. (2003) Informed trading, investment, and welfare. Journal of Business,
76:439–454.
587. Dow, J. and Rahi, R. (2000) Should speculators be taxed? Journal of Business, 73:89–107.
588. Dow, J. and Werlang, S. (1992) Uncertainty aversion, risk aversion, and the optimal choice
of portfolio. Econometrica, 60:197–204.
589. Dow, J. and Werlang, S. (1992) Excess volatility of stock prices and Knightian uncertainty.
590. Drees, B. and Eckwert, B. (1997) Intrinsic bubbles and asset price volatility. Economic
Theory, 9:499–510.
591. Duarte, J. and Young, L. (2009) Why is PIN priced? Journal of Financial Economics,
91:119–138.
592. Duffie, D. (1987) Stochastic equilibria with incomplete financial markets. Journal of
593. Duffie, D. (1996) Dynamic Asset Pricing Theory. Princeton University Press.
594. Duffie, D. and Epstein, L. (1992) Stochastic differential utility. Econometrica, 60:353–394.
595. Duffie, D. and Epstein, L. (1992) Asset pricing with stochastic differential utility. Review of
596. Duffie, D. and Huang, C. (1985) Implementing Arrow-Debreu equilibria by continuous
trading of few long-lived securities. Econometrica, 53:1337–1356.
597. Duffie, D. and Rahi, R. (1995) Financial market innovation and security design: an
introduction. Journal of Economic Theory, 65:1–42.
598. Duffie, D. and Shafer, W. (1985) equilibrium in incomplete markets I. A basic model of
generic existence. Journal of Mathematical Economics, 14:285–300.
599. Duffie, D. and Shafer, W. (1986) Equilibrium in incomplete markets II. Generic existence in
stochastic economies. Journal of Mathematical Economics, 15:199–216.
600. Duffie, D. and Zame, W. (1989) The consumption-based capital asset pricing model.
601. Dumas, B., Kurshev, A. and Uppal, R. (2009) Equilibrium portfolio strategies in the presence
of sentiment risk and excess volatility. Journal of Finance, 64:579–629.
788 References
602. Dunn, K. and Singleton, K. (1986) Modeling the term structure of interest rates under non-
separability of preferences and durability of goods. Journal of Financial Economics, 17:27–
55.
603. Dutta, P. and Madhavan, A. (1995) Price continuity rules and insider trading. Journal of
Financial and Quantitative Analysis, 30:199–223.
604. Dutta, P. and Madhavan, A. (1997) Competition and collusion in dealer markets. Journal of
Finance, 52:245–276.
605. Dybvig, P. (1983) An explicit bound on individual assets’ deviations from Apt pricing in a
finite economy. Journal of Financial Economics, 12:483–496.
606. Dybvig, P. (1984) Short sales restrictions and kinks on the mean–variance frontier. Journal
of Finance, 39:239–244.
607. Dybvig, P. and Ingersoll, J. (1982) Mean-variance theory in complete markets. Journal of
Business, 55:233–251.
608. Dybvig, P. and Liu, H. (2010) Lifetime consumption and investment: Retirement and
constrained borrowing. Journal of Economic Theory, 145:885–907.
609. Dybvig, P. and Ross, S. (1985) Yes, the APT is testable. Journal of Finance, 40:1173–1187.
610. Dybvig, P. and Ross, S. (1986) Tax clienteles and asset pricing. Journal of Finance, 41:751–
763.
611. Dybvig, P. and Ross, S. (1992) Arbitrage. In The New Palgrave Dictionary of Money and
Finance, J. Eatwell, M. Milgate and P. Newman (eds.), Palgrave Macmillan, London.
612. Dynan, K. (1993) How prudent are consumers? Journal of Political Economy, 101:1104–
1113.
613. Dynan, K. (2000) Habit formation in consumer preferences: evidence from panel data. The
614. Dynan, K., Skinner, J. and Zeldes, S. (2004) Do the rich save more? Journal of Political
Economy, 112:397–444.
615. Easley, D., Hvidkjaer, S. and O’Hara, M. (2002) Is information risk a determinant of asset
returns? Journal of Finance, 57:2185–2221.
616. Easley, D., Kiefer, N. and O’Hara, M. (1997) The information content of the trading process.
Journal of Empirical Finance, 4:159–186.
617. Easley, D., Kiefer, N. and O’Hara, M. (1997) One day in the life of a very common stock.
618. Easley, D., Kiefer, N., O’Hara, M. and Paperman, J. (1996) Liquidity, information, and
infrequently traded stocks. Journal of Finance, 51:1405–1432.
619. Easley, D. and O’Hara, M. (1987) Price, trade size, and information in securities markets.
620. Easley, D. and O’Hara, M. (2004) Information and the cost of capital. Journal of Finance,
59:1553–1583.
621. Easley, D. and O’Hara, M. (2009) Ambiguity and nonparticipation: the role of regulation.
622. Easley, D. and O’Hara, M. (2010) Liquidity and valuation in an uncertain world. Journal of
623. Easley, D. and O’Hara, M. (2010) Microstructure and ambiguity. Journal of Finance,
65:1817–1846.
624. Eckbo, E. and Smith, D. (1998) The conditional performance of insider trades. Journal of
Finance, 53:467–498.
625. Eckwert, B. and Zilcha, I. (2001) The value of information in production economies. Journal
626. Eckwert, B. and Zilcha, I. (2003) Incomplete risk sharing arrangements and the value of
information. Economic Theory, 21:43–58.
627. Edelen, R. and Gervais, S. (2003) The role of trading halts in monitoring a specialist market.
628. Ederington, L. and Lee, H. (1993) How markets process information: news releases and
volatility. Journal of Finance, 48:1161–1191.
References 789
629. Edwards, R. (2008) Health risk and portfolio choices. Journal of Business & Economic
Statistics, 26:472–485.
630. Eeckhoudt, L. and Gollier, C. (2000) The effects of change in risk on risk taking: A survey.
Mimeo.
631. Eeckhoudt, L., Gollier, C. and Schlesinger, H. (2005) Economic and Financial Decisions
under Risk. Princeton University Press.
632. Eeckhoudt, L. and Kimball, M. (1992) Background risk, prudence and the demand for
insurance. In Contributions to Insurance Economics, G. Dionne (ed.), Kluwer Academic
Publishers.
633. Eichenbaum, M. and Hansen, L. (1990) Estimating models with intertemporal substitution
using aggregate time-series data. Journal of Business and Economic Statistics, 8: 53–69.
634. Eichenbaum, M., Hansen L. and Singleton, K. (1988) A time series analysis of representative
agent models of consumption and leisure. Quarterly Journal of Economics, 103:51–78.
635. Eleswarapu, V. (1997) Cost of transacting and expected returns in the NASDAQ market.
636. Ellsberg, D. (1961) Risk, ambiguity, and the Savage axioms. Quarterly Journal of Eco-
nomics, 75:647–679.
637. Elmendorf, D. and Kimball, M. (2000) Taxation of labor income and the demand for risky
assets. International Economic Review, 41:801–833.
638. Elton, E., Gruber, M. and Blake, C. (1995) Fundamental economic variables, expected
returns, and bond fund performance. Journal of Finance, 50:1229–1257.
639. Elton, E., Gruber, M. and Busse, J. (1998) Do investors care about sentiment? Journal of
Business, 71:477–499.
640. Epstein, L. (1999) A definition of uncertainty aversion. Review of Economic Studies, 66:579–
608.
641. Epstein, L. (2001) Sharing ambiguity. The American Economic Review, 91:45–50.
642. Epstein, L. and Melino, A. (1995) A revealed preference analysis of asset pricing under
recursive utility. Review of Economic Studies, 62:597–618.
643. Epstein, L. and Miao, J. (2003) A two-person dynamic equilibrium under ambiguity. Journal
of Economic Dynamics and Control, 27:1253–1288.
644. Epstein, L. and Schneider, M. (2007) Learning under ambiguity. Review of Economic
Studies, 74:1275–1303.
645. Epstein, L. and Schneider, M. (2008) Ambiguity, information quality, and asset pricing.
646. Epstein, L. and Wang, T. (1994) Intertemporal asset pricing under Knightian uncertainty.
647. Epstein, L. and Wang, T. (1995) Uncertainty, risk-neutral measures and security price booms
and crashes. Journal of Economic Theory, 67:40–82.
648. Epstein, L. and Zhang, J. (2001) Subjective probabilities on subjectively unambiguous
events. Econometrica, 69: 265–306.
649. Epstein, L. and Zin, S. (1989) Substitution, risk aversion, and the temporal behavior of
consumption and asset returns I: a theoretical framework. Econometrica, 57:937–969.
650. Epstein, L. and Zin, S. (1990) ‘First-order’ risk aversion and the equity premium puzzle.
651. Epstein, L. and Zin, S. (1991) Substitution, risk aversion, and the temporal behavior of
consumption and asset returns II: an empirical analysis. Journal of Political Economy,
57:937–969.
652. Epstein, L. and Zin, S. (2001) The independence axiom and asset returns. Journal of
Empirical Finance, 6:537–572.
653. Etner, J., Jeleva, M. and Tallon, J.-M. (2012) Decision theory under ambiguity. Journal of
Economic Surveys, 26:234–270.
654. Evans, G. (1986) Selection criteria for models with non-uniqueness. Journal of Monetary
790 References
655. Evans, G. (1989) The fragility of sunspots and bubbles. Journal of Monetary Economics,
23:297–317.
656. Evans, G. (1991) Pitfalls in testing for explosive bubbles in asset prices. The American
657. Evans, G. and Honkapohja, S. (1998) Learning dynamics. In Handbook of Macroeconomics,
J. Taylor and M. Woodford (eds.), North Holland, Amsterdam.
658. Evans, M. (1994) Expected returns, time-varying risk, and risk premia. Journal of Finance,
49:654–679.
659. Evans, M. (1998) Dividend variability and stock market swings. Review of Economic
Studies, 65:711–740.
660. Fama, E. (1965) The behaviour of stock market prices. Journal of Business, 38:34–105.
661. Fama, E. (1970) Efficient capital markets: a review of theory and empirical work. Journal of
Finance, 25:383–417.
662. Fama, E. (1990) Stock returns, expected returns, and real activity. Journal of Finance,
45:1089–1107.
663. Fama, E. (1991) Efficient capital markets: II. Journal of Finance, 46:1575–1618.
664. Fama, E. (1998) Market efficiency, long-term returns, and behavioural finance. Journal of
665. Fama, E. and Blume, M. (1966) Filter rules and stock market trading profits. Journal of
Business, 39:226–241.
666. Fama, E. and French, K. (1988) Permanent and temporary components of stock prices.
667. Fama, E. and French, K. (1988) Dividend yields and expected stock returns. Journal of
668. Fama, E. and French, K. (1989) Business conditions and expected returns on stocks and
bonds. Journal of Financial Economics, 25:23–49.
669. Fama, E. and French, K. (1992) The cross-section of expected stock returns. Journal of
Finance, 47:427–465.
670. Fama, E. and French, K. (1993) Common risk factors in the returns of stocks and bonds.
671. Fama, E. and French, K. (1995) Size and book to market factors in earnings and returns.
672. Fama, E. and French, K. (1996) The CAPM is wanted, dead or alive. Journal of Finance,
51:1947–1958.
673. Fama, E. and French, K. (1996) Multifactor explanations of asset pricing anomalies. Journal
of Finance, 51:55–85.
674. Fama, E. and French, K. (1998) Value versus growth: the international evidence. Journal of
Finance, 53:1975–1999.
675. Fama, E. and French, K. (2002) The equity premium. Journal of Finance, 57:637–660.
676. Fama, E. and French, K. (2006) The Value Premium and the CAPM. Journal of Finance,
61:2163–2186.
677. Fama, E. and French, K. (2012) Size, value, and momentum in international stock returns.
Journal of Financial Economics, 105: 457–472.
678. Fama, E. and MacBeth, J. (1973) Risk, return and equilibrium: empirical tests. Journal of
679. Fama, E. and Schwert, G. (1977) Asset returns and inflation. Journal of Financial Eco-
nomics, 5:115–146.
680. Farhi, E. and Panageas, S. (2007) Saving and investing for early retirement: A theoretical
analysis. Journal of Financial Economics, 83: 87–121.
681. Faure-Grimaud, A. and Gromb, D. (2004) Public trading and private incentives. Review of
682. Fernandes, N. and Ferreira, M. (2009) Insider trading laws and stock price informativeness.
683. Fernholz, R. (2002) Stochastic Portfolio Theory. Springer, New York.
References 791
684. Ferris, S., Haugen, R. and Makhija, A. (1988) Predicting contemporary volume with historic
volume at differential price levels: evidence supporting the disposition effect. Journal of
Finance, 43:677–697.
685. Ferson, W. (1989) Changes in expected security returns, risk, and the level of interest rates.
686. Ferson, W. (1990) Are the latent variables in time varying expected returns compensation
for consumption risk? Journal of Finance, 45:397–427.
687. Ferson, W. (1995) Theory and empirical testing of asset pricing models. In Handbooks
in Operations Research and Management Science, Vol. 9 (Finance), R.A. Jarrow, V.
Maksimovic and W.T. Ziemba (eds.), North Holland, Amsterdam.
688. Ferson, W. (2003) Tests of multifactor pricing models, volatility bounds and portfolio
performance. In Handbook of the Economics of Finance, Vol. 21, G. Constantinides, M.
Harris and R. Stulz (eds.), Elsevier Science Publisher, 743–802.
689. Ferson, W. and Constantinides, G. (1991) Habit persistence and durability. Journal of
690. Ferson, W., Foerster, S. and Keim, D. (1993) General tests of latent variable models and
mean variance spanning. Journal of Finance, 48:131–156.
691. Ferson, W. and Harvey, C. (1991) The variation of economic risk premiums. Journal of
692. Ferson, W. and Harvey, C. (1992) Seasonality and consumption-based asset pricing. Journal
of Finance, 47:511–551.
693. Ferson, W. and Harvey, C. (1999) Conditioning variables and the cross-section of stock
694. Ferson, W., Kandel, S. and Stambaugh, R. (1987) Tests of asset pricing with time-varying
expected risk premiums and market betas. Journal of Finance, 42:201–220.
695. Ferson, W. and Korajczyk, R. (1995) Do arbitrage pricing models explain the predictability
of stock returns? Journal of Business, 68:309–349.
696. Ferson, W. and Jagannathan, R. (1996) Econometric evaluation of asset pricing models.
In Handbook of Statistics (Vol. 14), G.S. Maddala and C.R. Rao (eds.), Elsevier Science
Publisher.
697. Ferson, W., Sarkissian, S. and Simin, T. (1999) The alpha factor asset pricing model: a
parable. Journal of Financial Markets, 2:49–68.
698. Ferson, W. and Siegel, A. (2001) The efficient use of conditioning information in portfolios.
699. Ferson, W. and Siegel, A. (2003) Stochastic discount factor bounds with conditioning
information. Review of Financial Studies, 16:567–595.
700. Fielding, D. and Stracca, L. (2007) Myopic loss aversion, disappointment aversion, and the
equity premium puzzle. Journal of Economic Behavior and Organization, 64:250–268.
701. Figlewski, S. (1978) Market “efficiency” in a market with heterogeneous information.
702. Figlewski, S. (1979) Subjective information and market efficiency in a betting market.
703. Figlewski, S. (1981) The informational effects of restrictions on short sales: some empirical
evidence. Journal of Financial and Quantitative Analysis 16:463–476.
704. Fischer, P. (1992) Optimal contracting and insider trading restrictions. Journal of Finance,
47:673–694.
705. Fischer, P. and Verrecchia, R. (1999) Public information and heuristic trade. Journal of
Accounting and Economics, 27:89–124.
706. Fishburn, P. (1970) Utility Theory for Decision Making. Wiley, New York.
707. Fishburn, P. and Porter, B. (1976) Optimal portfolios with one safe and one risky asset:
effects of changes in rate of return and risk. Management Science, 22:1064–1073.
708. Fishman, M. and Hagerty, K. (1989) Disclosure decisions by firms and the competition for
price efficiency. Journal of Finance, 46:633–646.
792 References
709. Fishman, M. and Hagerty, K. (1992) Insider trading and the efficiency of stock prices. Rand
710. Fishman, M. and Hagerty, K. (1995) The mandatory disclosure of trades and market
liquidity. Review of Financial Studies, 8:637–676.
711. Fishman, M. and Hagerty, K. (1995) The incentive to sell financial market information.
Journal of Financial Intermediation, 4:95–115.
712. Fishman, M. and Longstaff, F. (1992) Dual trading in futures markets. Journal of Finance,
47:643–671.
713. Flannery, M. and Protopapadakis, A. (2002) Macroeconomic factors do influence aggregate
stock returns. Review of Financial Studies, 15:751–782.
714. Flavin, M. (1981) The adjustment of consumption to changing expectations about future
income. Journal of Political Economy, 89:974–1009.
715. Flavin, M. (1983) Excess volatility in financial markets: a reassessment of the empirical
evidence. Journal of Political Economy, 91:929–956.
716. Flavin, M. and Nakagawa, S. (2008) A model of housing in the presence of adjustment costs:
a structural interpretation of habit persistence. The American Economic Review, 98:474–495.
717. Flood, M., Huisman, R., Koedijk, K. and Mahieu, R. (1999) Quote disclosure and price
discovery in multiple-dealer financial markets. Review of Financial Studies, 12:37–59.
718. Flood, R. and Hodrick, R. (1986) Asset price volatility, bubbles, and prices switching.
719. Flood, R. and Hodrick, R. (1990) On testing for speculative bubbles. Journal of Economic
Perspective, 4:85–101.
720. Flood, R., Hodrick, R. and Kaplan, P. (1986) An evaluation of recent evidence on stock
market bubbles. NBER working paper, no.1971.
721. Föllmer, H. and Schied, A. (2011) Stochastic Finance - An Introduction in Discrete Time.
3rd edition. Walter de Gruyter, Berlin.
722. Föllmer, H. and Schweizer, M. (1991) Hedging of contingent claims under incomplete
information. In Applied Stochastic Analysis, M.H.A. Davis and R.J. Elliott (eds.), Gordon &
Breach, London/New York, 389–414.
723. Föllmer, H. and Sondermann, D. (1986) Hedging of non-redundant contingent claims. In
Contributions to Mathematical Economics, W. Hildenbrand and A. Mas-Colell (eds.), North
Holland, Amsterdam, 205–223.
724. Fontana, C. (2015) Weak and strong no-arbitrage conditions for continuous financial
markets. International Journal of Theoretical and Applied Finance, 18:1550005.
725. Fontana, C. and Schweizer, M. (2012) Simplified mean-variance portfolio optimisation.
Mathematics and Financial Economics, 6:125–153.
726. Fornari, F. (2002) The size of the equity premium. Temi di Discussione no.447, Banca
d’Italia.
727. Forster, M. and George, T. (1992) Anonymity in securities markets. Journal of Financial
Intermediation, 2:168–206.
728. Forsythe, R. and Lundholm, R. (1990) Information aggregation in an experimental market.
729. Foster, D. and Viswanathan, S. (1990) A theory of intraday variations in volume, variance
and trading costs in securities markets. Review of Financial Studies, 3:593–624.
730. Foster, D. and Viswanathan, S. (1993) The effect of public information and competition on
trading volume and price volatility. Review of Financial Studies, 6:23–56.
731. Foster, D. and Viswanathan, S. (1993) Variations in trading volume, returns volatility, and
trading costs: evidence of recent price formation models. Journal of Finance, 48:187–211.
732. Foster, D. and Viswanathan, S. (1996) Strategic trading when agents forecast the forecasts
of others. Journal of Finance, 51:1437–1478.
733. Foucault, T. (1999) Order flow composition and trading costs in a dynamic limit order
market. Journal of Financial Markets, 2:99–134.
734. Foucault, T., Kadan, O. and Kandel, E. (2005) Limit order book as a market for liquidity.
References 793
735. Foucault, T., Pagano, M. and Röell, A. (2013) Market Liquidity: Theory, Evidence and
Policy. Oxford University Press.
736. Frederick, S., Loewenstein, G. and O’Donoghue, T. (2002) Time discounting and time
preference: a critical review. Journal of Economic Literature, 40:351–401.
737. French, E. (2005) The effects of health, wealth, and wages on labour supply and retirement
behaviour. Review of Economic Studies, 72:395–427.
738. French, E. and Jones, J.B. (2011) The effects of health and self-insurance on retirement
behavior. Econometrica, 79:693–732.
739. French, K. (1980) Stock returns and the weekend effect. Journal of Financial Economics,
8:55–69.
740. French, K. and Poterba, J. (1991) Investor diversification and international equity markets.
741. French, K. and Roll, R. (1986) Stock return and variances: the arrival of information and
reaction of traders. Journal of Finance, 17:5–26.
742. French, K., Schwert, G. and Stambaugh, R. (1987) Expected stock returns and volatility.
743. Friedman, M. (1953) Essays in Positive Economics, University of Chicago Press.
744. Friedman, M. and Savage, L. (1948) The utility analysis of choice involving risk. Journal of
745. Froot, K. and Obstfeld, M. (1991) Intrinsic bubbles: the case of stock prices. The American
746. Fuydenberg, D. and Tirole, J. (1991) Game Theory. MIT Press, Cambridge.
747. Gabaix, X. (2012) Variable rare disasters: an exactly solved framework for ten puzzles in
macro-finance. Quarterly Journal fo Economics, 127:645–700.
748. Gabaix, X. and Laibson, D. (2001) The 6D bias and the equity premium puzzle. NBER
Macroeconomics Annual 2001, 257–328.
749. Gali, J. (1994) Keeping up with the Joneses: consumption externalities, portfolio choice, and
asset prices. Journal of Money, Credit, and Banking, 26:1–9.
750. Gallant, R., Hansen, L. and Tauchen, G. (1990) Using conditional moments of asset
payoffs to infer the volatility of intertemporal marginal rates of substitution. Journal of
Econometrics, 45:141–179.
751. Gallant, R., Rossi, P. and Tauchen, G. (1992) Stock prices and volume. Review of Financial
Studies, 5:199–242.
752. Gallant, R., Rossi, P. and Tauchen, G. (1993) Nonlinear dynamic structures. Econometrica,
61:871–907.
753. Gallant, R. and Tauchen, G. (1989) Seminonparametric estimation of conditionally con-
strained heterogeneous processes: asset pricing applications. Econometrica, 57:1091–1120.
754. Ganguli, J. V. and Yang, L. (2009) Complementarities, multiplicity, and supply information.
Journal of the European Economic Association, 7:90–115.
755. Garcia, D. and Sangiorgi, F. (2011) Information sales and strategic trading. Review of
756. Garcia, D. and Strobl, G. (2011) Relative wealth concerns and complementarities in
information acquisition. Review of Financial Studies, 24:169–207.
757. Garcia, D. and Vanden, J. (2009) Information acquisition and mutual funds. Journal of
758. Garfinkel, J. and Nimalendran, M. (2003) Market structure and trader anonymity: an analysis
of insider trading. Journal of Financial and Quantitative Analysis, 38:591–610.
759. Garlappi, L., Uppal, R. and Wang, T. (2007) Portfolio selection with parameter and model
uncertainty: a multi-prior approach. Review of Financial Studies, 20:41–81.
760. Garleanu, N. (2009) Portfolio choice and pricing in illiquid markets. Journal of Economic
Theory, 144:532–564.
761. Garleanu, N. and Pedersen, L.H. (2013) Dynamic trading with predictable returns and
transaction costs. Journal of Finance, 68:2309–2340.
794 References
762. Gaunersdorfer, A. (2000) Endogenous fluctuations in a simple asset pricing model with
heterogeneous agents. Journal of Economic Dynamics and Control, 24:799–831.
763. Geanakoplos, J. (1990) An introduction to general equilibrium with incomplete markets.
764. Geanakoplos, J. and Mas-Colell, A. (1989) Real indeterminancy with financial assets.
765. Geanakoplos, J. and Polemarchakis, H. (1986) Existence, regularity and constrained subop-
timality of competitive allocations when the asset market is incomplete. In Essays in Honor
of K. Arrow, W. Heller, R. Starr and D. Starrett (eds.), Cambridge University Press, 65–95.
766. Gemmill, G. (1994) Transparency and liquidity: a study of block trades on the London Stock
exchange under different publication rules. Journal of Finance, 51:1765–1790.
767. Gennotte, G. (1986) Optimal portfolio choice under incomplete information. Journal of
Finance, 61:733–749.
768. George, T., Kaul, G. and Nimalendran, M. (1991) Estimation of the bid-ask spread and its
components: a new approach. Review of Financial Studies, 4:623–656.
769. George, T., Kaul, G. and Nimalendran, M. (1994) Trading volume and transaction costs in
specialist markets. Journal of Finance, 49:1489–1505.
770. Gerard, B. and Nanda, V. (1993) Trading and manipulation around seasoned equity offerings.
771. Gerety, M. and Mulherin, J.H. (1992) Trading halts and market activity: an analysis of
volume at the open and the close. Journal of Finance, 47:1765–1784.
772. Gerety, M. and Mulherin, J.H. (1994) Price formation on stock exchanges: the evolution
trading within the day. Review of Financial Studies, 6:609–629.
773. Gervais, S. and Odean, T. (2001) Learning to be overconfident. Review of Financial Studies,
14:1–27.
774. Ghirardato, P. and Marinacci, M. (2002) Ambiguity made precise: A comparative founda-
tion. Journal of Economic Theory, 102:251–289.
775. Gibbons, M. (1982) Multivariate tests of financial models: a new approach. Journal of
776. Gibbons, M. and Ferson, W. (1985) Testing asset pricing models with changing expectations
and an unobservable market portfolio. Journal of Financial Economics, 14:217–236.
777. Gibbons, M., Ross, S. and Shanken, J. (1989) A test of the efficiency of a given portfolio.
778. Gilboa, I. (1987) Expected utility with purely subjective non-additive probabilities. Journal
of Mathematical Economics, 16:65–88.
779. Gilboa, I. and Marinacci, M. (2013) Ambiguity and the Bayesian paradigm. In Advances in
Economics and Econometrics, D. Acemoglu, M. Arellano and E. Dekel (eds.), Cambridge
University Press.
780. Gilboa, I. and Schmeidler, D. (1989) Maxmin expected utility with nonunique prior. Journal
of Mathematical Economics, 18:141–153.
781. Gilles, C. and LeRoy, S. (1991) On the arbitrage pricing theory. Economic Theory, 1:213–
229.
782. Gilles, C. and LeRoy, S. (1991) Econometric aspects of variance-bounds tests: a survey.
783. Gilles, C. and LeRoy, S. (1992) Bubbles and charges. International Economic Review,
33:323–339.
784. Gilles, C. and LeRoy, S. (1997) Bubbles as payoffs at infinity. Economic Theory, 9:261–281.
785. Giovannini, A. and Weil, P. (1989) Risk aversion and intertemporal substitution in the capital
asset pricing model. NBER working paper no.2824.
786. Glaser, M. and Weber, M. (2007) Overconfidence and trading volume. The Geneva Risk and
Insurance Review, 32:1–36.
787. Glosten, L. (1989) Insider trading, liquidity, and the role of the monopolist specialist.
Journal of Business, 62:211–235.
References 795
788. Glosten, L. (1994) Is the electronic open limit order book inevitable? Journal of Finance,
49:1127–1162.
789. Glosten, L. and Harris, L. (1988) Estimating the components of the bid-ask spread. Journal
790. Glosten, L. and Milgrom, P. (1985) Bid, ask and transaction prices in a specialist market
with heterogeneously informed traders. Journal of Financial Economics, 14:71–100.
791. Gneezy, U., Kapteyn, A. and Potters, J. (2003) Evaluation periods and asset prices in a
market experience. Journal of Finance, 58:821–838.
792. Gneezy, U. and Potters, J. (1997) An experiment on risk taking and evaluation periods.
793. Godek, P. (1996) Why Nasdaq market makers avoid odd-eight quotes. Journal of Financial
794. Goettler, R., Parlour, C. and Rajan, U. (2005) Equilibrium in a dynamic limit order market.
795. Goetzmann, W. and Jorion, P. (1993) Testing the predictive power of dividend yields. Journal
of Finance, 48:663–679.
796. Goetzmann, W. and Jorion, P. (1995) A longer look at dividend yields. Journal of Business,
68:482–508.
797. Goetzmann, W. and Jorion, P. (1999) Global stock markets in the twentieth century. Journal
of Finance, 54:953–980.
798. Goldstein, M. and Kavajecz, K. (2000) Eighths, sixteenths, and market depth: changes in tick
size and liquidity provision on the NYSE. Journal of Financial Economics, 56:125–149.
799. Gollier, C. (1995) The comparative statics of changes in risk revisited. Journal of Economic
Theory, 66:522–535.
800. Gollier, C. (2001) The Economics of Risk and Time. MIT Press, Cambridge.
801. Gollier, C. (2001) Wealth inequality and asset pricing. Review of Economic Studies, 68:181–
203.
802. Gollier, C. (2009) Portfolio choices and asset prices. The comparative statics of ambiguity
aversion. Review of Economic Studies, 78:1329–1344.
803. Gollier, C. and Pratt, J. (1996) Risk vulnerability and tempering effect of background risk.
804. Gollier, C. and Schlesinger, H. (2002) Changes in risk and asset prices. Journal of Monetary
805. Gomes, F. (2005) Portfolio choice and trading volume with loss-averse investors. Journal of
Business, 78:675–706.
806. Gomes, F., Kotlikoff, L. and Viceira, L. (2008) Optimal life-cycle investing with flexible
labor supply: a welfare analysis of life-cycle funds. The American Economic Review: Papers
and Proceedings, 98:297–303.
807. Gomes, F. and Michaelides, A. (2003) Portfolio choice with internal habit formation: a life-
cycle model with uninsurable labor income risk. Review of Economic Dynamics, 6:729–766.
808. Gomes, F. and Michaelides, A. (2005) Optimal life cycle asset allocation: understanding the
empirical evidence. Journal of Finance, 60:869–904.
809. Gomes, F. and Michaelides, A. (2008) Asset pricing with limited risk sharing and heteroge-
neous agents. Review of Financial Studies, 21:415–448.
810. Gomes, J., Kogan, L. and Zhang, L. (2003) Equilibrium cross-section of returns. Journal of
811. Gomes, J., Yaron, A. and Zhang, L. (2006) Asset pricing implications of firms’ financing
812. Goodhart, C. and O’Hara, M. (1997) High frequency data in financial markets: issues and
applications. Journal of Empirical Finance, 4:73–114.
813. Gottardi, P. and Rahi, R. (2014) Value of information in competitive economies with
incomplete markets. International Economic Review, 55:1468–2354.
814. Gould, J. and Verrecchia, R. (1985) The information content of specialist pricing. Journal of
796 References
815. Gourinchas, P. and Parker, J. (2002) Consumption over the life cycle. Econometrica, 70:47–
89.
816. Gourio, F. (2007) Disasters and recoveries. The American Economic Review 98:68–73.
817. Goyal, A. and Welch, I. (2008) A comprehensive look at the empirical performance of equity
premium prediction. Review of Financial Studies, 21:1455–1508.
818. Graham, B. and Dodd, D. (1934) Security Analysis. McGraw-Hill, New York.
819. Graham, J., Harvey, C. and Huang, H. (1996) Investor competence, trading frequency, and
home bias. Management Science, 55:1094–1106.
820. Green, J. (1973) Information, efficiency and equilibrium. Discussion paper n.284. Harvard
University.
821. Green, J. (1981) Value of Information with Sequential Futures Markets Econometrica,
49:335–358.
822. Green, J. and Jarrow, R. (1987) Spanning and completeness in markets with contingent
claims. Journal of Economic Theory, 41:202–210.
823. Greene, J. and Smart, S. (1999) Liquidity provision and noise trading: evidence from the
investment dartboard column. Journal of Finance, 54:1885–1899.
824. Greenwald, B. and Stein, J. (1991) Transactional risk, market crashes, and the role of circuit
breakers. Journal of Business, 64:443–462.
825. Griffin, J. (2002) Are the Fama and French factors global or country specific? Review of
826. Griffin, J. and Lemmon, M. (2002) Book-to-market equity, distress risk, and stock returns.
Journal fo Finance, 57:2317–2336.
827. Grinblatt, M. and Han, B. (2005) Prospect theory, mental accounting, and momentum.
828. Griffin, J., Harris, J., Shu, T. and Topaloglu, S. (2011) Who drove and burst the tech bubble?
829. Grinblatt, M. and Keloharju, M. (2001) What makes investors trade? Journal of Finance,
56:589–616.
830. Grinblatt, M. and Keloharju, M. (2004) Tax-loss trading and wash sales. Journal of Financial
831. Grinblatt, M. and Keloharju, M. (2009) Sensation seeking, overconfidence, and trading
activity. Journal of Finance, 69:549–578.
832. Grinblatt, M., Keloharju, M. and Linnainmaa, J. (2011) IQ and stock market participation,
833. Grinblatt, M. and Moskowitz T. (2004) Predicting stock price movements from past returns:
the role of consistency and tax-loss selling. Journal of Financial Economics 71:541–579.
834. Grinblatt, M. and Titman, S. (1983) Factor pricing in a finite economy. Journal of Financial
835. Grinblatt, M. and Titman, S. (1985) Approximate factor structures. Interpretations and
implications for empirical tests. Journal of Finance, 40:1367–1373.
836. Grinblatt, M. and Titman, S. (1987) The relation between mean-variance efficiency and
arbitrage pricing. Journal of Business, 60:97–112.
837. Gromb, D. and Vayanos, D. (2002) Equilibrium and welfare in markets with financially
constrained arbitrageurs. Journal of Financial Economics, 66:361–407.
838. Gross, D. and Souleles, N. (2001) Do liquidity constraints and interest rates matter for
consumer behavior? Evidence from credit card data. Quarterly Journal of Economics,
117:149–186.
839. Grossman, S. (1976) On the efficiency of competitive stock markets where traders have
diverse information. Journal of Finance, 31:573–585.
840. Grossman, S. (1978) Further results on the informational efficiency of competitive stock
markets. Journal of Economic Theory, 18:81–101.
841. Grossman, S. (1981) An introduction to the theory of rational expectations under asymmetric
information. Review of Economic Studies, 48:541–559.
842. Grossman, S. (1989) The Informational Role of Prices. MIT Press, Cambridge.
References 797
843. Grossman, S. (1992) The informational role of upstairs and downstairs trading. Journal of
Business, 65:509–530.
844. Grossman, S. and Laroque, G. (1990) Asset pricing and optimal portfolio choice in the
presence of illiquid durable consumption goods. Econometrica, 58:25–51.
845. Grossman, S., Melino A. and Shiller, R. (1987) Estimating the continuous time consumption
based asset pricing model. Journal of Business and Economic Statistics, 5:315–327.
846. Grossman, S. and Miller, M. (1988) Liquidity and market structure. Journal of Finance,
43:617–637.
847. Grossman, S., Miller, M., Cone, K., Fischel, D. and Ross, D. (1997) Clustering and
competition in dealer markets. Journal of Law and Economics, 40:23–60.
848. Grossman, S. and Shiller, R. (1981) The determinants of the variability of stock markets
prices. The American Economic Review, 71:222–227.
849. Grossman, S. and Stiglitz, J. (1980) On the impossibility of informationally efficient
markets. The American Economic Review, 70:393–408.
850. Grundy, B. and Martin, S. (2001) Understanding the nature of the risks and the source of the
rewards to momentum investing. Review of Financial Studies, 14:29–78.
851. Grundy, B. and Martin, S. (2001) Trade and revelation of information through prices and
direct disclosure. Review of Financial Studies, 2:495–526.
852. Grundy, B. and Kim, Y. (2002) Stock market volatility in a heterogeneous information
economy. Journal of Financial and Quantitative Analysis, 37:1–27.
853. Guesnerie, R. and Jaffray, J. Y. (1974) Optimality of equilibrium of plans, prices and price
expectations. In Allocation under uncertainty: equilibrium and optimality, J. Drèze (ed.),
Palgrave Macmillan, London.
854. Guidolin, M.(2006) Pessimistic beliefs under rational learning: Quantitative implications for
the equity premium puzzle. Journal of Economics and Business, 58:85–118.
855. Guidolin, M. and Timmermann, A. (2007) Properties of equilibrium asset prices under
alternative learning schemes. Journal of Economic Dynamics and Control, 31:161–217.
856. Guidolin, M. and Rinaldi, F. (2013) Ambiguity in asset pricing and portfolio choice: a review
of the literature. Theory and Decisions, 74:183–217.
857. Guiso, L., Haliassos, M. and Jappelli, T. eds. (2001) Household Portfolios. MIT Press,
Cambridge.
858. Guiso, L. and Jappelli, T. (2005) Awareness and stock market participation. Review of
Finance, 9:537–567.
859. Guiso, L., Jappelli, T. and Terlizzese, D. (1992) Earnings uncertainty and precautionary
saving. Journal of Monetary Economics, 30:307–332.
860. Guiso, L., Jappelli, T. and Terlizzese, D. (1996) Income risk, borrowing constraints, and
portfolio choice. The American Economic Review, 86:158–172.
861. Guiso, L., Sapienza, P. and Zingales, L. (2008) Trusting the stock market. Journal of
Finance, 63:2557–2600.
862. Guiso, L. and Sodini, C. (2012) Household finance. An emerging field. Mimeo.
863. Gul, F. (1991) A theory of disappointment aversion. Econometrica, 59:667–686.
864. Gul, F. and Pesendorfer, W. (2001) Temptation and self-control. Econometrica, 69:1403–
1435.
865. Gul, F. and Pesendorfer, W. (2004) Self-control and the theory of consumption. Economet-
rica, 72:119–158.
866. Gurkaynak, R. (2008) Econometric tests of asset price bubbles: taking stock. Journal of
Economic Surveys, 22:166–186.
867. Guvenen, F. (2006) Reconciling conflicting evidence on the elasticity of intertemporal
substitution: A macroeconomic perspective. Journal of Monetary Economics, 53:1451–
1472.
868. Guvenen, F. (2009) A parsimonious macroeconomic model for asset pricing. Econometrica,
77:1711–1750.
869. Hadar, J. and Russell, W. (1971) Stochastic dominance and diversification. Journal of
798 References
870. Hadar, J. and Russell, W. (1974) Diversification of interdependent prospects. Journal of

871. Hadar, J., Russell, W. and Seo, T. (1977) Gains from diversification. Review of Economic
Studies, 44:363–368.
872. Hadar, J. and Seo, T. (1990) The effects of shifts in a return distribution on optimal portfolios.
International Economic Review, 31:721–736.
873. Hagerty, K. (1991) Equilibrium bid-ask spreads in markets with multiple assets. Review of
874. Hahn, F. (1973) On the Notion of Equilibrium in Economics. Cambridge University Press.
875. Hakansson, N. (1970) Optimal investment and consumption under risk for a class of utility
functions. Econometrica, 38:587–607.
876. Haliassos, M. and Bertaut, C. (1995) Why does so few hold stocks? The Economic Journal,
105:1110–1129.
877. Haliassos, M. and Michaelides, A. (2003) Portfolio choice and liquidity constraints.
International Economic Review, 44:143–177.
878. Hall, R. (1978) Stochastic implications of the life cycle-permanent income hypothesis:
theory and evidence. Journal of Political Economy, 86:971–987.
879. Hall, R. and Mishkin, F. (1982) The sensitivity of consumption to transitory: estimates from
panel data households. Econometrica, 50:461–481.
880. Hall, R. (1988) Intertemporal substitution in consumption. Journal of Political Economy,
96:339–357.
881. Hamilton, J. (1994) Time Series Analysis. Princeton University Press.
882. Hamilton, J. and Whiteman, C. (1985) The observable implications of self-fulfilling
expectations. Journal of Monetary Economics, 16:353–373.
883. Han, Y. (2006) Asset allocation with a high dimensional latent factor stochastic volatility
model. Review of Financial Studies, 19:237–271.
884. Han, Y., Ke, Y. and Zhou G. (2013) A new anomaly: the cross-sectional profitability of
technical analysis. Journal of Financial and Quantitative Analysis, 48:1433–1461.
885. Handa, P., Kothari, S. and Wasley, C. (1993) Sensitivity of multivariate tests of the capital
asset pricing model to the returns measurement interval. Journal of Finance, 48:1543–1551.
886. Handa, P. and Linn, S. (1993) Arbitrage pricing with estimation risk. Journal of Financial
887. Handa, P. and Tiwari, A. (2006) Does stock return predictability imply improved asset
allocation and performance? Evidence from the U.S. stock market (1954–2002). The Journal
of Business, 79:2423–2468.
888. Hankansson, N., Kunkel, G. and Ohlson, J. (1982) Sufficient and necessary conditions for
information to have social value in pure exchange. Journal of Finance, 37:1169–1181.
889. Hansen, L., Heaton, J. and Li, N. (2008) Consumption strikes back? Measuring long run
risk. Journal of Political Economy, 116:260–302.
890. Hansen, L., Heaton, J. and Lutmer, E. (1995) Econometric evaluation of asset pricing
models. Review of Financial Studies, 8:237–274.
891. Hansen, L. and Jagannathan, R. (1991) Implications of security market data for models of
dynamic economies. Journal of Political Economy, 91:249–265.
892. Hansen, L. and Jagannathan, R. (1997) Assessing specification errors in stochastic discount
factors models. Journal of Finance, 52:557–590.
893. Hansen, L. and Sargent, T. (2001) Robust control and model uncertainty. The American
894. Hansen, L., Sargent, T. and Tallarini, T. (1999) Robust permanent income and pricing.
895. Hansen, L. and Singleton, K. (1982) Generalized instrumental variables estimation of
nonlinear rational expectations models. Econometrica, 50:1269–1286.
896. Hansen, L. and Singleton, K. (1983) Stochastic consumption, risk aversion, and the temporal
behavior of asset returns. Journal of Political Economy, 91:249–265.
References 799
897. Harris, C. and Laibson, D. (2001) Dynamic choices of hyperbolic consumers. Econometrica,
69:935–957.
898. Harris, L. (1986) A transaction data study of weekly and intradaily patterns in stock returns.
899. Harris, L. (1994) Minimum price variations, discrete bid-ask spreads, and quotation sizes.
900. Harris, L. and Hasbrouck, J. (1996) Market vs. limit orders: The SuperDOT evidence on
order submission strategy. Journal of Financial and Quantitative Analysis, 31:213–231.
901. Harris, M. and Raviv, A. (1993) Differences of opinion make a horse race. Review of
902. Harris, M. and Venkatesh, P. (2005) The information content of the limit order book:
evidence from NYSE specialist trading decisions. Journal of Financial Markets, 8:25–67.
903. Harrison, M. and Kreps, D. (1978) Speculative investor behavior in a stock market with
heterogeneous expectations. Quarterly Journal of Economics, 92:323–336.
904. Harrison, M. and Kreps, D. (1979) Martingales and arbitrage in multiperiod securities
markets. Journal of Economic Theory, 28:183–91.
905. Harrison, M. and Pliska, S. (1981) Martingales and stochastic integrals in the theory of
continuous trading. Stochastic Processes and their Applications, 11:215–260.
906. Hart, O. (1975) On the optimality of equilibrium when the market structure is incomplete.
907. Hart, O. and Kreps, D. (1986) Price destabilizing speculation. Journal of Political Economy,
94:927–952.
908. Harvey, C. (1989) Time varying conditional covariances in tests of asset pricing models.
909. Harvey, C. (1991) The world price of covariance risk. Journal of Finance, 46:111–157.
910. Harvey, C. and Siddique, A. (2000) Conditional skewness in asset pricing tests. Journal of
Finance, 55:1263–1296.
911. Hasbrouck, J. (1988) Trades, quotes, inventories, and information. Journal of Financial
912. Hasbrouck, J. (1991) Measuring the information content of stock trades. Journal of Finance,
46:179–207.
913. Hasbrouck, J. and Sofianos, G. (1993) The trades of market makers: an empirical analysis
of NYSE specialists. Journal of Finance, 48:1565–1595.
914. Hasbrouck, J. and Schwartz, R. (1988) Liquidity and execution costs in equity markets.
Journal of Portfolio Management, 14:10–16.
915. Haugen, R. and Baker, N. (1996) Commonality in the determinants of expected stock returns.
916. Hawawini, G. and Keim, D. (1995) On the predictability of common stock returns: world-
wide evidence. In Handbooks in Operations Research and Management Science, Vol. 9
(Finance), R.A. Jarrow, V. Maksimovic and W.T. Ziemba (eds.), North Holland, Amsterdam.
917. Hayashi, F. (1985) The effect of liquidity constraints on consumption: a cross-sectional
analysis. Quarterly Journal of Economics, 100:183–206.
918. He, H. and Modest, D. (1995) Market frictions and consumption-based asset pricing. Journal
of Political Economy, 103:94–117.
919. He, H. and Pearson, N.D. (1991) Consumtpion and portfolio choices with incomplete
markets and short-sale constraints: the infinite dimensional case. Journal of Economic
Theory, 54:259–304.
920. He, H. and Wang, J. (1995) Differential information and dynamic behavior of stock trading
volume. Review of Financial Studies, 8:919–972.
921. He, J., Kan, R., Ng, L. and Zhang, C. (1996) Tests of the relations among marketwide factors,
firm-specific variables, and stock returns using a conditional asset pricing model. Journal of
Finance, 51:1891–1908.
800 References
922. Healy, P. and Palepu, K. (2001) Information asymmetry, corporate disclosure, and the
capital markets: a review of the empirical disclosure literature. Journal of Accounting and
923. Heaton, J. (1993) The interaction between time-nonseparable preferences and time aggrega-
tion. Econometrica, 61:353–385.
924. Heaton, J. (1995) An empirical investigation of asset pricing with temporally dependent
preference specifications. Econometrica, 63:681–717.
925. Heaton, J. and Lucas, D. (1992) The effects of incomplete insurance markets and trading
costs in a consumption-based asset pricing model. Journal of Economic Dynamics and
Control, 16:601–620.
926. Heaton, J. and Lucas, D. (1995) The importance of investor heterogeneity and financial
market imperfections for the behavior of asset prices. Carnegie-Rochester Conference series
on Public policy, 42:1–32.
927. Heaton, J. and Lucas, D. (1996) Evaluating the effects of incomplete markets on risk sharing
and asset pricing. Journal of Political Economy, 104:443–487.
928. Heaton, J. and Lucas, D. (1997) Market frictions, saving behavior and portfolio choice.
Macroeconomic Dynamics, 1:76–101.
929. Heaton, J. and Lucas, D. (2000) Portfolio choice and asset prices: the importance of
entrepreneurial risk. Journal of Finance, 55:1163–1197.
930. Hellwig, C. and Lorenzoni, G. (2009) Bubbles and self-enforcing debt. Econometrica, 77:
1137–1164.
931. Hellwig, M. (1980) On the aggregation of information in competitive markets. Journal of
932. Hellwig, M. (1982) Rational expectations equilibrium with conditioning on past prices: a
mean variance example. Journal of Economic Theory, 26:279–312.
933. Hendershott, T. and Mendelson, H. (2000) Crossing networks and dealer markets: competi-
tion and performance. Journal of Finance, 55:2071–2115.
934. Hendershott, T. and Seasholes, M. (2007) Market maker inventories and stock prices. The
935. Henderson V. (2005) Explicit solutions to an optimal portfolio choice problem with
stochastic income. Journal of Economic Dynamics and Control, 29:1237–1266.
936. Henderson, V. and Hobson, D. (2008) Utility indifference pricing: an overview. In Indiffer-
ence Pricing, R. Carmona (ed.), Princeton University Press, 44–74.
937. Hens, T. and Pilgrim, B. (2003) General Equilibrium Foundations of Finance: Structure of
Incomplete Markets Models, Kluwer Academic Publishers.
938. Hens, T. and Rieger M.O. (2010) Financial Economics, Springer, Berlin/Heidelberg.
939. Hernandez, A., and Santos, M. (1996) Competitive equilibria for infinite-horizon economies
with incomplete markets. Journal of Economic Theory, 71:102–130.
940. Heston, S., Loewenstein, M. and Willard G. (2007) Options and Bubbles. Review of
941. Heston, S. and Sadka, R. (2008) Seasonality in the cross-section of stock returns. Journal of
942. Hindy, A. (1995) Viable prices in financial markets with solvency constraints. Journal of
Mathematical Economics, 24:105–135.
943. Hindy, A. and Huang, C. (1993) Optimal consumption and portfolio rules with durability
and local substitution. Econometrica, 61:85–121.
944. Hindy, A., Huang, C. and Zhu, S. (1997) Optimal consumption and portfolio rules with
durability and habit formation. Journal of Economic Dynamics and Control, 21:525–550.
945. Hirshleifer, D. (2001) Investor psychology and asset pricing. Journal of Finance, 56:1533–
1597.
946. Hirshleifer, D. and Luo, G. (2001) On the survival of overconfident traders in a competitive
securities market. Journal of Financial Markets, 4:73–84.
References 801
947. Hirshleifer, D., Subrahmanyam, A. and Titman, S. (1994) Security analysis and trading
patterns when some investors receive information before others. Journal of Finance,
49:1664–1698.
948. Hirshleifer, J. (1971) The private and social value of information and the reward to inventive
activity. The American Economic Review, 61:561–574.
949. Ho, T. and Stoll, H. (1981) Optimal dealer pricing under transactions and return uncertainty.
950. Ho, T. and Stoll, H. (1983) The dynamics of dealer markets under competition. Journal of
Finance, 38:1053–1074.
951. Hodrick, R. (1992) Dividend yields and expected stock returns: alternative procedures for
inference and measurement. Review of Financial Studies, 5:357–386.
952. Hodrick, R. and Zhang, X. (2001) Evaluating the specification errors of asset pricing models.
953. Holden, C. and Subrahmanyan, A. (1992) Long-lived private information and imperfect
competition. Journal of Finance, 47:247–270.
954. Holden, C. and Subrahmanyan, A. (1994) Risk aversion, imperfect competition and long-
lived information. Economics Letters, 44:181–190.
955. Hollifield, B., Miller, R. and Sandas, P. (2004) Empirical analysis of limit order markets.
956. Holmstrom, B. and Myerson, R.B. (1983) Efficient and durable decision rules with
incomplete information. Econometrica, 51:1799–1819.
957. Hong, H., Kubik, J. and Stein, J. (2004) Social interaction and stock-market participation.
958. Hong, H., Lim, T. and Stein, J. (2000) Bad news travels slowly: size, analyst coverage and
the profitability of momentum strategies. Journal of Finance, 55:-265–295.
959. Hong, H. and Rady, S. (2002) Strategic trading and learning about liquidity. Journal of
Financial Markets, 5:419–450.
960. Hong, H., Scheinkman, J. and Xiong, W. (2006) Asset float and speculative bubbles. Journal
of Finance, 61:1073–1117.
961. Hong, H. and Stein, J. (1999) A unified theory of underreaction, momentum trading and
overreaction in asset markets. Journal of Finance, 44:2143–2184.
962. Hong, H. and Stein, J. (2003) Differences of opinion, rational arbitrage and market crashes.
963. Hong, H. and Stein, J. (2007) Disagreement and the stock market. Journal of Economic
Perspectives, 21:109–128.
964. Hong, H. and Stein, J. (2007) Simple forecasts and paradigm shifts. Journal of Finance,
62:1207–1242.
965. Hong, H. and Wang, J. (2000) Tradinng and returns under periodic market closures. Journal
of Finance, 55:297–320.
966. Horneff, W., Maurer, R. and Stamos, M. (2008) Life-cycle asset allocation with annuity
markets. Journal of Economic Dynamics and Control 32:3590–3612.
967. Hou, K. (2007) Industry information diffusion and the lead-lag effect in stock returns.
968. Hou, K. and Moskowitz, T. (2005) Market frictions, price delay, and the cross-section of
expected returns. Review of Financial Studies, 18:981–1020.
969. Hou, K. and Robinson, D. (2006) Industry concentration and average stock returns. The
Journal fo Finance, 61:1927–1955.
970. Hsieh, D. and Miller, M. (1990) Margin regulations and stock market volatility. Journal of
Finance, 45:3–30.
971. Huang, C. and Litzenberger, R. (1988) Foundations for Financial Economics. North-
Holland, Amsterdam.
972. Huang, K. and Werner, J. (2000) Asset price bubbles in Arrow-Debreu and sequential
equilibrium. Economic Theory, 15:253–278.
802 References
973. Huang, K. and Werner, J. (2004) Implementing Arrow-Debreu equilibria by trading

infinitely-lived securities. Economic Theory, 19:35–64.
974. Huang, K., Liu, Z. and Zhu, J. (2015) Temptation and self-control: some evidence and
applications. Journal of Money Credit and Banking, 47:581–615.
975. Huang, M. (2003) Liquidity shocks and equilibrium liquidity premia. Journal of Economic
Theory, 109:104–129.
976. Huang, R. and Stoll, H. (1994) Market microstructure and stock return predictions. Review
977. Huang, R. and Stoll, H. (1996) Dealer versus auction markets: a paired comparison of
execution costs on NASDAQ and the NYSE. Journal of Financial Economics, 41:313–357.
978. Huang, R. and Stoll, H. (1997) The components of the bid-ask spread: a general approach.
979. Huberman, G. (1982) A simple approach to arbitrage pricing theory. Journal of Economic
Theory, 28:183–91.
980. Huberman, G. (1992) A simple approach to arbitrage pricing theory. In The New Palgrave
Dictionary of Money and Finance, J. Eatwell, M. Milgate and P. Newman (eds.), Palgrave
Macmillan, London.
981. Huberman, G. (2001) Familiarity breeds investment. Review of Financial Studies, 14:659–
680.
982. Huberman, G. and Kandel, S. (1987) Mean-variance spanning. Journal of Finance, 52:873–
888.
983. Huberman, G. and Kandel, S. (1990) Market efficiency and Value Line’s record. Journal of
Business, 63:187–215.
984. Huberman, G., Kandel, S. and Stambaugh. R. (1987) Mimicking portfolios and exact
arbitrage pricing. Journal of Finance, 52:1–9.
985. Huberman, G. and Ross, S. (1983) Portfolio turnpike theorems, risk aversion, and regularly
varying utility functions. Econometrica, 51:1345–1361.
986. Huberman, G. and Schwert, W. (1985) Information aggregation, inflation, and the pricing of
indexed bonds. Journal of Political Economy, 93:92–114.
987. Huddart, S., Hughes, J. and Brunnermeier, M. (1999) Disclosure requirements and stock
exchange listing choice in an intertemporal context. Journal of Accounting Economics,
26:237–269.
988. Huddart, S., Hughes, J. and Levine, C. (1999) Public disclosure of trades by corporate
insiders in financial markets and tacit coordination. Mimeo.
989. Huddart, S., Hughes, J. and Levine, C. (2001) Public disclosure and dissimulation of insider
trades. Econometrica, 69:665–682.
990. Huddart, S. and Ke, B. (2007) Information asymmetry and cross-sectional determinants of
insider trading. Contemporary Accounting Research, 24:195–232.
991. Huddart, S., Ke, B. and Shi, C. (2007) Jeopardy, non-public information, and insider trading
around SEC 10-K and 10-Q filings. Journal of Accounting and Economics, 43:3–36.
992. Huggett, M. (1993) The risk-free rate in heterogeneous-agent incomplete-insurance
economies. Journal of Economic Dynamics and Control, 17:953–969.
993. Huggett, M. (2001) Aggregate precautionary savings: when is the third derivative irrelevant?
994. Hussman, J. (1992) Market efficiency and inefficiency in rational expectations equilibria.
995. Hvidkjaer, S. (2006) A trade-based analysis of momentum. Review of Financial Studies,
19:458–491.
996. Ibbotson, R. and Ritter, J. (1995) Initial public offerings. In Handbooks in Operations
Research and Management Science, Vol. 9 (Finance), R.A. Jarrow, V. Maksimovic and W.T.
Ziemba (eds.), North Holland, Amsterdam.
997. Ikeda, S. and Shibata, A. (1992) Fundamentals-dependent bubbles in stock prices. Journal
of Monetary Economics, 30:143–168.
References 803
998. Ikenberry, D., Lakonishok, J. and Vermaelen, T. (1995) Market underreaction to open market
share repurchases. Journal of Financial Economics, 39:181–208.
999. Ingersoll, J. (1984) Some results in the theory of arbitrage pricing. Journal of Finance,
39:1021–1039.
1000. Ingersoll, J. (1987) Theory of Financial Decision Making. Rowman & Littlefield, New
Jersey.
1001. Ingrao, B. and Israel, G. (1987) La Mano Invisibile. Laterza, Roma.
1002. Jackson, M. (1991) Equilibrium, price formation, and the value of private information.
1003. Jackson, M. (1994) A proof of the existence of speculative equilibria. Journal of Economic
Theory, 64:221–233.
1004. Jackson, M. and Peck, J. (1991) Speculation and price fluctuations with private, extrinsic
signals. Journal of Economic Theory, 55:274–295.
1005. Jacobs, F. (1999) Incomplete markets and security prices: do asset-pricing puzzles result
from aggregation problems? Journal of Finance, 46:123–163.
1006. Jacobs, K. and Wang, K. (2004) Idiosyncratic consumption risk and the cross section of asset
1007. Jaffe, J. (1974) Special information and insider trading. Journal of Business, 47:410–428.
1008. Jaffe, J. and Westerfield, R. (1999) The week-end effect in common stock returns: the
international evidence. Journal of Finance, 40:433–454.
1009. Jagannathan, R. and Kocherlakota, N. (1996) Why should older people invest less in stocks
than younger people? Federal Reserve Bank of Minneapolis Quarterly Review, 20:11–23.
1010. Jagannathan, R., Kubita, K. and Takehara, H. (1998) Relationship between labor-income
risk and average returns: empirical evidence from the Japanese stock market. Journal of
Business, 71:319–347.
1011. Jagannathan, R., McGrattan, E. and Scherbina, A. (2001) The declining U.S. equity
premium. NBER working paper, no.8172.
1012. Jagannathan, R. and Wang, Z. (1996) The conditional CAPM and the cross-section of
expected returns. Journal of Finance, 51:3–55.
1013. Jagannathan, R. and Wang, Z. (2007) Lazy investors, discretionary consumption, and the
cross-section of stock returns. Journal of Finance, 62:1623–1661.
1014. Jang, B., Keun Koo, H., Liu, H. and Loewenstein, M. (2007) Liquidity premia and
transaction costs. Journal of Finance, 62:2329–2366.
1015. Jappelli, T. (1990) Who is credit constrained in the US economy? Quarterly Journal of
Economics, 105:219–234.
1016. Jappelli, T. and Pagano, M. (1994) Saving, growth, and liquidity constraints. Quarterly
1017. Jappelli, T., Pischke, J. and Souleles, N. (1998) Testing for liquidity constraints in Euler
equations with complementary data sources. The Review of Economics and Statistics,
80:251–262.
1018. Jarrow, R. (1986) The relationship between arbitrage and first order stochastic dominance.
1019. Jarrow, R. (1988) Preferences, continuity, and the arbitrage pricing theory. Review of
1020. Jarrow, R. (1992) Market manipulation, bubbles, corners, and short squeezes. Journal of
1021. Jarrow, R. (1994) Derivative security markets, market manipulation, and option pricing
theory. Journal of Financial and Quantitative Analysis, 29:241–261.
1022. Jarrow, R. and Zhao, F. (2006) Downside loss aversion and portfolio management. Manage-
ment Science, 52:558–566.
1023. Jegadeesh, N. (1990) Evidence of predictable behavior of security returns. Journal of
Finance, 45:881–898.
1024. Jegadeesh, N. (1991) Seasonality in stock price mean reversion: evidence from the US and
UK. Journal of Finance, 46:1427–1444.
804 References
1025. Jegadeesh, N. (1992) Does market risk really explain the size effect. Journal of Financial
1026. Jegadeesh, N. and Titman, S. (1993) Returns to buying winners and selling losers:
implications for stock market efficiency. Journal of Finance, 48:65–91.
1027. Jegadeesh, N. and Titman, S. (1995) Overreaction, delayed reaction, and contrarian profits.
1028. Jegadeesh, N. and Titman S. (2001) Profitability of momentum strategies: an evaluation of
alternative explanations. Journal of Finance, 56:699–720.
1029. Jegadeesh, N. and Titman S. (2002) Cross-sectional and time-series determinants of
momentum returns. Review of Financial Studies, 15:143–158.
1030. Jeng, L., Metrick, A. and Zeckauser, R. (2003) The profits to insider trading: a performance-
evaluation perspective. Review of Economics and Statistics, 85:453–471.
1031. Jensen, M. (1978) Some anomalous evidence regarding market efficiency. Journal of
1032. Jermann, U. (1998) Asset pricing in production economies. Journal of Monetary Economics,
41:257–275.
1033. Jobson, J. and Korkie, B. (1982) Potential performance and tests of portfolio efficiency.
1034. John, K. and Narayanan, R. (1997) Market manipulation and the role of insider trading
regulations. Journal of Business, 70:217–247.
1035. John, K. and Lang, L. (1991) Insider trading around dividend announcements: theory and
evidence. Journal of Finance, 46:1361–1389.
1036. Johnson, T. (2004) Forecast dispersion and the cross section of expected returns. Journal of
Finance, 59:1957–1978.
1037. Johnson, T. (2008) Volume, liquidity, and liquidity risk. Journal of Financial Economics,
87:388–417.
1038. Jones, C. (2001) Extracting factors from heteroskedastic asset returns. Journal of Financial
1039. Jones, C. (2001) A century of stock market liquidity and trading costs. Mimeo.
1040. Jones, C., Kaul, G. and Lipson, M. (1994) Information, trading, and volatility. Journal of
1041. Jones, C., Kaul, G. and Lipson, M. (1994) Transactions, volume and volatility. Review of
1042. Jones, C. and Lamont, O. (2002) Short sale constraints and stock returns. Journal of
1043. Jones, C. and Lipson, M. (2001) Sixteenths: direct evidence on institutional execution costs.
1044. Jones, C. and Seguin, P. (1997) Transaction costs and price volatility: evidence from
commission deregulation. The American Economic Review, 87:728–737.
1045. Jordan, J. (1982) The generic existence of rational expectations equilibrium in the higher
dimensional case. Journal of Economic Theory, 26:2224–2243.
1046. Jordan, J. (1983) On the efficient markets hypothesis. Econometrica, 51:1325–1343.
1047. Jordan, J. and Radner, R. (1982) Rational expectations in microeconomic models: an
overview. Journal of Economic Theory, 26:201–223.
1048. Jorion, P. and Giovannini, A. (1993) Time-series tests of a non-expected-utility model of
asset pricing. European Economic Review, 37:1083–1100.
1049. Jouini, E. and Kallal, H. (1995) Martingales and arbitrage in securities markets with
transaction costs. Journal of Economic Theory, 66:178–197.
1050. Jouini, E. and Kallal, H. (1995) Arbitrage in securities markets with short sales constraints.
Mathematical Finance, 5:197–232.
1051. Jouini, E. and Napp, C. (2006) Heterogeneous beliefs and asset pricing in discrete time.
Journal of Economic Dynamics and Control 30:1233–1266.
1052. Jouini, E. and Napp, C. (2007) Consensus consumer and intertemporal asset pricing with
heterogeneous beliefs. Review of Economic Studies, 74:1149–1174.
References 805
1053. Ju, N. and Miao, J. (2012) Ambiguity, learning, and asset returns. Econometrica, 80:559–
591.
1054. Julliard, C. and Ghosh, A. (2012) Can rare events explain the equity premium puzzle?
1055. Jurek, J. and Viceira, L. (2011) Optimal value and growth tilts in long-horizon portfolios
1056. Kabanov, Y. and Safarian, M. (2009) Markets with Transaction Costs - Mathematical
Theory. Springer, Berlin/Heidelberg.
1057. Kahle, K. (2000) Insider trading and the long run performance of new security issues.
Journal of Corporate Finance, 6:25–53.
1058. Kahn, C. and Winton, A. (1998) Ownership structure, speculation, and shareholder interven-
tion. Journal of Finance, 53:99–129.
1059. Kahn, J. (1990) Moral hazard, imperfect risk-sharing, and the behavior of asset returns.
1060. Kahneman, D. and Tversky, A. (1979) Prospect theory: an analysis of decision under risk.
1061. Kahneman, D. and Tversky, A. (1982) Intuitive prediction: biases and corrective procedures.
In Judgement under Uncertainty: Heuristics and Biases, D. Kahneman, P. Slovic and A.
Tversky (eds.), Cambridge University Press.
1062. Kamihigashi, T. (1998) Uniqueness of asset prices in an exchange economy with unbounded
utility. Economic Theory, 12:103–122.
1063. Kan, R. and Zhang, C. (1997) Two pass tests of asset pricing models with useless factors.
1064. Kandel, E. and Marx, L. (1997) Nasdaq market structure and spread patterns. Journal of
1065. Kandel, E. and Marx, L. (1999) Payments for order flow on Nasdaq. Journal of Finance,
54:35–66.
1066. Kandel, S. and Pearson, N. (1995) Differential interpretation of public signals and trade in
speculative markets. Journal of Political Economy, 103:831–872.
1067. Kandel, S. and Stambaugh, R. (1987) On the correlations and inferences about mean
variance efficiency. Journal of Financial Economics, 18:61–90.
1068. Kandel, S. and Stambaugh, R. (1990) Expectations and volatility of consumption and asset
returns. Review of Financial Studies, 3:207–232.
1069. Kandel, S. and Stambaugh, R. (1991) Asset returns and intertemporal preferences. Journal
1070. Kandel, S. and Stambaugh, R. (1995) Portfolio inefficiency and the cross-section of expected
1071. Kandel, S. and Stambaugh, R. (1996) On the predictability of stock returns: an asset
allocation perspective. Journal of Finance, 51:385–424.
1072. Kang, J. and Stulz, R. (1997) Why is there a home bias? An analysis of foreign portfolio
equity ownership in Japan. Journal of Financial Economics, 46:3–28.
1073. Karatzas, I. and Shreve, S. (1998) Methods of Mathematical Finance. Springer, New York.
1074. Karolyi, A. (2006) The world of cross-listings and cross-listings of the world: challenging
conventional wisdom. Review of Finance, 10:99–152.
1075. Karpoff, J. (1987) The relation between price changes and trading volume: a survey. Journal
of Financial and Quantitative Analysis, 22:109–126.
1076. Kaul, G. (1996) Predictable components in stock returns. In Handbook of Statistics (Vol.14),
G.S. Maddala and C.R. Rao (eds.), Elsevier Science Publisher.
1077. Ke, B., Huddart, S. and Petroni, K. (2003) What insiders know about future earnings and how
they use it: evidence from insider traders. Journal of Accounting and Economics, 35:315–
346.
1078. Kehoe, T. and Levine, D. (1993) Debt constrained asset markets. Review of Economic
Studies, 60:865–888.
806 References
1079. Keim, D. (1983) Size-related anomalies and stock return seasonality: further empirical
evidence. Journal of Financial Economics, 12:13–32.
1080. Keim, D. (1989) Trading patterns, bid-ask spreads, and estimated security returns: the case
of common stock at calendar turning points. Journal of Financial Economics 25:75–97.
1081. Keim, D. and Madhavan, A. (1996) The upstairs market for large-block transactions: analysis
and measurement of price effects. Review of Financial Studies, 9:1–36.
1082. Keim, D. and Stambaugh, R. (1984) A further invetsigation of the weekend effect in stock
1083. Keim, D. and Stambaugh, R. (1986) Predicting returns in the stock and bond markets.
1084. Kelly, B. and Ljungqvist, A. (2012) Testing asymmetric-information asset pricing models.
1085. Kelly, M. (1997) Do noise traders influence stock prices? Journal of Money Credit and
Banking, 29:251–363.
1086. Kelsey, D. and Milne, F. (1995) The arbitrage pricing theorem with non-expected utility
preferences. Journal of Economic Theory, 65:557–574.
1087. Keynes, J.M. (1936) The General Theory of Unemployment, Interest and Money. Harcourt,
Brace & World.
1088. Khanna, N., Slezak, S. and Bradley, M. (1994) Insider trading, outside search, and resource
allocation: why firms and society may disagree on insider trading restrictions. Review of
1089. Kihlstrom, R. and Mirman, L. (1974) Risk aversion with many commodities. Journal of
1090. Kihlstrom, R. and Mirman, L. (1981) Constant, increasing, decreasing risk aversion with
many commodities. Review of Economic Studies, 48:271–280.
1091. Kim, D. (1995) The errors in the variables problem in the cross-section of expected stock
1092. Kim, D. (1997) A Reexamination of firm size, book-to-market, and earnings price in the
cross-section of expected returns. Journal of Financial and Quantitative Analysis, 32:463–
491.
1093. Kim, J., Nelson, C. and Startz, R. (1991) Mean reversion in stock prices? A reappraisal of
the empirical evidence. Review of Economic Studies, 58:515–528.
1094. Kim, K. (2001) Price limits and stock market volatility. Economics Letters, 71:131–136.
1095. Kim, O. and Verrecchia, R. (1991) Market reaction to anticipated announcements. Journal
1096. Kim, O. and Verrecchia, R. (1991) Trading volume and price reactions to public announce-
ments. Journal of Accounting Research, 29:302–321.
1097. Kim, O. and Verrecchia, R. (1994) Market liquidity and volume around earnings announce-
ments. Journal of Accounting and Economics, 17:41–67.
1098. Kim, S., Lin, J. and Slovin, M. (1997) Market structure, informed trading, and analysts’
recommendations. Journal of Financial and Quantitative Analysis, 32:507–532.
1099. Kim, T. and Omberg, E. (1996) Dynamic nonmyopic portfolio behavior. Review of Financial
Studies, 9:141–161.
1100. Kimball, M. (1990) Precautionary saving in the small and in the large. Econometrica, 58:53–
73.
1101. Kimball, M. (1993) Standard risk aversion. Econometrica, 61:589–611.
1102. Kirby, C. (1997) Measuring the predictable variation in stock and bond returns. Review of
1103. Kirby, C. (1998) The restrictions on predictability implied by rational asset pricing models.
1104. Kleidon, A. (1986) Variance bounds tests and stock price valuation models. Journal of
1105. Klibanoff, P., Marinacci, M. and Mukerji, S. (2005) A smooth model of decision making
under ambiguity. Econometrica, 73:1849–1892.
References 807
1106. Knez, P. and Ready, M. (1996) Estimating the profits from trading strategies. Review of
1107. Knez, P. and Ready, M. (1997) On the robustness of size and book-to-market in cross-
sectional regressions. Journal of Finance, 52:1355–1382.
1108. Knight, F. (1921) Risk, Uncertainty, and Profit. Houghton Mifflin, Boston.
1109. Kocherlakota, N. (1990) On tests of representative consumer asset pricing models. Journal
1110. Kocherlakota, N. (1990) Disentangling the coefficient of relative risk aversion from the
elasticity of intertemporal substitution: an irrelevance result. Journal of Finance, 45:175–
190.
1111. Kocherlakota, N. (1992) Bubbles and constraints on debt accumulation. Journal of Economic
Theory, 57:245–256.
1112. Kocherlakota, N. (1996) The equity premium: it’s still a puzzle. Journal of Economic
Literature, 34:42–71.
1113. Kocherlakota, N. (1998) The effects of moral hazard on asset prices when financial markets
are complete. Journal of Monetary Economics, 41:39–56.
1114. Kocherlakota, N. (2008) Injecting rational bubbles. Journal of Economic Theory, 142:218–
232.
1115. Kogan, L., Makarov, I. and Uppal, R. (2007) The equity risk premium and the riskfree rate
in an economy with borrowing constraints. Mathematics and Financial Economics, 1:1–19.
1116. Kogan, L., Ross, S., Wang, J. and Westerfield, M. (2006) The price impact and survival of
irrational traders. Journal of Finance 61:195–229.
1117. Kogan, L. and Wang, T. (2003) A simple theory of aset pricing under model uncertainty.
MIT working paper.
1118. Kondor, P. (2012) The more we know about the fundamental, the less we agree on the price.
1119. Koo, H. (1999) Consumption and portfolio selection with labor income: a discrete time
approach. Mathematical Methods of Operations Research, 50:219–243.
1120. Korajczyk, R. A. and Sadka, R. (2004) Are momentum profits robust to trading costs?
1121. Koszegi, B. and Rabin, M. (2006) A model of reference-dependent preferences. Quarterly
1122. Koski, J. and Michaely, R. (2000) Prices, liquidity, and the information content of trades.
1123. Kothari, S. (2001) Capital markets research in accounting. Journal of Accounting and
1124. Kothari, S. and Shanken, J. (1992) Stock return variation and expected dividends. Journal of
1125. Kothari, S. and Shanken, J. (1997) Book-to-market, dividend yield, and expected market
returns: a time series analysis. Journal of Financial Economics, 44:169–203.
1126. Kothari, S., Shanken, J. and Sloan, R. (1995) Another look at the cross-section of expected
stock returns. Journal of Finance, 50:185–224.
1127. Kothari, S. and Warner, J. (1997) Measuring long-horizon security price performance.
1128. Kramkov, D. and Schachermayer, W. (1999) The asymptotic elasticity of utility functions
and optimal investment in incomplete markets. Annals of Applied Probability, 9:904–950.
1129. Kramkov, D. and Schachermayer, W. (2003) Necessary and sufficient conditions in the
problem of optimal investment in incomplete markets. Annals of Applied Probability,
13:1504–1516.
1130. Krasa, S. (1987) Existence of competitive equilibrium for option markets. Journal of
1131. Krasa, S. and Werner, J. (1991) Equilibrium with options: existence and indeterminacy.
808 References
1132. Krebs, T. (2004) Testable implications of consumption-based asset pricing models with
incomplete markets. Journal of Mathematical Economics, 40:191–206.
1133. Krebs, T. (2007) Rational expectations equilibrium and the strategic choice of costly
information. Journal of Mathematical Economics, 43:532–548.
1134. Kreps, D. (1977) A note on fulfilled expectations equilibria. Journal of Economic Theory,
14:32–44.
1135. Kreps, D. (1982) Multiperiod securities and the efficient allocation of risk: a comment on the
Black and Scholes option pricing model. In The Economics of Uncertainty and Information,
J. McCall (ed.), University of Chicago Press.
1136. Kreps, D. (1988) Notes on the Theory of Choice. Westview Press, Boulder.
1137. Kroll, Y., Levy, H. and Rapoport, A. (1988) Experimental tests of the separation theorem
and the capital asset pricing model. The American Economic Review, 78:500–519.
1138. Krueger, D. and Lustig, H. (2010) When is market incompleteness irrelevant for the price of
aggregate risk (and when is it not)? Journal of Economic Theory, 145:1–41.
1139. Krusell, P., Kuruscu, B. and Smith, A. (2002) Time orientation and asset prices. Journal of
1140. Krusell, P. and Smith, A. (1997) Income and wealth heterogeneity, portfolio choice, and
equilibrium asset returns. Macroeconomic Dynamics, 1:387–422.
1141. Kuhn, (1970) The Structure of Scientific Revolutions. University of Chicago Press.
1142. Kumar, P. and Seppi, D. (1992) Futures manipulation with “cash settlement”. Journal of
Finance, 47:1485–1502.
1143. Kupiec, P. (1997) Margin requirements, volatility, and market integrity: what have we
learned since the crash? Mimeo.
1144. Kupiec, P. and Sharpe, S. (1991) Animal spirits, margin requirements, and stock price
1145. Kurz, M. and Motolese, M. (2001) Endogenous uncertainty and market volatility. Economic
Theory, 17:497–544.
1146. Kyle, A. (1984) A theory of futures market manipulation. In The Industrial Organization of
Futures Markets, R. Anderson (ed.), Lexington Books, 141–175.
1147. Kyle, A. (1985) Continuous auctions and insider trading. Econometrica, 53:1315–1334.
1148. Kyle, A. (1989) Informed speculation with imperfect competition. Review of Economic
Studies, 56:317–356.
1149. Kyle, A. and Viswanathan, S. (2008) How to define illegal price manipulation. The American
1150. Kyle, A. and Wang, A. (1997) Speculation duopoly with agreement to disagree: can
overconfidence survive the market test? Journal of Finance, 52:2073–2090.
1151. Kyle, A. and Xiong, W. (2001) Contagion as a wealth effect. Journal of Finance, 56:1401–
1440.
1152. Labadie, P. (1989) Stochastic inflation and the equity premium. Journal of Monetary
1153. Laffont, J. (1985) On the welfare analysis of rational expectations equilibria with asymmetric
information. Econometrica, 53:1–30.
1154. Laffont, J. (1989) The Economics of Uncertainty and Information. MIT Press, Cambridge.
1155. Laffont, J. and Maskin, E. (1990) The efficient market hypothesis and insider trading on the
stock market. Journal of Political Economy, 98:70–89.
1156. Laibson, D. (1997) Golden eggs and hyperbolic discounting. The Quarterly Journal of
Economics, 112:443–477.
1157. Laibson, D., Repetto, A. and Tobacman, J. (1998) Self-control and saving for retirement.
Brookings Papers on Economic Activity, 1:91–196.
1158. Lakonishok, J. and Lee, D. (2001) Are insiders’ trades informative? Review of Financial
Studies, 14:79–111.
1159. Lakonishok, J., Shleifer, A. and Vishny, R. (1994) Contrarian investment, extrapolation and
risk. Journal of Finance, 49:1541–1578.
References 809
1160. Lakonishok, J. and Smidt, A. (1988) Are seasonal anomalies real? A 90 years perspective.
1161. Lambert, R., Leuz, C. and Verrecchia, R. (2007) Accounting information, disclosure, and
the cost of capital. Journal of Accounting Research, 45:385–420.
1162. Lamont, O. (1998) Earnings and expected returns. Journal of Finance, 53:1563–1587.
1163. Lamont, O., Polk, C. and Saa-Requejo, J. (2001) Financial constraints and stock returns.
1164. Lamoureux, C. and Schnitzlen, C. (1997) When it’s not the only game in town: the effect of
bilateral search on the quality of a dealer market. Journal of Finance, 52:683–712.
1165. Lamoureux, C. and Zhou, G. (1996) Temporary components of stock returns: what do the
data tell us? Review of Financial Studies, 9:1033–1059.
1166. Lang, L., Litzenberger, R. and Madrigal, V. (1992) Testing financial market equilibrium
under asymmetric information. Journal of Political Economy, 100:317–348.
1167. La Porta, R. (1996) Expectations and the cross-section of stock returns. Journal of Finance,
51:1715–1742.
1168. La Porta, R., Lakonishok, J., Shleifer, A. and Vishny, R. (1997) Good news for value stocks:
further evidence on market efficiency. Journal of Finance, 52:859–874.
1169. Lauterbach, B. and Ben-Zion, U. (1993) Stock market crashes and the performance of circuit
breakers: empirical evidence. Journal of Finance, 48:1909–1920.
1170. Leach, C. and Madhavan, A. (1992) Intertemporal price discovery by market makers: active
versus passive learning. Journal of Financial Intermediation, 2:207–235.
1171. Leach, J. (1991) Rational speculation. Journal of Political Economy, 99:131–144.
1172. LeBaron, B., Arthur, B. and Palmer, R. (1999) Time series properties of an artificial stock
market. Journal of Economic Dynamics and Control, 23:1487–1516.
1173. Lee, C., Ready, M. and Seguin, P. (1994) Volume, volatility, and New York Stock Exchange
trading halts. Journal of Finance, 49:183–213.
1174. Lee, C., Shleifer, A. and Thaler, R. (1991) Investor sentiment and closed-end fund puzzle.
1175. Lee, C. and Swaminathan, B. (2000) Price momentum and trading volume. Journal of
Finance, 55:2017–2069.
1176. Lee, W., Jiang, C. and Indro, D. (2002) Stock market volatility, excess returns, and the role
of investor sentiment. Journal of Banking and Finance, 26:2277–2299.
1177. Lehman, B. (1990) Fads, martingales, and market efficiency. Quarterly Journal of Eco-
nomics, 105:1–28.
1178. Lehmann, B. and Modest, D. (1988) The empirical foundations of the arbitrage pricing
theory. Journal of Financial Economics, 21:213–254.
1179. Leippold, M., Trojani, F. and Vanini, P. (2008) Learning and asset prices under ambiguous
information. Review of Financial Studies, 21:2565–2597.
1180. Leland, H. (1972) On the existence of optimal policies under uncertainty. Journal of
1181. Leland, H. (1992) Insider trading: should it be prohibited? Journal of Political Economy,
100:859–887.
1182. Lengwiler, Y. (2004) Microfoundations of Financial Economics. Princeton University Press.
1183. LeRoy, S. (1973) Risk aversion and the martingale property of stock prices. International
1184. LeRoy, S. (1989) Efficient capital markets and martingales. Journal of Economic Literature,
27:1583–1621.
1185. LeRoy, S. (2004) Rational exuberance. Journal of Economic Literature, 42: 783–804.
1186. LeRoy, S. and Gilles, C. (1992) Asset price bubbles. In The New Palgrave Dictionary of
Money and Finance, J. Eatwell, M. Milgate and P. Newman (eds.), Palgrave Macmillan,
London.
1187. LeRoy, S. and LaCivita, C. (1981) Risk aversion and the dispersion of asset prices. Journal
810 References
1188. LeRoy, S. and Parke, W. (1992) Stock price volatility; tests based on the geometric random
walk. The American Economic Review, 82:981–992.
1189. LeRoy, S. and Porter, R. (1981) The present-value relation: tests based on implied variance
bounds. Econometrica 49:555–574.
1190. LeRoy, S. and Steigerwald, D. (1995) Volatility. In Handbooks in Operations Research and
Management Science, Vol. 9 (Finance), R.A. Jarrow, V. Maksimovic and W.T. Ziemba (eds.),
North Holland, Amsterdam.
1191. LeRoy, S. and Werner, J. (2001) Principles of Financial Economics, Cambridge University
Press.
1192. Letendre, M. and Smith, G. (2001) Precautionary saving and portfolio allocation: DP and
GMM. Journal of Monetary Economics, 48:197–215.
1193. Lettau, M. (2002) Idiosyncratic risk and volatility bounds, or, can models with idiosyncratic
risk solve the equity premium puzzle? Review of Economics and Statistics, 84:376–380.
1194. Lettau, M. and Ludvigson, S. (2001) Consumption, aggregate wealth, and expected stock
1195. Lettau, M. and Ludvigson, S. (2001) Resurrecting the (C)CAPM: a cross-sectional test when
risk premia are time varying. Journal of Political Economy, 109:1238–1287.
1196. Lettau, M. and Ludvigson, S. (2005) Expected returns and expected dividend growth.
1197. Lettau, M., Ludvigson, S. and Wachter J. (2008) The declining equity premium: what role
does macroeconomic risk play? Review of Financial Studies, 21:1653–1687.
1198. Lettau, M. and Uhlig, H. (2000) Can habit formation be reconciled with business cycle facts?
1199. Lettau, M. and Uhlig, H. (2000) Sharpe ratios and preferences an analytical approach.
Macroeconomic Dynamics, 6:242–265.
1200. Lettau, M. and Van Nieuwerburgh, S. (2008) Reconciling the return predictability evidence.
1201. Lettau, M. and Wachter, J. (2007) Why is long-horizon equity les risky? A duration based
explanation of the value premium. Journal of Finance, 62:55–92.
1202. Leuz, C. and Verrecchia, R. (2000) The economic consequences of increased disclosure.
Journal of Accounting Research, 38:91–124.
1203. Levine, D. and Zame, W. (2002) Does market incompleteness matter? Econometrica,
70:1805–1839.
1204. Levy, H. (1992) Stochastic dominance and expected utility: survey and analysis. Manage-
ment Science, 38:555–593.
1205. Levy, H. (2016) Stochastic Dominance: Investment Decision Making under Uncertainty, 3rd
edition. Springer, Cham.
1206. Levy, H. and Levy, M. (2002) Prospect theory: much ado about nothing? Management
Science, 48:1334–1349.
1207. Levy, H. and Levy, M. (2004) Prospect theory and mean-variance analysis. Review of
1208. Levy, H. and Wiener, Z. (1998) Stochastic dominance and prospect dominance with
subjective weighting functions. Journal of Risk and Uncertainty, 12:147–163.
1209. Lewellen, J. (1999) The time series relations among expected return, risk, and book-to-
market. Journal of Financial Economics, 54:5–43.
1210. Lewellen, J. (2002) Momentum and autocorrelation in stock returns. Review of Financial
Studies, 15:533–564.
1211. Lewellen, J. (2004) Predicting returns with financial ratios. Journal of Financial Economics,
74:209–235.
1212. Lewellen, J. and Nagel, S. (2006) The conditional CAPM does not explain asset-pricing
anomalies. Journal of Financial Economics, 82:289–314.
1213. Lewellen, J., Nagel, S. and Shanken, J. (2010) A skeptical appraisal of asset pricing tests.
Journal of Financial Economics 96:175–194.
References 811
1214. Lewellen, J. and Shanken, J. (2002) Learning, asset-pricing tests, and market efficiency.
1215. Lewis, K. (1996) What can explain the apparent lack of international consumption risk
sharing? Journal of Political Economy, 104:267–297.
1216. Lewis, K. (1999) Trying yo explain home bias in equities and consumption. Journal of
Economic Literature, 37:571–608.
1217. Li, Y. (2001) Expected returns and habit persistence. Review of Financial Studies, 14:861–
899.
1218. Liew, J. and Vassalou, M. (2000) Can book to market, size, and momentum be risk factors
that predict economic growth? Journal of Financial Economics, 57:221–245.
1219. Lin, J., Sanger, G. and Booth, G. (1995) Trade size and components of the bid-ask spread.
1220. Lintner, J. (1956) Distribution of incomes of corporations among dividends, retained
earnings, and taxes. The American Economic Review, 61:97–113.
1221. Lintner, J. (1965) Security prices, risk and maximal gains from diversification. Journal of
Finance, 20:587–615.
1222. Litzenberger, R. and Ramaswamy, K. (1979) On distributional restrictions for two fund
separation. TIMS Studies in the Management Science, 11:99–107.
1223. Liu, H. (2011) Dynamic portfolio choice under ambiguity and regime switching mean
returns. Journal of Economic Dynamics and Control, 35:623–640.
1224. Liu, H. and Loewenstein, M. (2002) Optimal portfolio selection with transaction costs and
finite horizons. Review of Financial Studies, 15:805–835.
1225. Liu, H. and Miao, J. (2015) Growth uncertainty, generalized disappointment aversion and
production-based asset pricing. Journal of Monetary Economics, 69:70–89.
1226. Liu, J. (2006) Portfolio selection in stochastic environments. Review of Financial Studies,
20:1–39.
1227. Liu, J., Pan, J. and Wang, T. (2006) An equilibrium model of rare-event premia and its
implication for option smirks. Review of Financial Studies, 18:131–164.
1228. Liu, L., Whited, T. and Zhang, L. (2009) Investment-based expected stock returns. Journal
1229. Liu, W. (2006) A liquidity-augmented capital asset pricing model. Journal of Financial
1230. Liu, J. and Longstaff, F. (2004) Losing money on arbitrage: optimal dynamic portfolio
choice in markets with arbitrage opportunities. Review of Financial Studies, 17:611–641.
1231. Ljungqvist, L. and Sargent, T. (2004) Recursive Macroeconomic Theory, 2nd edition. MIT
Press, Cambridge.
1232. Llorente, G., Michaely, R., Saar, G. and Wang, J. (2002) Dynamic volume-return relation of
individual stocks. Review of Financial Studies, 15:1005–1047.
1233. Lo, A. and MacKinlay, C. (1988) Stock market prices do not follow random walks: evidence
from a simple specification test. Review of Financial Studies, 1:41–66.
1234. Lo, A. and MacKinlay, C. (1989) The size and power of the variance ratio test in finite
samples. A Monte Carlo investigation. Journal of Econometrics, 40:203–238.
1235. Lo, A. and MacKinlay, C. (1990) Data-snooping biases in tests of financial asset pricing
models. Review of Financial Studies, 3:431–467.
1236. Lo, A. and MacKinlay, C. (1990) An econometric analysis of nonsynchronous trading.
Journal of Econometrics, 45:181–211.
1237. Lo, A. and MacKinlay, C. (1990) When are contrarian profits due to stock market
overreaction? Review of Financial Studies, 3:175–206.
1238. Lo, A. Mamaysky H. and Wang, J. (2001) Foundations of technical analysis: computa-
tional algorithms, statistical inference, and empirical implementation. Journal of Finance,
55:1705–1756.
1239. Lo, A. Mamaysky H. and Wang, J. (2001) Asset prices and trading volume under fixed
transactions costs. NBER working paper, no.8311.
812 References
1240. Lo, A. and Wang, J. (2000) Trading volume: definitions, data analysis and implications of
portfolio theory. Review of Financial Studies, 13:257–300.
1241. Locke, P., Sarkar, A. and Wu, L. (1999) Market liquidity and trader welfare in multiple dealer
markets: evidence from dual trading restrictions. Journal of Financial and Quantitative
Analysis, 34:35–47.
1242. Loewenstein, M. and Willard, G. (2000) Rational equilibrium asset-pricing bubbles in
continuous trading models. Journal of Economic Theory, 91:17–58.
1243. Loewenstein, M. and Willard, G. (2000) Local martingales, arbitrage, and viability. Eco-
nomic Theory, 16:135–161.
1244. Longstaff, F. (2001) Optimal portfolio choice and the valuation of illiquid securities. Review
1245. Longstaff, F. (2009) Portfolio claustrophobia: asset pricing in markets with illiquid assets.
1246. Loughran, T. (1997) Book to market across firm size, exchange, and seasonality: is there an
effect? Journal of Financial and Quantitative Analysis, 32:249–268.
1247. Loughran, T. and Ritter, J. (1996) Long-term market overreaction: the effect of low priced
stocks. Journal of Finance, 51:1959–1970.
1248. Loewenstein, G. and Prelec, D. (1992) Anomalies in intertemporal choice: evidence and an
interpretation. The Quarterly Journal of Economics, 107:573–597.
1249. Loewenstein, M. and Willard, G. (2006) The limits of investor behavior. Journal of Finance,
61:231–258.
1250. Lucas, D. (1994) Asset pricing with undiversifiable income risk and short sales constraints
deepening the equity premium puzzle. Journal of Monetary Economics, 34:325–341.
1251. Lucas, R. (1972) Expectations and the neutrality of money. Journal of Economic Theory,
2:103–104.
1252. Lucas, R. (1978) Asset pricing in an exchange economy. Econometrica, 46:1429–1445.
1253. Ludvigson, S. and Michaelides, A. (2001) Does buffer stock saving explain the smoothness
and excess sensitivity of consumption? The American Economic Review, 91:631–647.
1254. Luo, G. (1998) Market efficiency and natural selection in a commodity futures market.
1255. Luttmer, E. (1996) Asset pricing in economies with frictions. Econometrica, 64:1439–1467.
1256. Luttmer, E. (1999) What level of fixed costs can reconcile consumption and stock returns?
1257. Lynch, A. (1996) Decision frequency and synchronization across agents: implications for
aggregate consumption and equity returns. Journal of Finance, 51:1479–1497.
1258. Lynch, A. (2001) Portfolio choice and equity characteristics: characterizing the hedging
demands induced by returns predictability. Journal of Financial Economics, 62:67–130.
1259. Lynch, A. and Balduzzi, P. (2000) Predictability and transaction costs: the impact on
rebalancing rules and behavior. Journal of Finance, 55:2285–2309.
1260. Lynch, A. and Tan, S. (2011) Labor income dynamics at business cycle frequencies:
implications for portfolio choice. Journal of Financial Economics, 101:333–359.
1261. Lyons, R. (1996) Optimal transparency in a dealer market with an application to foreign
exchange. Journal of Financial Intermediation, 5:225–254.
1262. Maccheroni, F., Marinacci, M. and Rustichini, A. (2006) Ambiguity aversion, robustness,
and the variational representation of preferences. Econometrica, 74:1447–1498.
1263. Maccheroni, F., Marinacci, M. and Rustichini, A. (2006) Dynamic variational preferences.
1264. Maccheroni, F., Marinacci, M., Rustichini, A. and Taboga, M. (2009) Portfolio selection
with monotone mean-variance preferences. Mathematical Finance, 19:487–521.
1265. Mace, B. (1991) Full insurance in the presence of aggregate uncertainty. Journal of Political
Economy, 99:928–957.
1266. Machina, M. (1982) “Expected utility” analysis without the independence axiom. Econo-
metrica, 50:277–323.
References 813
1267. Machina, M. (1987) Choice under uncertainty: problems solved and unsolved. Journal of
1268. Machina, M. (1989) Dynamic consistency and non-expected utility models of choice under
uncertainty. Journal of Economic Literature, 27:1622–1668.
1269. MacKinlay, C. (1995) Multifactor models do not explain deviations from the CAPM. Journal
1270. MacKinlay, C. (1997) Event studies in economics and finance. Journal of Economic
Literature, 35:13–40.
1271. MacKinlay, C. and Richardson, M. (1991) Using generalized methods of moments to test
mean-variance efficiency. Journal of Finance, 46:511–527.
1272. Madhavan, A. (1992) Trading mechanisms in securities markets. Journal of Finance,
47:607–641.
1273. Madhavan, A. (1995) Consolidation, fragmentation, and the disclosure of trading informa-
tion. Review of Financial Studies, 8:579–603.
1274. Madhavan, A. (1996) Security prices and market transparency. Journal of Financial
1275. Madhavan, A. (2000) Market microstructure: a survey. Journal of Financial Markets, 3:205–
258.
1276. Madhavan, A. and Cheng, M. (1997) In search of liquidity: block trades in the upstairs and
downstairs markets. Review of Financial Studies, 10:175–203.
1277. Madhavan, A., Porter, D. and Weaver, D. (2005) Should securities markets be transparent?
1278. Madhavan, A., Richardosn, M. and Roomans, M. (1997) Why do security prices change? A
transaction-level analysis of NYSE stocks. Review of Financial Studies, 10:1035–1064.
1279. Madhavan, A. and Smidt, S. (1991) A Bayesian model of intraday specialist pricing. Journal
1280. Madhavan, A. and Smidt, S. (1993) An analysis of changes in specialist inventories and
quotations. Journal of Finance, 48:1595–1628.
1281. Madhavan, A. and Sofianos, G. (1998) An empirical analysis of NYSE specialist trading.
1282. Madrigal, V. and Scheinkman, J. (1997) Price crashes, information aggregation, and market-
making. Journal of Economic Theory, 75:16–63.
1283. Madrigal, V. and Smith, S. (1995) On fully revealing prices when markets are incomplete.
1284. Maenhout, P. (2004) Robust portfolio rules and asset pricing. Review of Financial Studies,
17:951–983.
1285. Magill, M. and Quinzii, M. (1996) Incomplete markets over an infinite horizon: long-lived
securities and speculative bubbles. Journal of Mathematical Economics, 26:133–170.
1286. Magill, M. and Quinzii, M. (1998) Theory of Incomplete Markets. MIT Press, Cambridge.
1287. Magill, M. and Shafer, W. (1985) Characterization of generically complete real asset
structures. Journal of Mathematical Economics, 19:167–194.
1288. Magill, M. and Shafer, W. (1991) Incomplete markets. In Handbook of Mathematical
Economics (Vol.4), W. Hildenbrand and H. Sonnenschein (eds.), Elsevier Science Publisher.
1289. Makarov, I. and Rytchkov, O. (2012) Forecasting the forecasts of others: implications for
asset pricing. Journal of Economic Theory, 147:941–966.
1290. Malkiel, B. (2003) The efficient market hypothesis and its critics. Journal of Economic
Perspectives, 17:59–82.
1291. Malloy, C., Moskowitz, T. and Vissing-Jorgensen, A. (2009) Long-run stockholder con-
sumption risk and asset returns. Journal of Finance, 64:2427–2479.
1292. Manaster, S. and Mann, S. (1996) Life in the pits: competitive market making and inventory
control. Review of Financial Studies, 9:953–975.
1293. Mankiw, G. (1986) The equity premium and the concentration of aggregate shocks. Journal
814 References
1294. Mankiw, G., Romer, D. and Shapiro, M. (1985) An unbiased reexamination of stock market
1295. Mankiw, G., Romer, D. and Shapiro, M. (1991) Stock market forecastability and volatility:
a statistical appraisal. Review of Economic Studies, 38:455–477.
1296. Mankiw, G., Rotemberg, J. and Summers, L. (1985) Intertemporal substitution in macroe-
conomics. Quarterly Journal of Economics, 100:225–252.
1297. Mankiw, G. and Shapiro, M. (1986) Risk and return: consumption beta versus market beta.
Review of Economics and Statistics, 48:452–459.
1298. Mankiw, G. and Zeldes, S. (1991) The consumption of stockholders and nonstockholders.
Journal of Financial Economics 29:97–112.
1299. Manne, H. (1966) Insider Trading: Regulation, Enforcement and Prevention. Free Press,
New York.
1300. Manove, M. (1989) The harm from insider trading and informed speculation. Quarterly
1301. Manzano, C. and Vives, A. (2011) Public and private learning from prices, strategic
substitutability and complementarity, and equilibrium multiplicity. Journal of Mathematical
1302. Marcet, A. and Singleton, K. (1999) Equilibrium asset prices and savings of heterogeneous
agents in the presence of incomplete markets and portfolio constraints. Macroeconomic
Dynamics, 3:243–277.
1303. Markovitz, H. (1952) Portfolio selection. Journal of Finance, 7:77–91.
1304. Marin, J. and Rahi, R. (1999) Speculative securities. Economic Theory, 14:653–668.
1305. Marin, J. and Rahi, R. (2000) Information revelation and market incompleteness. Review of
1306. Marsh, T. and Merton, R. (1986) Dividend variability and variance bounds tests for the
rationality of stock market prices. The American Economic Review, 76:483–498.
1307. Marsh, T. and Merton, R. (1987) Dividend behavior for the aggregate stock market. Journal
1308. Marshall, D. (1992) Inflation and asset returns in a monetary economy. Journal of Finance,
47:1315–1342.
1309. Marshall, D. and Parekh, N. (1999) Can costs of consumption adjustment explain asset
pricing puzzles? Journal of Finance, 54:623–654.
1310. Mas-Colell, A., Whinston, M. and Green, J. (1995) Microeconomic Theory. Oxford Univer-
sity Press.
1311. Massa, M. and Simonov, A. (2006) Hedging, familiarity and portfolio choice. Review of
1312. Maug, E. (1998) Large shareholders as monitors: is there a trade-off between liquidity and
control? Journal of Finance, 53:65–98.
1313. Mayers, D. (1972) Nonmarketable assets and capital market equilibrium under uncertainty.
In Studies in the Theory of Capital Markets, M. Jensen (ed.), Praeger, New York.
1314. McKelvey R. and Page, T. (1990) Public and private information: an experimental study of
information pooling. Econometrica, 58:1321–1339.
1315. McQueen, G., Pinegar, M. and Thorley, S. (1996) Delayed reaction to good news and the
cross-autocorrelation of portfolio returns. Journal of Finance, 51:889–918.
1316. Mech, T. (1993) Portfolio return autocorrelation. Journal of Financial Economics, 34:307–
344.
1317. Medrano, L.A. and Vives, X. (2004) Regulating insider trading when investment matters.
1318. Meghir, C. and Weber, G. (1996) Intertemporal nonseparability or borrowing restrictions? A
disaggregate analysis using a U.S. consumption panel. Econometrica, 64:1151–1181.
1319. Mehra, R. and Prescott, E. (1985) The equity premium: a puzzle. Journal of Monetary
1320. Mehra, R. and Prescott, E. (2003) The equity premium in retrospect. In Handbook of the
Economics of Finance (Vol. 2B), G. Constantinides, M. Harris and R. Stulz, Elsevier Science
Publisher, 889–938.
References 815
1321. Mehar, R. and Sah, R. (2002) Mood fluctuations, projection bias, and volatility of equity
prices. Journal of Economic Dynamics and Control, 26:869–897.
1322. Mei, J. (1993) A semiautoregression approach to the arbitrage pricing theory. Journal of
Finance, 48:599–619.
1323. Mendelson, H. (1987) Consolidation, fragmentation, and market performance. Journal of
1324. Mendelson, H. and Tunca, T. (2004) Strategic trading, liquidity, and information acquisition.
1325. Menzly, L., Santos, T. and Veronesi, P. (2004) Understanding predictability. Journal of
1326. Merrick, J., Naik, N. and Yadav, P. (2005) Strategic trading behavior and price distortion in
a manipulated market: anatomy of a squeeze. Journal of Financial Economics, 77:171–218.
1327. Merton, R. (1969) Lifetime portfolio selection: the continuous time case. Review of
Economics and Statistics, 51:247–257.
1328. Merton, R. (1971) Optimal consumption and portfolio rules in a continuous time model.
1329. Merton, R. (1972) An analytic derivation of the efficient portfolio frontier. Journal of
1330. Merton, R. (1973) An intertemporal capital asset pricing model. Econometrica, 41:867–887.
1331. Merton, R. (1974) On the pricing of corporate debt: the risk structure of interest rates.
1332. Merton, R. (1987) A simple model of capital market equilibrium with incomplete informa-
tion. Journal of Finance, 42:483–510.
1333. Meulbroek, L. (1992) An empirical analysis of illegal insider trading. Journal of Finance,
47:1661–1699.
1334. Meyer, J. (1987) Two-moment decision models and expected utility maximization. The
1335. Michaelides, A. (2001) Portfolio choice, liquidity constraints and stock market mean
reversion. Mimeo.
1336. Michaely, R., Thaler, R. and Womack, K. (1995) Price reactions to dividend initiations and
omissions: overreaction or drift? Journal of Finance, 50:573–607.
1337. Michaely, R. and Vila, J. (1996) Trading volume with private valuation: evidence from the
ex-dividend day. Review of Financial Studies, 9:471–509.
1338. Michaely, R., Vila, J. and Wang, J. (1996) A model of trading volume with tax-induced
heterogeneous valuation and transaction costs. Journal of Financial Intermediation, 5:340–
371.
1339. Michaely, R. and Womack, K. (1999) Conflict of interest and the credibility of underwriter
analysts recommendations. Review of Financial Studies, 12:573–608.
1340. Milevsky, M. and Young, V. (2007) Annuitization and asset allocation. Journal of Economic
Dynamics and Control, 31:3138–3177.
1341. Milgrom, P. and Stokey, N. (1982) Information, trade and common knowledge. Journal of
1342. Miller, E. (1977) Risk, uncertainty, and divergence of opinion. Journal of Finance 32:1151–
1168.
1343. Miller, M. and Scholes, M. (1972) Rates of return in relation to risk: a re-examination of
some recent findings. In Studies in the Theory of Capital Markets, M. Jensen (ed.), Praeger,
New York.
1344. Milne, F. (1979) Consumer preferences, linear demand functions and aggregation in
competitive asset markets. Review of Economic Studies, 50:407–417.
1345. Milne, F. (1988) Arbitrage and diversification in a general equilibrium asset economy.
1346. Milne, F. (1995) Finance Theory and Asset Pricing. Clarendon Press, Oxford.
1347. Mitchell, M. and Mulherin, H. (1994) The impact of public information on the stock market.
816 References
1348. Modigliani, F. and Miller, M. (1958) The cost of capital, corporate finance, and the theory
of investment. The American Economic Review, 48:261–297.
1349. Modigliani, F. and Miller, M. (1963) Corporate income taxes and the cost of capital: a
correction. The American Economic Review, 53:433–443.
1350. Moinas, S. and Pouget, S. (2013) The bubble game: an experimental study of speculation.
1351. Montrucchio, L. and Privileggi, F. (2001) On the fragility of bubbles in equilibrium asset
pricing models. Journal of Economic Theory, 101:158–188.
1352. Morris, S. (1994) Trade with heterogeneous prior beliefs and asymmetric information.
1353. Morris, S. (1996) Speculative investor behavior and learning. Quarterly Journal of Eco-
nomics, 111:1111–1131.
1354. Moskowitz, T. (2003) An analysis of risk and pricing anomalies. Review of Financial Studies,
16:417–457.
1355. Moskowitz, T. and Grinblatt, M. (1999) Do industries explain momentum? Journal of
Finance, 44:1249–1290.
1356. Moskowitz, T. and Vissing-Jorgensen, A. (2002) The returns to entrepreneurial investment:
a private equity premium puzzle? The American Economic Review, 92:745–778.
1357. Mossin, J. (1966) Equilibrium in a capital asset market. Econometrica, 34:261–276.
1358. Mossin, J. (1968) Aspects of rational insurance purchasing. Journal of Political Economy,
79:553–568.
1359. Muendler, M.A. (2007) The possibility of informationally efficient markets. Journal of
1360. Muhle-Karbe, J. and Guasoni, P. (2013) Portfolio choice with transaction costs: a user’s
guide. Paris-Princeton Lectures on Mathematical Finance 2013, V. Henderson and R. Sircar
(eds.), Springer, Cham.
1361. Mukerji, S. and Tallon, J. (2001) Ambiguity aversion and incompleteness of financial
markets. Review of Economic Studies, 68:883–904.
1362. Mukerji, S. and Tallon, J. (2004) Ambiguity aversion and the absence of indexed debt.
1363. Munk, C. (2013) Financial Asset Pricing Theory. Oxford University Press.
1364. Musiela, M. and Rutkowski, M. (1997) Martingale Methods in Financial Modelling.
Springer, Berlin/Heidelberg.
1365. Nachman, D. (1989) Spanning and completeness with options. Review of Financial Studies,
1:311–328.
1366. Naik, N. (1997) On the aggregation of information in competitive markets: the dynamic
case. Journal of Economic Dynamics and Control, 21:1199–1227.
1367. Naik, N. (1997) Multi-period information markets. Journal of Economic Dynamics and
Control, 21:1229–1258.
1368. Naik, N., Neuberger, A. and Viswanathan, S. (1999) Trade disclosure regulation in markets
with negotiated trades. Review of Financial Studies, 12:873–900.
1369. Naik, V. (1995) Finite state securities market models and arbitrage. In Handbooks in Opera-
tions Research and Management Science, Vol. 9 (Finance), R.A. Jarrow, V. Maksimovic and
W.T. Ziemba (eds.), North Holland, Amsterdam.
1370. Naranjo, A., Nimalendran, M. and Ryngaert, M. (1998) Stock returns, dividend yields, and
taxes. Journal of Finance, 53:2029–2057.
1371. Narayanan, R. (2000) Insider trading and the voluntary disclosure of information by firms.
1372. Nawalkha, M. (1997) A multibeta representation theorem for linear asset pricing Theories.
1373. Nelson, C. and Kim, M. (1993) Predictable stock returns: the role of small sample bias.
1374. Nguyen, P. and Portait, R. (2002) Dynamic asset allocation with mean variance preferences
and a solvency constraint. Journal of Economic Dynamics and Control, 26:11–32.
References 817
1375. Nielsen, L. (1990) Existence of equilibrium in CAPM. Journal of Economic Theory, 32:223–
229.
1376. Nielsen, L. (1987) Portfolio selection in the mean-variance model: a note. Journal of
Finance, 42:1371–1376.
1377. Nielsen, L. (1992) Positive prices in CAPM. Journal of Finance, 47:791–808.
1378. Noe, C. (1999) Voluntary disclosures and insider transactions. Journal of Accounting
1379. Odean, T. (1998) Volume, volatility, price, and profit when all traders are above the average.
1380. Odean, T. (1998) Are investors reluctant to realize their losses? Journal of Finance, 53:1775–
1798.
1381. Odean, T. (1999) Do investors trade too much? The American Economic Review, 89:1279–
1298.
1382. Ofek, E. and Richardson M. (2003) DotComMania: the rise and fall of internet stock prices.
1383. Ogaki, M. and Zhang, Q. (2001) Decreasing relative risk aversion and tests of risk sharing.
1384. O’Hara, M. (1995) Market Microstructure Theory. Blackwell Publishers.
1385. O’Hara, M. and Oldfield, G. (1986) The microeconomics of market making. Journal of
1386. Orosel, G. (1996) Informational efficiency and welfare in the stock market. European
1387. Orosel, G. (1998) Participation costs, trend chasing, and volatility of stock prices. Review of
1388. Otrok, C., Ravikumar, B. and Whiteman, C. (2002) Habit formation: a resolution of the
equity premium puzzle? Journal of Monetary Economics 49:1261–1288.
1389. Owen, J. and Rabinovitch, R. (1983) On the class of elliptical distributions and their
applications to the theory of portfolio choice. Journal of Finance, 37:745–752.
1390. Ozguz, A. (2009) Good times or bad times? Investors’ uncertainty and stock returns. Review
1391. Pagan, A. (1996) The econometrics of financial markets. Journal of Empirical Finance,
3:15–102.
1392. Pagan, A. and Schwert, W. (1990) Alternative models for conditional stock volatility.
Journal of Econometrics, 45:267–290.
1393. Pagano, M. (1989) Endogenous market thinness and stock price volatility. Review of
1394. Pagano, M. (1989) Trading volume and asset liquidity. Quarterly Journal of Economics,
104:255–274.
1395. Pagano, M. and Roell, A. (1996) Transparency and liquidity: a comparison of auction and
dealer markets with informed trading. Journal of Finance, 51:123–145.
1396. Paiella, M. (2004) Heterogeneity in financial market participation: appraising its implica-
tions for the C-CAPM. Review of Finance, 8:445–480.
1397. Palfrey, T. and Wang, S. (2012) Speculative overpricing in asset markets with information
flows. Econometrica, 80:1937–1976.
1398. Palomino, F. (1996) Noise trading in small markets. Journal of Finance, 51:1537–1550.
1399. Palumbo, M. (1999) Uncertain medical expenses and precautionary saving near the end of
the life cycle. Review of Economic Studies, 66:395–421.
1400. Park, C. and Irwin, S. (2007) What do we know about the profitability of technical analysis?
Journal of Economic Surveys, 21:786–826.
1401. Parker, J. and Julliard, C. (2005) Consumption risk and the cross section of expected returns.
1402. Parlour, C. (1998) Price dynamics in limit order markets. Review of Financial Studies,
11:789–816.
818 References
1403. Parlour, C. and Seppi, D. (2003) Liquidity-based competition for order flow. Review of
1404. Pascucci, A. and Runggaldier, W. (2012) Financial Mathematics: Theory and Problems for
Multi-period Models. Springer, Milan.
1405. Pástor, L. (2000) Portfolio selection and asset pricing models. Journal of Finance, 55:179–
223.
1406. Pástor, L. and Stambaugh, R. (2000) The equity premium and structural breaks. Journal of
Finance, 56:1207–1239.
1407. Pástor, L. and Stambaugh, R. (2003) Liquidity risk and expected stock returns. Journal of
1408. Pástor, L. and Veronesi, P. (2003) Stock valuation and learning about profitability. Journal
of Finance, 58:1749–1789.
1409. Pástor, L. and Veronesi, P. (2006) Was there a Nasdaq bubble in the late 1990s? Journal of
1410. Patelis, A. (1997) Stock return predictability and the role of monetary policy. Journal of
Finance, 52:1951–1972.
1411. Pennacchi, G. (2008) Theory of Asset Pricing. Pearson.
1412. Peress, J. (2004) Wealth, information acquisition and portfolio choice. Review of Financial
Studies, 17:879–914.
1413. Peress, J. (2011) Erratum to “Wealth, information acquisition and portfolio choice”. Review
1414. Pesaran, H. (1989) The Limits to Rational Expectations. Blackwell Publishers.
1415. Pesaran, H. and Timmermann, A. (1995) Predictability of stock returns: robustness and
economic significance. Journal of Finance, 50:1201–1227.
1416. Petkova, R. (2006) Do the Fama-French factors proxy for innovations in predictive
variables? Journal of Finance, 61:581–612.
1417. Petkova, R. and Zhang, L. (2005) Is value riskier than growth? Journal of Financial
1418. Pham, H. (2000) On quadratic hedging in continuous time. Mathematical Methods of
Operations Research, 51:315–339.
1419. Pietra, T. and Siconolfi, P. (1998) Fully revealing equilibria in sequential economies with
asset markets. Journal of Mathematical Economics, 29:211–223.
1420. Platen, E. and Heath, D. (2006) A Benchmark Approach to Quantitative Finance. Springer,
Berlin/Heidelberg.
1421. Pliska, S. (1997) Introduction to Mathematical Finance. Blackwell Publishers.
1422. Plott, C. and Sunder, S. (1988) Rational expectations and the aggregation of diverse
information in laboratory security markets. Econometrica, 56:1085–1118.
1423. Po-Hsuan, H. and Chung-Ming, K. (2005) Reexamining the profitability of technical
analysis with data snooping checks. Journal of Financial Econometrics, 3:606–628.
1424. Polemarchakis, H. and Ku, B. (1990) Options and equilibrium. Journal of Mathematical
1425. Polemarchakis, H. and Siconolfi, P. (1993) Asset markets and the information revealed by
prices. Economic Theory, 3:645–661.
1426. Polkovnichenko, V. (2007) Life-cycle portfolio choice with additive habit formation prefer-
ences and uninsurable labor income risk. Review of Financial Studies, 20:83–124.
1427. Pontiff, J. and Schall, L. (1998) Book-to-market ratios as predictors of market returns.
1428. Porter, D. and Weaver, D. (1998) Post-trade transparency on Nasdaq’s national market
system. Journal of Financial Economics, 50:231–252.
1429. Poterba, J. and Samwick, A. (1999) Household portfolio allocation over the life-cycle. NBER
working paper, no. 6185.
1430. Poterba, J. and Summers, L. (1988) Mean reversion in stock prices: evidence and implica-
tions. Journal of Financial Economics, 22:27–60.
References 819
1431. Poterba, J. and Weisbenner, S. (2001) Capital gains tax rules, tax loss trading, and turn-of-
the-year returns. Journal of Finance 56:353–368.
1432. Pratt, J. (1964) Risk aversion in the small and in the large. Econometrica, 32:122–136.
1433. Pratt, J. and Zeckauser, R. (1987) Proper risk aversion. Econometrica, 55:143–154.
1434. Presidential Task Force on Market Mechanism (1988) Report of the Presidential Task Force
on Market Mechanism. United States Government, Washington.
1435. Protter, P. (2013) A mathematical theory of financial bubbles. Paris-Princeton Lectures on
Mathematical Finance, Henderson, V. and Sircar, R. (eds.), Springer, Cham, 1–108.
1436. Quiggin, J. (1982) A theory of anticipated utility. Journal of Economic Behavior and
Organization, 3:323–343.
1437. Radner, R. (1972) Existence of equilibrium of plans, prices, and price expectations in a
sequence of markets. Econometrica, 40:289–303.
1438. Radner, R. (1979) Rational expectations equilibrium: generic existence and the information
revealed by price. Econometrica, 47:655–678.
1439. Radner, R. (1982) Equilibrium under uncertainty. In Handbook of Mathematical Economics
(Vol.I), K. Arrow and M. Intriligator (eds.), Elsevier Science Publisher.
1440. Rahi, R. (1985) Partially revealing rational expectations equilibria with nominal assets.
1441. Reinganum, M. (1981) The arbitrage pricing theory: some empirical results. Journal of
Finance, 36:313–322.
1442. Reinganum, M. (1981) A new empirical perspective on the CAPM. Journal of Financial and
Quantitative Analysis, 16:439–462.
1443. Reinganum, M. (1981) Misspecification of capital asset pricing : Empirical anomalies based
on earnings’ yields and market values. Journal of Financial Economics, 9:19–46.
1444. Reinganum, M. (1983) The anomalous stock market behavior of small firms in January:
Empirical tests for tax-loss selling effects. Journal of Financial Economics, 12:89–104.
1445. Reisman, H. (1992) Reference variables, factor structure, and the approximate multibeta
representation. Journal of Finance, 47:1303–1314.
1446. Repullo, R. (1999) Some remarks on Leland’s model of insider trading. Economica, 66:359–
374.
1447. Richards, A. (1997) Winner-loser reversals in national stock market indices: can they be
explained? Journal of Finance, 52:2129–2144.
1448. Richardson, M. (1993) Temporary components of stock prices: a skeptic view. Journal of
Business and Economic Statistics, 11:199–207.
1449. Richardson, M. and Smith, T. (1994) A unified approach to testing for serial correlation in
stock returns. Journal of Business, 67:371–399.
1450. Richardson, M. and Stock, J. (1989) Drawing inferences from statistics based on multi-year
asset returns. Journal of Financial Economics, 25:323–348.
1451. Rietz, T. (1988) The equity risk premium: a solution. Journal of Monetary Economics,
22:117–131.
1452. Rigotti, L. and Shannon, C. (2005) Uncertainty and risk in financial markets. Econometrica,
73:203–243.
1453. Rigotti, L., Strzalecki, T. and Shannon, C. (2008) Subjective beliefs and ex-ante trade.
1454. Rinaldi, F. (2009) Endogenous incompleteness of financial markets: the role of ambiguity
and ambiguity aversion. Journal of Mathematical Economics, 45:872–893.
1455. Rochet, J. and Vila, J. (1994) Insider trading without normality. Review of Economic Studies,
61:131–152.
1456. Roell, A. (1990) Dual capacity and the quality of the market. Journal of Financial
1457. Roll, R. (1977) A critique of the asset’s pricing theory’s test: part I. Journal of Financial
1458. Roll, R. (1983) Vas ist das? The turn-of-the-year effect and the return premia of small firms.
Journal of Portfolio Management 9:18–28.
820 References
1459. Roll, R. (1984) Orange juice and weather. The American Economic Review, 74:861–880.
1460. Roll, R. (1988) R2 . Journal of Finance, 43:541–566.
1461. Roll, R. and Ross, S. (1980) An empirical investigation of the arbitrage pricing theory.
1462. Roll, R. and Ross, S. (1994) On the cross-sectional relation between expected returns and
betas. Journal of Finance, 49:101–121.
1463. Romer, D. (1993) Rational asset-price movements without news. The American Economic
Review, 83:1112–1130.
1464. Ross, S. (1976) The arbitrage pricing theory of capital asset pricing. Journal of Economic
Theory, 13:341–360.
1465. Ross, S. (1976) Options and efficiency. Quarterly Journal of Economics, 90:75–89.
1466. Ross, S. (1977) Return, risk and arbitrage. In Risk and Return in Finance, I. Friend and
J. Bicksler (eds.), Ballinger, Cambridge.
1467. Ross, S. (1978) Mutual fund separation in financial theory: the separating distributions.
1468. Ross, S. (1981) Some stronger measures of risk aversion in the small and large with
applications. Econometrica, 49:621–638.
1469. Ross, S. (1987) Arbitrage and martingales with taxation. Journal of Political Economy,
95:371–393.
1470. Ross, S. (1989) Information and volatility: the no arbitrage martingale approach to timing
and resolution irrelevance. Journal of Finance, 44:1–18.
1471. Rosu, I. (2009) A dynamic model of the limit order book. Review of Financial Studies,
22:4601–4641.
1472. Rothschild, M. and Stiglitz, J. (1970) Increasing risk I: a definition. Journal of Economic
Theory, 2:225–243.
1473. Rothschild, M. and Stiglitz, J. (1971) Increasing risk II: its economic consequences. Journal
1474. Routledge, B. (1999) Adaptive learning financial markets. Review of Financial Studies,
12:1165–1202.
1475. Routledge, B. and Zin, S. (2009) Model uncertainty and liquidity. Review of Economic
Dynamics 12:543–566.
1476. Routledge, B. and Zin, S. (2010) Generalized disappointment aversion and asset prices.
1477. Rouwenhorst, G. (1995) Asset pricing implications of equilibrium business cycle models. In
Frontiers of Business Cycle Research, T. Cooley (ed.), Princeton University Press, 294–330.
1478. Rouwenhorst, G. (1998) International momentum strategies. Journal of Finance, 53:267–
284.
1479. Rouwenhorst, G. (1999) Local return factors and turnover in emerging stock markets.
1480. Rozeff, M. (1984) Dividend yields are equity risky premiums. Journal of Portfolio Manage-
ment, 11:68–75.
1481. Rozeff, M. and Kinney, W. (1976) Capital market seasonality: the case of stock returns.
1482. Rozeff, M. and Zaman, M. (1988) Market efficiency and insider trading: new evidence.
Journal of Business, 61:25–44.
1483. Rozeff, M. and Zaman, M. (1998) Overreaction and insider trading: evidence from growth
and value portfolios. Journal of Finance, 53:701–716.
1484. Rozen, K. (2010) Foundations of intrinsic habit formation. Econometrica, 78:1341–1373.
1485. Rubinstein, M. (1974) An aggregation theorem for securities markets. Journal of Financial
1486. Rubinstein, M. (1975) Securities market efficiency in an Arrow-Debreu economy. The
1487. Rubinstein, M. (1976) The valuation of uncertain income streams and the pricing of options.
Bell Journal of Economics, 7:407–425.
References 821
1488. Runkle, D. (1991) Liquidity constraints and the permanent income hypothesis: evidence
from panel data. Journal of Monetary Economics, 27:73–98.
1489. Sadka, R. and Scherbina, A. (2007) Analyst disagreement, mispricing, and liquidity. Journal
of Finance, 62: 2367–2403.
1490. Salyer, K. (1998) Crash states and the equity premium: solving one puzzle raises another.
1491. Samuelson, P. (1965) Proof that properly anticipated prices fluctuate randomly. Industrial
Management Review, 6:41–49.
1492. Samuelson, P. (1967) General proof that diversification pays. Journal of Financial and
Quantitative Analysis, 2:1–13.
1493. Samuelson, P. (1969) Lifetime portfolio selection by dynamic stochastic programming.
Review of Economics and Statistics, 51:239–246.
1494. Sandas, P. (2001) Adverse selection and competitive market making: empirical evidence
from a limit order market. Review of Financial Studies, 14:705–734.
1495. Sandroni, A. (2000) Do markets favor agents able to make accurate predictions? Economet-
rica, 68:1303–1341.
1496. Santos, M. and Woodford, M. (1997) Rational asset pricing bubbles. Econometrica, 65:19–
58.
1497. Santos, T. and Veronesi, P. (2006) Labor income and predictable stock returns. Review of
1498. Santos, T. and Veronesi, P. (2010) Habit formation, the cross section of stock returns and the
cash flow risk puzzle. Journal of Financial Economics, 98:385–413.
1499. Savage, L. (1954) The Foundations of Statistics. Wiley, New York.
1500. Schachermayer, W. (2004) The fundamental theorem of asset pricing under proportional
transaction costs in finite discrete time. Mathematical Finance, 14:19–48.
1501. Schachermayer, W. (2010) The fundamental theorem of asset pricing. In Encyclopedia of
Quantitative Finance, Vol. 2, R. Cont (ed.), Wiley, Chichester, 792–801.
1502. Scheinkman, J. and Xiong, W. (2003) Overconfidence and speculative bubbles. Journal of
1503. Scheinkman, J. and Xiong, W. (2004) Heterogeneous beliefs, speculation and trading in
financial markets. Paris-Princeton Lectures on Mathematical Finance 2003, Lecture Notes
in Mathematics Volume 1847, 217–250.
1504. Schlee, E. (2001) The value of information in efficient risk-sharing arrangements The
1505. Schmeidler, D. (1989) Subjective probability and expected utility without additivity. Econo-
metrica, 57:571–587.
1506. Schmeidler, D. (1989) A rational expectations equilibrium with informative trading volume.
1507. Schnitzlen, C. (1996) Call and continuous trading mechanisms under asymmetric informa-
tion: an experimental investigation. Journal of Finance, 51:613–636.
1508. Schweizer, M. (2000) A guided tour through quadratic hedging approaches. In Option
Pricing, Interest Rates and Risk Management, E. Jouini, J. Cvitanic and M. Musiela (eds.),
Cambridge University Press, 538–574.
1509. Schwert, W. (1989) Why does stock market volatility change over time? Journal of Finance,
44:1115–1153.
1510. Schwert, W. (2001) Anomalies and market efficiency. In Handbook of the Economics of
Finance, Vol. 2, G. Constantinides, M. Harris and R. Stulz (eds.), Elsevier Science Publisher,
939–974.
1511. Schwert, W. and Seguin, P. (1990) Heteroskedasticity in stock returns. Journal of Finance,
45:1129–1155.
1512. Schwert, W. and Seguin, P. (1993) Securities transaction taxes: an overview of costs, benefits
and unresolved questions. Financial Analysts Journal, 49:27–35.
1513. Scott, L. (1985) The present value model of stock prices: regression tests and Monte Carlo
results. The Review of Economics and Statistics, 67:599–605.
822 References
1514. Segal, U. and Spivak, A. (1990) First order versus second order risk aversion. Journal of
1515. Sentana, E. and Wadhwani, S. (1992) Feedback traders and stock return autocorrelations:
evidence from a century daily data. The Economic Journal, 102:415–425.
1516. Seo, K. (2009) Ambiguity and second-order belief. Econometrica, 77:1575–1605.
1517. Seppi, D. (1990) Equilibrium block trading and asymmetric information. Journal of Finance,
45:73–94.
1518. Seppi, D. (1992) Block trading and information revelation around quarterly earnings
announcements. Review of Financial Studies, 5:281–305.
1519. Seppi, D. (1997) Liquidity provision with limit orders and strategic specialist. Review of
1520. Seyhun, H. (1986) Outsiders’ profits, costs of trading, and the markets efficiency. Journal of
1521. Seyhun, H. (1998) Investment Intelligence from Insider trading. MIT Press, Cambridge.
1522. Shalen, C. (1993) Volume, volatility, and the dispersion of beliefs. Review of Financial
Studies, 6:405–34.
1523. Shanken, J. (1982) The arbitrage pricing theory: Is it testable? Journal of Finance, 37:1129–
1140.
1524. Shanken, J. (1985) Multi-beta CAPM or equilibrium-APT? A reply. Journal of Finance,
40:1189–1196.
1525. Shanken, J. (1987) Multivariate proxies and asset pricing relations: living with the Roll
critique. Journal of Financial Economics, 18:91–110.
1526. Shanken, J. (1992) On the estimation of beta-pricing models. Review of Financial Studies,
5:1–55.
1527. Shanken, J. (1996) Statistical methods in tests of portfolio efficiency: a synthesis. In Hand-
book of Statistics, Vol.14, G.S. Maddala and C.R. Rao (eds.), North-Holland, Amsterdam.
1528. Shanken, J. and Zhou, G. (2007) Estimating and testing beta pricing models: alternative
methods and their performance in simulations. Journal of Financial Economics, 84:40–86.
1529. Sharpe, W. (1964) Capital asset prices: a theory of market equilibrium under conditions of
risks. Journal of Finance, 19:425–442.
1530. Sharpe, W. (1991) Capital asset prices with and without negative holdings. Journal of
Finance, 46:489–509.
1531. Shefrin, H. and Statman, M. (1985) The disposition to sell winners too early and ride losers
too long. Journal of Finance, 40:777–790.
1532. Shefrin, H. and Statman, M. (1994) Behavioral capital asset pricing. Journal of Financial
1533. Shiller, R. (1981) Do stock prices move too much to be justified by subsequent changes in
dividends? The American Economic Review, 71:421–435.
1534. Shiller, R. (1981) The use of volatility measures in assessing market efficiency. Journal of
Finance, 36:291–304.
1535. Shiller, R. (1984) Stock prices and social dynamics. Brookings Papers on Economic Activity,
2:457–498.
1536. Shiller, R. (1986) The Marsh-Merton model of managers smoothing of dividends. The
1537. Shiller, R. (1988) The probability of gross violations of a present value variance inequality.
1538. Shiller, R. (1989) Market Volatility. MIT Press, Cambridge.
1539. Shin, J. (1996) The optimal regulation of insider trading. Journal of Financial Intermedia-
tion, 5:49–73.
1540. Shleifer, A. and Summers, L. (1990) The noise trader approach to finance. Journal of
1541. Shleifer, A. and Vishny, R. (1997) The limits of arbitrage. Journal of Finance, 52:35–55.
1542. Siegel, J. (1988) Stocks for the Long Run. McGraw-Hill.
References 823
1543. Siegel, J. (1999) The shrinking equity premium. Journal of Portfolio Management, 26:10–
17.
1544. Simaan, Y., Weaver, D. and Whitcomb, D. (2003) Market maker quotation behavior and
pretrade transparency. Journal of Finance, 58:1247–1268.
1545. Simon H. (1955) A behavioral model of rational choice. Quarterly Journal of Economics,
69:99–118.
1546. Simon, H. (1976) From substantive to procedural rationality. In Method and Appraisal in
Economics, S. Latsis (ed.), Cambridge University Press.
1547. Simsek, A. (2013) Belief disagreements and collateral constraints. Econometrica, 81:1–53.
1548. Singleton, K. (1987) Asset prices in a time series model with disparately informed competi-
tive traders. In New Approaches to Monetary Economics, W. Barnett and K. Signleton (eds.),
North-Holland, Amsterdam.
1549. Singleton, K. (1990) Specification and estimation of intertemporal asset pricing models. In
Handbook of Monetary Economics, Vol.1, B. Friedman and F. Hahn (eds.), North-Holland,
Amsterdam.
1550. Skinner, J. (1988) Risky income, life cycle consumption, and precautionary saving. Journal
1551. Smith, W. (2001) How does the spirit of capitalism affect stock market prices? Review of
1552. Smith, V., Suchanek, G. and Williams, W. (1988) Bubbles, crashes and endogenous
expectations in experimental spot asset markets. Econometrica, 56:1119–1151.
1553. Snow, K. (1991) Diagnosis asset pricing models using the distribution of asset returns.
1554. Sogner, L. and Mitlohner, H. (2002) Consistent expectations equilibria and learning in a
stock market. Journal of Economic Dynamics and Control, 26:171–185.
1555. Solnik, (1974) An equilibrium model of the international capital market. Journal of
Economic Theory 8:500–524.
1556. Solow, R. (1956) A contribution to the theory of economic growth. Quarterly Journal of
1557. Song, F. and Zhang, J. (2005) Securities transaction tax and market volatility. The Economic
Journal, 115:1103–1120.
1558. Spiegel, M. and Subrahmanyam, A. (1992) Informed speculation and hedging in a noncom-
petitive securities market. Review of Financial Studies, 5:307–329.
1559. Spiegel, M. and Subrahmanyam, A. (2000) Asymmetric information and news disclosure
rules. Journal of Financial Intermediation, 9:363–403.
1560. Sriboonchitta, S., Wong, W.-K., Dhompongsa, S. and Nguyen, H.T. (2010) Stochastic
Dominance and Applications to Finance, Risk and Economics. CRC Press, Boca Raton.
1561. Stambaugh, R. (1982) On the exclusion of assets from tests of the two parameter model.
1562. Stambaugh, R. (1983) Arbitrage pricing with information. Journal of Financial Economics,
12:357–369.
1563. Stambaugh, R., Yu, J. and Yuan, Y. (2012) The short of it: investor sentiment and anomalies.
1564. Starmer, C. (2000) Developments in non-expected utility theory: the hunt for a descriptive
theory of choice under risk. Journal of Economic Literature, 38:332–382.
1565. Statman, M., Thorley, S. and Vorkink, K. (2006) Investor overconfidence and trading
volume. Review of Financial Studies, 19:1531–1565.
1566. Stattman, D. (1980) Book values and stock returns. The Chicago MBA: A Journal of Selected
Papers, 4:25–45.
1567. Stein, J. (1987) Informational externalities and welfare-reducing speculation. Journal of
1568. Stiglitz, J. (1982) The inefficiency of the stock market equilibrium. Review of Economic
Studies, 49:241–261.
824 References
1569. Stiglitz, J. (1989) Using tax policy to curb speculative short term trading. Journal of
Financial Services Research, 3:101–115.
1570. Stoll, H. (1978) The supply of dealer services in securities markets. Journal of Finance,
33:1133–1151.
1571. Stoll, H. and Whaley, R. (1990) Stock market structure and volatility. Review of Financial
Studies, 3:37–71.
1572. Storesletten, K., Telmer, C. and Yaron, A. (2004) Consumption and risk sharing over the life
cycle. Journal of Monetary Economics, 51:609–633.
1573. Storesletten, K., Telmer, C. and Yaron, A. (2004) Asset pricing with idiosyncratic risk and
overlapping generations. Review of Economic Dynamics, 10:519–548.
1574. Stulz, R. (1981) A model of international asset pricing. Journal of Financial Economics,
9:383–406.
1575. Subrahmanyam, A. (1991) Risk aversion, market liquidity, and price efficiency. Review of
1576. Subrahmanyam, A. (1994) Circuit breakers and market volatility: a theoretical perspective.
1577. Subrahmanyam, A. (1998) Transaction taxes and financial market equilibrium. Journal of
Business, 71:81–118.
1578. Summers, L. (1986) Does the stock market rationality reflect fundamental values? Journal
of Finance, 41:591–601.
1579. Summers, L. and Summers, V. (1989) When financial markets work too well: a cautious case
for a securities transactions tax. Journal of Financial Services Research, 3:261–268.
1580. Sundaresan, S. (1989) Intertemporally dependent preferences and the volatility of consump-
tion and wealth. Review of Financial Studies, 2:73–88.
1581. Sunder, S. (1992) Market for information: experimental evidence. Econometrica, 60:667–
695.
1582. Tallarini, T. and Zhang, H. (2005) External habit and the cyclicality of expected stock
returns. Journal of Business, 78:1023–1048.
1583. Telmer, C. (1993) Asset pricing puzzles and incomplete markets. Journal of Finance,
48:1803–1832.
1584. Teoh, S. and Hwang, C. (1991) Nondisclosure and adverse disclosure as signals of firm
value. Review of Financial Studies, 4:283–313.
1585. Tepla, L. (2000) Optimal hedging and valuation of nontraded assets. European Finance
Review 4:231–251.
1586. Tetlock, P. (2010) Does public financial news resolve asymmetric information. Review of
1587. Thaler, R. (1999) Mental accounting matters. Journal of Behavioral Decision Making,
12:183–206.
1588. Thaler, R., Tversky, A., Kahneman, D. and Schwartz, A. (1997) The effect of myopia and
loss aversion on risk taking: an experimental test. Quarterly Journal of Economics, 112:647–
661.
1589. Thorisson, H. (1995) Coupling methods in probability theory. Scandinavian Journal of
Statistics, 22:159–182.
1590. Timmermann, A. (1993) How learning in financial markets generates excess volatility and
predictability in stock prices. Quarterly Journal of Economics, 108:1135–1145.
1591. Timmermann, A. (1994) Present value models with feedback. Journal of Economic Dynam-
ics and Control, 18:1093–1119.
1592. Timmermann, A. (1994) Can agents learn to form rational expectations? Some results on
convergence and stability of learning in the UK stock market. The Economic Journal,
104:777–797.
1593. Timmermann, A. (1994) Excess volatility and predictability of stock prices in autoregressive
dividend models with learning. Review of Economic Studies, 63:523–557.
1594. Tirole, J. (1982) On the possibility of speculation under rational expectations. Econometrica,
50:1163–1181.
References 825
1595. Tirole, J. (1985) Asset bubbles and overlapping generations. Econometrica, 53:1499–1528.
1596. Tobin, J. (1958) Liquidity preferences as behavior towards risk. Review of Economic Studies,
25:65–86.
1597. Tobin, J. (1984) On the efficiency of the financial system. Lloyds Bank Review, 153:1–25.
1598. Townsend, R. (1983) Forecasting the forecasts of others. Journal of Political Economy,
91:546–588.
1599. Townsend, R. (1994) Risk and insurance in village India. Econometrica, 62:539–591.
1600. Trojani, F. and Vanini, P. (2004) Robustness and ambiguity aversion in general equilibrium.
1601. Tsay, R. (2010) Analysis of Financial Time Series, 3rd ed. Wiley, Chichester.
1602. Tuckman, B. and Vila, J. (1992) Arbitrage with holding costs: a utility based approach.
1603. Tversky, A. and Kahneman, D. (1992) Advances in prospect theory: cumulative representa-
tion of uncertainty. Journal of Risk and Uncertainty, 5:297–323.
1604. Umlauf, S. (1993) Transaction taxes and the behavior of the Swedish stock market. Journal
1605. Uppal, R. and Wang, T. (2003) Model misspecification and under diversification. Journal of
Finance, 58:2465–2486.
1606. Van Ness, B., Van Ness, R. and Pruitt, S. (2000) The impact of the reduction in tick
increments in major US markets on spreads, depth, and volatility. Review of Quantitative
Finance and Accounting, 15:153–167.
1607. Van Nieuwerburgh, S. and Veldkamp, L. (2009) Information immobility and the home bias
puzzle. Journal of Finance 64:1187–1215.
1608. Van Nieuwerburgh, S. and Veldkamp, L. (2010) Information acquisition and under-
diversification. Review of Economic Studies, 77:779–805.
1609. van Rooij, M., Lusardi, A. and Alessie, R. (2011) Financial literacy and stock market
participation. Journal of Financial Economics, 101:449–472.
1610. Varian, H. (1984) Microeconomic Analysis. Norton, New York.
1611. Varian, H. (1989) Differences in opinion in financial markets. In Financial Risk: Theory,
Evidence and Implications, C. Stone (ed.), Kluwer Academic Publishers.
1612. Vassalou, M. (2003) News related to future GDP growth as a risk factor in equity returns.
1613. Vassalou, M. and Xing, Y. (2004) Default risk in equity returns. Journal of Finance, 59:831–
868.
1614. Vayanos, D. (1998) Transaction costs and asset prices: a dynamic equilibrium model. Review
1615. Veldkamp, L. (2006) Information markets and the comovement of asset prices. Review of
1616. Veronesi, P. (1999) Stock market overreaction to bad news in good times: a rational
expectations equilibrium model. Review of Financial Studies, 12:975–1007.
1617. Veronesi, P. (2004) The Peso problem hypothesis and stock market returns. Journal of
1618. Verrecchia, R. (1982) Information acquisition in a noisy rational expectations economy.
1619. Verrecchia, R. (1993) Insider trading and the voluntary disclosure of information by firms.
1620. Verrecchia, R. (1993) Discretionary disclosure. Journal of Accounting and Economics,
5:179–194.
1621. Verrecchia, R. (2001) Essays on disclosure. Journal of Accounting and Economics, 32:97–
180.
1622. Viceira, L. (2001) Optimal portfolio choice for long horizon investors with nontradable labor
income. Journal of Finance, 56:433–470.
1623. Vila, J. (1989) Simple games of market manipulation. Economics Letters, 29:21–26.
826 References
1624. Villaverde, J.F. and Krueger, D. (2011) Consumption and saving over the life cycle: how
important are consumer durables? Macroeconomic Dynamics, 15:725–770.
1625. Vissing-Jorgensen, A. (2000) Towards an explanation of household portfolio choice het-
erogeneity: nonfinancial income and participation cost structures. NBER working paper,
no.8884.
1626. Vissing-Jorgensen, A. (2002) Limited asset market participation and the elasticity of
intertemporal substitution. Journal of Political Economy, 110:825–853.
1627. Viswanathan, S. and Wang, J. (1999) Market architecture: limit order books versus dealer-
ship markets. Journal of Financial Markets, 5:127–167.
1628. Vives, X. (1995) Short-term investment and the informational efficiency of the market.
1629. Vives, X. (1995) The speed of information revelation in a financial market mechanism.
1630. Vives, X. (2008) Information and Learning in Markets. Princeton University Press.
1631. Vives, X. (2014) On the possibility of informationally efficient markets. Journal of the
European Economic Association, 12:1200–1239.
1632. Von Neumann, J. and Morgenstern, O. (1953) Theory of Games and Economic Behaviour.
Princeton University Press.
1633. Vuolteenaho, T. (2002) What drives firm level stock returns? Journal of Finance, 57:233–
264.
1634. Wachter, J. (1999) Risk aversion and allocation to long-term bonds. Journal of Economic
Theory, 112:325–333.
1635. Wachter, J. (2006) A consumption-based model of the term structure of interest rates.
1636. Wachter, J. (2001) Portfolio and consumption decisions under mean-reverting returns: an
exact solution for complete markets. Journal of Financial and Quantitative Analysis, 37:63–
91.
1637. Wachter, J. (2013) Can time-varying risk of rare disasters explain aggregate stock market
volatility? Journal of Finance, 68:987–1035.
1638. Wachter, J. and Warusawitharana, M. (2009) Predictable returns and asset allocation: Should
a skeptical investor time the market? Journal of Econometrics, 148:162–178.
1639. Wachter, J. and Yogo, M. (2010) Why do household portfolio shares rise in wealth? Review
1640. Wahal, S. (1997) Entry, exit, market makers, and the bid-ask spread. Review of Financial
Studies, 10:871–901.
1641. Wang, A. (2001) Overconfidence, investor sentiment and evolution. Journal of Financial
1642. Wang, F. (1998) Strategic trading, asymmetric information and heterogeneous beliefs.
1643. Wang, J. (1993) A model of intertemporal asset prices under asymmetric information.
1644. Wang, J. (1994) A model of competitive stock trading volume. Journal of Political Economy,
102:127–168.
1645. Wang, S. (1993) The local recoverability of risk aversion and intertemporal substitution.
1646. Watanabe, M. (2008) Price volatility and investor behavior in an overlapping generations
model with information asymmetry. Journal of Finance, 63:229–272.
1647. Weil, P. (1987) Confidence and the real value of money in overlapping generation models.
1648. Weil, P. (1989) The equity premium puzzle and the risk-free rate puzzle. Journal of Monetary
1649. Weil, P. (1990) On the possibility of price decreasing bubbles. Econometrica, 58:1467–1474.
1650. Weil, P. (1992) Equilibrium asset prices with undiversifiable labor income risk. Journal of
References 827
1651. Weitzman, M. (2007) Subjective expectations and asset-return puzzles. The American
1652. Werner, I. and Kleidon, A. (1996) U.K. and U.S. trading of British cross-listed stocks: an
intraday analysis of market integration. Review of Financial Studies, 9:619–664.
1653. Werner, J. (1985) Equilibrium in economies with incomplete financial markets. Journal of
1654. Werner, J. (2014) Rational asset pricing bubbles and debt constraints. Journal of Mathemat-
ical Economics, 53:145–152.
1655. West, D. (1987) A specification test for speculative bubbles. Quarterly Journal of Eco-
nomics, 102:553–580.
1656. West, D. (1988) Bubbles, fads, and stock price volatility tests: a partial evaluation. Journal
of Finance, 43:639–656.
1657. West, D. (1988) Dividend innovations and stock price volatility. Econometrica, 56:37–61.
1658. Weston, J. (2000) Competition on the NASDAQ and the impact of recent market reforms.
1659. Wheatley, S. (1988) Some tests of the consumption-based asset pricing model. Journal of
1660. Willard, G. and Dybvig, P. (1999) Empty promises and arbitrage. Review of Financial
Studies, 12:807–834.
1661. Wilson, R. (1968) The theory of syndacates. Econometrica, 36:119–132.
1662. Womack, K. (1996) Do brokerage analysts’ recommendations have investment value?
1663. Yan, H. (2008) Natural selection in financial markets: Does it work? Management Science,
54:1935–1950.
1664. Yao, R. and Zhang, H. (2005) Optimal consumption and portfolio choices with risky housing
and borrowing constraints. Review of Financial Studies, 18:197–239.
1665. Yogo, M. (2006) A consumption-based explanation of expected stock returns. Journal of
Finance, 61:539–580.
1666. Yogo, M. (2008) Asset prices under habit formation and reference-dependent preferences.
Journal of Business and Economic Statistics, 26:131–143.
1667. Yogo, M. (2016) Portfolio choice in retirement: health risk and the demand for annuities,
housing, and risky assets. Journal of Monetary Economics, 80:17–34.
1668. Yu, J. (2011) Disagreement and return predictability of stock portfolios. Journal of Financial
1669. Xia, Y. (2001) Learning about predictability: the effects of parameter uncertainty on dynamic
asset allocation. Journal of Finance, 56:205–246.
1670. Xiong, W. (2001) Convergence trading with wealth effects: an amplification mechanism in
financial markets. Journal of Financial Economics, 62:247–292.
1671. Xiong, W. and Hongjun, Y. (2010) Heterogeneous expectations and bond markets. Review
1672. Zapatero, F. (1998) Effects of financial innovations on market volatility when beliefs are
heterogeneous. Journal of Economic Dynamics and Control, 22:597–626.
1673. Zarowin, P. (1989) Does the stock market overreact to corporate earnings information?
1674. Zarowin, P. (1990) Size, seasonality, and stock market overreaction. Journal of Financial
1675. Zeldes, S. (1989) Consumption and liquidity constraints: an empirical investigation. Journal
1676. Zeldes, S. (1989) Optimal consumption with stochastic income: deviations from certainty
equivalence. Quarterly Journal of Economics, 104:275–297.
1677. Zhang, F. (2006) Information uncertainty and stock returns. Journal of Finance, 61:105–137.
1678. Zhang, H. (1997) Endogenous borrowing constraints with incomplete markets. Journal of
Finance, 52:2187–2209.
828 References
1679. Zhang, H. (2000) Explaining bond returns in heterogeneous agent models: The importance
of higher-order moments. Journal of Economic Dynamics and Control, 24:1381–1404.
1680. Zhang, L. (2005) The value premium. Journal of Finance, 60:67–103.
1681. Zhou, C. (1998) Dynamic portfolio choice and asset pricing with differential information.
1682. Zhou, C. (1999) Informational asymmetry and market imperfections: another solution to the
equity premium puzzle. Journal of Financial and Quantitative Analysis, 34:445–464.
Index
-algebra, 256 asset return, 56

asymmetric information, 398
auction market, 594, 638
absolute risk aversion, 28
actuarially fair, 24 background risk, 108, 112, 114
adapted stochastic process, 258 backward recursion, 264, 266
adverse selection, 398, 422, 433, 447, 450, bank account, 56
594, 601, 605, 607, 617, 619, 637, batch auction market, 594
638 Bayes rule, 258
aggregate demand, 7 behavioral finance, 358, 360, 370, 372,
aggregate risk, 129 520–541
aggregation property, 12, 157, 159, 191, 291 Bellman equation, 266
Allais paradox, 48, 489 biased self-attribution, 529, 531
ambiguity aversion, 486 bid-ask spread, 560, 561, 607, 616, 619, 625,
ambiguity premium, 488 628, 638, 641
anonymity of trading, 640 information ambiguity, 485
anticipation, 500, 567 bifurcation index, 287
anxiety, 500 binomial model, 183
arbitrage opportunity, 148, 165, 175, 184, 209, multi-period, 308
231, 293, 304 blackout period, 632
asymptotic, 234 Blackwell effect, 407, 421, 422
of the first kind, 165, 293 bond, 183
of the second kind, 165, 293 Borch condition, 131, 136
arbitrage pricing theory, 230–245, 319, 381, borrowing constraints, 562
572 bounded rationality, 541–548
empirical analysis, 241 breadth of ownership, 454
book-market value ratio, 244 bubble, 330, 336, 468, 538
equilibrium models, 249 overconfidence, 533
macroeconomic factors, 244 budget constraint, 4
Shanken critique, 241 buffer stock behavior, 556, 561
size, 244 business cycle, 375
Arrow security, 102, 143, 150, 282
Arrow-Debreu equilibrium, 139
artificial economy, 415 Call option, 159
asset allocation puzzle, 556, 572 capacity, 484

Springer Finance, DOI 10.1007/978-1-4471-7322-9
830 Index
capital asset pricing model (CAPM), 206, 209, implicit collusion, 618
213–230, 319, 381, 505 market, 607, 638
conditional, 229 monopolist, 607
CAPM anomalies, 223, 366, 498, 519, 532 preferencing arrangements, 618
book-to-market value ratio, 223 Delta hedging strategy, 186
january effect, 223 demand schedule, 586
leverage, 223 derivative pricing, 309
liquidity, 224 destabilizing speculation, 540
mean reversion and momentum effect, 224 difference model, 512
price earnings ratio, 223 difference of opinions, 401, 451–457, 461,
size, 223 538, 551
CARA utility, 108 diffusion of information, 469
CARA,DARA,IARA utility, 28 disappointment, 500, 567
central dominance, 114 disaster event, 387
centralized market, 641 discount factor, 7
certainty effect, 489 disposition effect, 494, 498
certainty equivalent, 25 distorted beliefs, 541, 548, 551
certainty line, 104, 129 diversification principle, 43, 69, 71, 105, 203,
Choquet expected utility, 483 219, 239
circuit breakers, 642 downstairs-upstairs markets, 637
classical asset pricing theory, 347, 362 dual capacity, 647
co-monotonicity, 132, 282 durable goods, 510
coefficient of absolute risk aversion, 28 dynamic programming principle, 264, 266
coefficient of relative risk aversion, 28, 61
cointegration, 350
common knowledge, 124, 397, 403 Edgeworth box, 9, 128
comonotonic independence, 484 efficient market hypothesis, 362, 364
competitive equilibrium, 7 elasticity of intertemporal substitution of
completeness assumption, 3, 19 consumption, 207, 490, 507, 513
complex security, 146, 151 elementary event, 15
concentration/fragmentation of markets, 636 elliptical distributions, 50
conditional capital asset pricing model, 229 Ellsberg paradox, 482
conservatism, 549 endowment process, 280
consumption capital asset pricing model envelope condition, 268, 503
(CCAPM), 202–213, 312, 380, 505 equilibrium, 174
consumption complementarity, 511, 514 existence, 8
consumption smoothing, 111, 275 uniqueness, 14
consumption smoothing puzzle, 383, 555 equilibrium price functional, 412
contingent claim, 301, 309 equity premium puzzle, 322, 383
contingent good, 123, 139 disappointment, 567
continuity assumption, 3, 19 information ambiguity, 487
contract curve, 9, 128 labor risk, 558
contrarian strategy, 366, 369, 371, 373, 374, limited participation, 564
389 loss aversion, 498
convergence traders, 570 myopic loss aversion, 494
correlation effect, 556 recursive preferences, 507, 567
cross listing, 636 adverse selection, 447
cross-covariance effect, 368 borrowing constraints, 562
cum-dividend price, 259 bounded rationality, 548
difference of opinions, 457, 548, 551
disappointment, 500
DARA utility, 31, 61, 63, 108, 112 distorted beliefs, 551
dealer external habit, 519
competition, 618 habit persistence, 513
Index 831
hyperbolic discount, 520 first welfare theorem, 9

incomplete markets, 555 forward interest rate, 320
incomplete observation, 572 fragmented market, 641
information, 448 framing effect, 491
keeping up with the Joneses, 515 free riding behavior, 442
noise traders, 527 frequentist approach, 481
pessimism, 384, 551 Friedman’s conjecture, 537
restricted participation, 564 fundamental solution, 350, 360
short sale constraints, 562 fundamental theorem of asset pricing,
transaction costs, 561 164–183, 192, 292, 296
estimation risk, 571 bid-ask spread, 560
Euler condition, 268, 378, 503, 512 second, 174, 300
European Call option, 309 fundamental value, 326
event, 15, 16 fundamentalist approach, 350
event studies, 377 future market, 123, 125, 138, 141
event tree, 256
ex-dividend price, 259
ex-post rational price, 351, 352 gamble, 17
excess volatility, 349–362 general equilibrium theory, 7
agents heterogeneity, 539 generalized least squares, 226
information ambiguity, 566 global absolute risk aversion coefficient, 206
liquidity constraints, 573 Green-Lucas equilibrium, 399, 411–414, 587
recursive preferences, 507 asymmetric information, 431
adverse selection, 447 costly information, 434
behavioral finance, 360 existence, 462
bounded rationality, 543 fully revealing, 414, 416–427, 463, 594
bubbles, 351 heterogeneous information, 423, 427
classical asset pricing theory, 358 linear, 432
difference of opinions, 451 multi-period models, 445
dividend stationarity, 356 noise traders, 428
external habit, 519 partially revealing, 414, 420, 427, 447
feedback traders, 537 Grossman-Stiglitz paradox, 434, 442, 594
hyperbolic discount, 520 growth stock, 224, 227
noise traders, 527, 569
nuisance parameters, 355
overconfidence, 528 habit formation, 380, 510, 567
risk aversion, 358 habit persistence, 510
time varying returns, 360 Hansen-Jagannathan bound, 204, 314, 379
expectation hypothesis, 321 HARA utility function, 33
expected term premium, 322 herding behavior, 447
expected utility theory, 19–24, 480–520 heterogeneous agents, 520–541
experimental economics, 458, 641 heterogeneous beliefs, 219, 245, 399, 403, 646
exponential utility function, 12, 32, 70, 100 heterogeneous information, 398
external consistency, 7, 123 heterogeneous opinions, 401
external habit, 517 heteroskedasticity, 363
hierarchical information, 445
high volume puzzle, 459
factor analysis, 243 higher order beliefs, 445, 467
factor model, 231 Hirshleifer effect, 421, 422, 464, 635, 639
approximate, 239 home bias puzzle, 332, 466
feasibility constraint, 281 information ambiguity, 488
feedback traders, 534–537, 540, 570 homogeneous beliefs, 257
filtration, 256 horizon effect, 494, 556, 560, 572
financial innovation, 191 information ambiguity, 488
832 Index
household portfolios, 331, 556, 562, 564 investor sentiment, 462

human capital, 247, 567 irrationality, 361, 366, 372
hyperbolic discount, 519
hyperbolic utility functions, 100, 138, 159,
291, 463 January effect, 365, 371, 373, 389
Jensen’s inequality, 25
IARA utility, 63, 108

imperfect competition, 583, 587, 606 keeping up with the Joneses, 515
Inada conditions, 5, 270 Kyle model, 595, 619
income effect, 62
incomplete information, 571
incomplete markets, 159, 552–566
labor income, 556
incomplete observation, 571
law of one price, 165, 293
independence assumption, 20, 49, 483
lead-lag effect, 390
independence axiom, 489
learning, 541, 550, 623
indexed debt, 486
learning risk, 571
indifference curves, 4, 41, 104
leisure, 567
indifference sets, 4
leveraged firm, 187
indirect utility function, 5
life cycle model, 259–279
infinite regression problem, 445
asset allocation puzzle, 331
infinite time horizon, 325
durable goods, 334
infinitely lived economy, 325
empirical evidence, 331
information acquisition, 442, 594
health risk, 334
information aggregation, 398, 458
horizon effect, 331
imperfectly competitive market, 585
housing, 334
perfectly competitive market, 416
illiquid assets, 334
information ambiguity, 482–488, 646
labor income, 334
information disclosure, 467, 623, 632, 639
luxury goods, 334
information effect, 427
mean reversion, 332
information flow, 256
participation puzzle, 331
information risk, 462
stochastic volatility, 333
information transmission, 398, 412
life-cycle-permanent income hypothesis,
informational consistency, 126, 142
270
informational efficiency, 398, 415, 615
likelihood ratio, 170, 182, 209, 307
informational trading, 450
limit order, 596, 602, 606
insider trading, 398, 595, 609, 624, 629–633,
limit order book, 606, 646
647
linear risk tolerance, 137, 463
empirical analysis, 631
liquidity, 573, 597, 641
regulation, 631
liquidity traders, 428, 609, 626
insurance demand, 102–110
logarithmic utility function, 12, 33, 272
insurance principle, 45, 68, 71, 203, 219
long term bond, 333
internal consistency, 6, 123
long-lived security, 259, 285
internal habit, 568
loss aversion, 49, 490, 491, 566
internet bubble, 336
lottery, 16
intertemporal capital asset pricing model
(ICAPM), 319, 380
intertemporal consumption, 7, 110, 149
intertemporal hedging, 275, 279 margin requirement, 573, 635
intertemporal rate of substitution of marginal rate of substitution, 5, 104, 203, 311
consumption, 312, 502 market abuse, 629
intraday patterns, 625 market breakdown, 450, 464, 594, 601, 605,
intraweek patterns, 625, 628 617, 638, 640, 642
Index 833
market completeness, 139, 143, 146, 150, 152, mutual fund, 390, 467
155, 174, 282, 300 mutual fund separation, 96–100
Pareto optimality, 552 monetary, 97
dynamic, 286, 287 normal distribution, 98
effective, 162, 163, 291 two, 74, 76, 97
static, 283 utility functions, 100
market crash, 387, 442, 451, 457 mutuality principle, 131, 132, 190
market depth, 433, 589, 597 myopic behavior, 273, 276, 279
market efficiency, 364, 398 myopic loss aversion, 494
semi-strong form, 402
strong form, 402
weak form, 402 naive forecast, 356
market for information, 447, 594, 645 Nash equilibrium in trading strategies, 588
market frictions, 551 no arbitrage fundamental equation, 325
market imperfections, 560–563 no-trade equilibrium, 10, 155, 281, 290
market incompleteness, 159 no-trade theorem, 409, 451, 464
market irrationality, 228 noise traders, 428, 432, 522–528, 569, 586,
market maker 595, 597
competition, 596 nominal asset, 138, 191
monopolist, 605 non-additive probability measure, 485, 566
market manipulation, 623, 633, 648 non-informational trading, 450
market order, 596, 606 non-redundant assets, 146
market participation, 564 non-synchronous trading, 369
market portfolio, 162, 205, 221, 230, 313 numéraire asset, 138, 146
market risk, 219 numéraire portfolio, 274
market selection, 521, 537, 538
market timing, 540
market transparency, 639 objectivist approach, 480
market value, 147, 166, 293, 297 optimal portfolio problem, 57
market viability, 176 optimal saving and consumption, 110
markets for information, 443, 467 order-driven market, 594, 638
Markov chain, 322 overconfidence, 456, 528–534
Markov process, 276 overreaction, 228, 366, 372, 373, 537
martingale approach, 303, 331 liquidity constraints, 573
martingale process, 270, 274, 295, 614 bounded rationality, 546
maxmin expected utility, 483 distorted beliefs, 550
mean effect, 556 noise traders, 569
mean reversion, 224, 366, 371, 449, 528 overconfidence, 528
mean-preserving spread, 407
mean-variance analysis, 38–47
mean-variance portfolio selection, 76 Pareto optimality, 8–12, 281, 289, 411
mean-variance preferences, 39 constrained, 149
measurability, 256 ex-ante, 127
mental accounting, 499 ex-post, 127
mimicking portfolio, 236, 249 information ambiguity, 486
minimum correlation portfolio, 205, 314 sharing rule, 136
minimum variance portfolio, 81 two-period , 127–138
model misspecification premium, 488 constrained, 161
model uncertainty, 488, 571 empirical analysis, 557
Modigliani-Miller theorem, 187 ex-ante, 140
momentum effect, 224, 368, 500 interim, 408
overconfidence, 528 participation puzzle, 564
momentum strategy, 369, 371, 375, 389, 460 information ambiguity, 488
multi-factor models, 318 perfect diversification, 47, 69
834 Index
perfect foresight, 142, 284, 351 random walk, 270, 349, 356, 363
perfect foresight price, 352 rank dependent expected utility, 490, 566
pessimism, 534, 550, 573 rational bubble, 325, 326, 349, 350
Ponzi scheme, 330, 334 rational expectations, 124, 125, 141, 190, 284,
pooling equilibrium, 617 400, 541
portfolio, 55 rationality assumption, 3, 19
portfolio choices, 556, 560, 562, 572 real asset, 138
loss aversion, 494 recursive preferences, 501–509, 567
multi-period, 259 reflexivity assumption, 3, 19
portfolio frontier, 77–96, 213 regular economy, 190
borrowing constraints, 94 relative risk aversion, 28
efficient, 84 replicating portfolio, 146
risk free asset, 88 representative agent, 10, 133, 155, 182, 202,
risky assets, 77 281, 289, 307, 310, 359
transaction costs, 94 representativeness heuristic, 549
two risky assets, 88 resolution of uncertainty, 502
portfolio optimization, 55–76 return predictability, 347, 362–377, 646
portfolio problem agents heterogeneity, 539
multi-period, 262 liquidity constraints, 573
portfolio turnpike, 334 loss aversion, 498
power utility function, 12, 32, 207, 274, 276, recursive preferences, 507
316 adverse selection, 447
precautionary premium, 109 bounded rationality, 546
precautionary saving, 112, 555, 561 distorted beliefs, 551
predictable stochastic process, 260 dividend yield, 376
preference relation, 2 external habit, 519
price continuity rule, 644 feedback traders, 537
pricing functional, 166, 297 macroeconomic variables, 376
pricing kernel, 181, 307 noise traders, 527, 569
principal component analysis, 243 overconfidence, 528
private equity, 332 return serial correlation, 363, 366
private information, 398 Riesz representation, 212, 237
probability measure, 15, 16 risk, 16, 480
probability space, 15 risk adjusted discount factor, 212, 219, 233
proper risk aversion, 50 risk aversion, 24–33, 49, 59, 60
prospect stochastic dominance, 492 first order, 509
prospect theory, 490, 491 risk factor, 231
cumulative, 494 risk free rate, 56
efficient portfolio, 493 risk free rate puzzle, 383, 385
prospect utility functional, 491 loss aversion, 498
prudence, 109, 111 recursive preferences, 507, 567
psychology, 569 adverse selection, 447
borrowing constraints, 562
bounded rationality, 548
quadratic utility function, 32, 40, 70, 209, 216, difference of opinions, 457
315 distorted beliefs, 551
quasi-complete economy, 463 external habit, 519
quasi-homothetic preferences, 12 habit persistence, 513
quasi-hyperbolic discount, 519 incomplete markets, 555
quote-driven market, 607, 638 incomplete observation, 572
keeping up with the Joneses, 515
restricted participation, 565
Radner equilibrium, 141–149, 284, 286 short sale constraints, 562
random variable, 16, 256 transaction costs, 561
Index 835
risk neutral probability measure, 168, 179, 209, substitution effect, 62

294, 347 sufficient statistic, 419, 425
risk neutral valuation, 169, 301 sunshine trading, 640
risk premium, 24, 58, 59, 65, 203 sunspot equilibrium, 447, 468
risk sharing, 190, 282, 421, 630 super-replication price, 171
autarchy, 566 sure thing principle, 23
home bias, 557 survivorship bias, 225, 226, 385, 390
risk tolerance, 28 symmetric equilibrium, 432
riskless arbitrage opportunity, 165 linear, 589
robust-risk sensitive control, 488 systematic risk, 203
Roll’s critique, 222, 223, 225, 241
tâtonnement process, 13
Saint Petersburg paradox, 17 tangent portfolio, 90
saving, 7, 112 tax heterogeneity, 468
seasonality effects, 365 taxation, 365, 468, 573
second order approximation, 40 technical analysis, 369, 445, 466
second welfare theorem, 9 temporal resolution of uncertainty, 288
security analysts, 373, 390 term premium puzzle, 387, 391
security market line, 215, 246 term structure of interest rates, 320
selection bias problem, 222 tick size, 648
self control preferences, 520 time additive utility, 24
self-serving attribution bias, 534 time invariant utility, 24
separating equilibrium, 617 time preferences, 519
separating hyperplane theorem, 167 time varying returns, 347, 360, 373
sharing rule, 134, 162 trade size, 597, 617, 624
Sharpe ratio, 93, 204, 209, 314 trading halt, 642
short sales constraint, 446, 452, 560, 572, 573 trading volume, 459, 622, 624, 625, 628
social welfare function, 10 agents heterogeneity, 539
speculation, 446, 455, 540 loss aversion, 500
speculative bubble, 350, 446, 453 transaction costs, 560
spherical symmetry, 50 adverse selection, 448
spot market, 123, 125, 138, 141 difference of opinions, 451
standard risk aversion, 50 overconfidence, 528
state index portfolio, 159 trading-consumption strategy, 260
state of nature, 15 proportions of wealth, 271
state of the world, 15 self-financing, 260
state price, 154, 166, 167, 300 transaction costs, 369, 560, 572, 636
state space, 104 transaction tax, 635
Stein’s lemma, 68, 206, 313 transitivity assumption, 3, 19
stochastic discount factor, 181, 204, 307, 505 transversality condition, 327
one-period, 313, 518 two funds separation, 163
stochastic dominance, 34–38 two mutual funds separation, 138, 215
first order, 34 two pass approach, 220
second order, 35, 77
second order monotonic, 38
stochastic interest rate, 312, 332 uncertainty, 16, 482
stochastic process, 258 uncertainty aversion, 483, 485, 566
stock, 183 underreaction, 373, 377
style investing, 540, 570 loss aversion, 500
sub-replication price, 171 bounded rationality, 548
subjective probability, 17, 49, 482 distorted beliefs, 550
substantial rationality, 4, 520 overconfidence, 528
836 Index
uniqueness of representation property, 146 volatility, 363, 449, 459, 465, 625
unleveraged firm, 187 volume, 465
utility function, 3
Uzawa utility, 567
wealth process, 260
weekend effect, 365
welfare theorems, 9
value function, 264, 503 white noise, 367
value of information, 404, 462 window dressing, 365
value process, 264
value stock, 224
variance effect, 556 zero-ˇ capital asset pricing model, 216, 221
variance ratio test, 364 zero-correlation portfolio, 84
variance uncertainty, 450 zero-coupon bond, 320

Financial Markets Theory-Barucci

Uploaded by

Copyright:

Available Formats

Financial Markets Theory-Barucci

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Financial Markets Theory-Barucci

Uploaded by

Copyright:

Available Formats

Springer Finance

This subseries of Springer Finance consists of graduate textbooks.

More information about this series at http://www.springer.com/series/3674

Financial Markets Theory

ISSN 1616-0533 ISSN 2195-0687 (electronic)

Library of Congress Control Number: 2017940989

© Springer-Verlag London Ltd. 2003, 2017

Printed on acid-free paper

This Springer imprint is published by Springer Nature

Milano, Italy Emilio Barucci

Firenze/Pisa, Italy Emilio Barucci

5.3 Empirical Tests of the CAPM . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 220

10.4 Multi-Period Models of Market Microstructure . . . . . . . . . . . . . . . . . . . . 619

All theory depends on assumptions which are not quite true.

Once it has achieved the status of paradigm, a scientific theory

In this chapter, we provide a brief overview of the fundamental results on pure

© Springer-Verlag London Ltd. 2017 1

1.1 Choices Under Certainty

Let us consider an agent who makes choices under certainty. In an environment

According to the above assumption, an agent is rational if he is able to rank any

x R y ” u.x/ u.y/; for all x; y 2 X:

max u.x/: (MP0)

L.x; / D u.x/ C .w p> x/;

uxk .x / D pk ; for all k D 1; : : : ; LI

lim u0 .x/ D 0 and lim u0 .x/ D C1:

thus leading to the budget constraint

u.c0 / C ı u.c1 /; (1.4)

where ı 2 .0; 1/ represents a (subjective) discount factor, accounting for the

In particular, assuming that the function u is strictly concave, this optimality

c0 < c1 ” rf ı > 1;

1.2 General Equilibrium Theory

In a perfectly competitive economy, the market represents an anonymous

Equation (1.6) actually consists of L equations in L unknowns (prices), one for

Theorem 1.2 Let the preference relation R i be rational, continuous, P strictly

uixk .xi / pk

where xi denotes the equilibrium consumption bundle of agent i, for i D 1; : : : ; I.

1.3 Pareto Optimality and Aggregation

ui .xi / ui .xi /; for all i D 1; : : : ; I;

with strict inequality holding for some j 2 f1; : : : ; Ig.

The identification of Pareto optimal allocations as equilibrium allocations pro-

where the maximization is subject to the constraint

The coefficients ai 0, for i D 1; : : : ; I, represent the weights assigned to the utility

max u.x/; (1.10)

under the constraint . p /> .x e/ 0. Note that, for all k D 1; : : : ; L,

1.4 Notes and Further Readings

In economic theory, the focus on equilibrium analysis is due to several reasons.

Behaviour is substantively rational when it is appropriate to the

The optimal consumption problem introduced in the previous chapter is formulated

© Springer-Verlag London Ltd. 2017 15

P.E1 [ E2 [ : : : [ En /  P.E1 / C P.E2 / C : : : C P.En /;

with equality holding for disjoint events ( finite additivity).

We refer to Chung [448] for a classical account of probability theory.

Therefore, if an agent evaluates a random variable by taking the expectation of

2.1 Expected Utility Theory

xQ 1 R xQ 2 ” U.Qx1 / U.Qx2 /; (2.3)

where U./ is defined as in (2.2).

xQ D ˛ xQ C .1 ˛/Qx R ˛ xQ C .1 ˛/Qx 0 R ˛ xQ 0 C .1 ˛/Qx 0 D xQ 0 :

ˇ xQ C .1 ˇ/Qx P ˛ xQ C .1 ˛/Qx ” ˇ > ˛:

Indeed, by step 1 (for the strict preference relation P) we know that

P.E1 [ E2 [ : : : [ En / P.E1 / C P.E2 / C : : : C P.En /;

where U./ is defined as in (2.2).

and that, for any 2 .0; 1/,

xQ C .1 /.˛ xQ C .1 ˛/Qx / P ˛ xQ C .1 ˛/Qx :

To prove the ( implication, it is enough to choose D .ˇ ˛/=.1 ˛/, with

CEu .Qx/ D EŒQx u .Qx/:

Fig. 2.1 Expected utility and certainty equivalent, fx ; x C I 1=2; 1=2g

Since EŒQ D .1 /.w w0 / C .1 /.w0 w/ D 0, the gamble Q is actuarially

which can be rewritten as u .Qx/ 12 rua .x/ 2 , where we define

which can be rewritten as ur .Qx/ 12 rur .x/ 2 ; where

Note that EŒQ D 0. Then, by the definition of risk premium,

Since x; x0 2 R and 2 Œ0; 1 are arbitrary, this proves the concavity of g. t

u .x C Q / D ua .x C Q / ub .x C Q / D u .x0 C Q /;

xQ 01 WD inffy 2 Œ0; 1 W F1 .y/ uQ g and xQ 02 WD inffy 2 Œ0; 1 W F2 .y/ uQ g:

thus showing that xQ 0i d xQ i , for i D 1; 2. The claim then follows by letting Q WD

U.Qx1 / D EŒu.Qx02 C Q / EŒu.Qx02 / D U.Qx2 /;

EŒQx1 EŒQx2 D U.Qx1 / U.Qx2 / 0:

EŒQx1 D 6; 2 .Qx1 / D 20 and EŒQx2 D 6; 2 .Qx2 / 22:43:

EŒu.Qx1 / D 10 and EŒu.Qx2 / 10:07:

for some function V W R RC ! R increasing with respect to the first argument

EŒQx1 D 3:625; 2 .Qx1 / D 5:422 and EŒQx2 D 3:5; 2 .Qx2 / D 7:125;

EŒu.Qx1 / 0:406 and EŒu.Qx2 / 0:433:

The representation EŒu.Qx/ D V.; 2 / follows by noting that C zQ d xQ . The