Princeton University - Economic Theory Center

Research Paper No. 050-2013

Speculation, Trading and Bubbles

April 4, 2013

Jose A. Scheinkman
Princeton University

Electronic copy available at: http://ssrn.com/abstract=2227701

 Speculation, Trading and Bubbles
Third Annual Arrow Lecture
José A. Scheinkman ∗
Princeton University and NBER

April 4, 2013

Contents
1 Introduction 2

2 An example: The South Sea bubble 5

3 Three stylized facts 7

3.1 Bubbles and trading volume . . . . . . . . . . . . . . . . . . . 7
3.2 Asset price bubbles implosion and increases in asset supply . 9
3.3 Asset price bubbles and the arrival of “new technologies” . . 11

4 Evidence for costly short selling and overconfidence 12

5 Sketch of a model 13
5.1 Limited capital . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2 Leverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.3 Origins of optimism . . . . . . . . . . . . . . . . . . . . . . . 18
5.4 Executive compensation, risk-taking and speculation . . . . . 19

6 Some additional evidence 20

∗
I wish to thank the Committee on Global Thought for the extraordinary honor of
delivering this Arrow Lecture. Kenneth Arrow was the towering researcher in economic
theory during the second half of the 20th century, and the treatment of uncertainty in
economics is one of the many subjects in which his influence is deeply felt. I also want
to thank Ken Arrow, Patrick Bolton, Sandy Grossman, Joe Stiglitz and members of the
audience for comments during the Lecture, Glen Weyl and Wei Xiong for comments on an
earlier draft, and Matthieu Gomez and Michael D. Sockin for excellent research assistance.

Electronic copy available at: http://ssrn.com/abstract=2227701

7 Some final observations 23

A Formal model 24
A.1 A Simple model . . . . . . . . . . . . . . . . . . . . . . . . . . 24
A.2 Limited capital . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1 Introduction
The history of financial markets is strewed with periods in which asset
prices seem to vastly exceed fundamentals - events commonly called bub-
bles. Nonetheless, there is very little agreement among economists on what
are the economic forces that generate such occurrences. Numerous academic
papers and books have been written explaining why the prices attained in a
particular episode can be justified by economic actors rationally discounting
future streams of payoffs. Some proponents of efficient-markets even deny
that one can attach any meaning to bubbles.1
Part of the difficulty stems from the fact that economists’ discussions of
bubbles often concentrate solely on the behavior of asset prices. The most
common definition of a bubble is “a period in which prices exceed funda-
mental valuation”. Valuation however depends on a view on fundamentals
and efficient market advocates correctly point out that valuations are al-
most always ex post wrong. In addition, bubbles are frequently associated
with periods of technological or financial innovations that are of uncertain
value at the time of the bubble, making it possible, although often unrea-
sonable, to argue that buyers were paying a price that corresponded to a
fair valuation of future dividends, given the information at their disposal.
In this lecture I adopt the alternative approach of starting with a more
precise model of asset prices that allows for divergence between asset prices
and fundamental valuation and that has additional implications that are
easier to test empirically. The model is based on the presence of fluctuating
heterogeneous beliefs among investors and the existence of an asymmetry
between the cost of acquiring an asset and the cost of shorting that same
asset. The two basic assumptions of the model - differences in beliefs and
higher costs of going short - are far from being standard in the literature
on asset pricing. For many types of assets, including stocks, there are good
economic reasons why investors should have more difficulty going short than
going long, but most economic models assume no asymmetry. The existence
1
e.g. Eugene Fama in Cassidy (2010) “I don’t even know what a bubble means. These
words have become popular. I don’t think they have any meaning.”

Electronic copy available at: http://ssrn.com/abstract=2227701

of differences in beliefs is thought to be obvious for the vast majority of mar-
ket practitioners, but economists have produced a myriad of results showing
that investors cannot agree to disagree. One implication of “cannot agree
to disagree” results is that differences in private information per se do not
generate security transactions, since agents learn from observing security
prices that adjust to reflect the information of all parties. Arrow (1986)
appropriately calls this implication “[a conclusion] flatly contrary to obser-
vation2 . Because they are not standard, I discuss in Section 4 below some
empirical evidence supporting these two central assumptions of the model.
Heterogeneous beliefs make it possible for the coexistence of optimists
and pessimists in a market. The cost asymmetry between going long and
going short an asset implies that optimists’ views are expressed more fully
than pessimists’ views in the market and thus even when opinions are on
average unbiased, prices are biased upwards. Finally, fluctuating beliefs
give even the most optimistic the hope that, in the future, an even more
optimistic buyer may appear. Thus a buyer would be willing to pay more
than the discounted value she attributes to an asset” future payoffs, because
the ownership of the asset gives her the option to resell the asset to a future
optimist.
The difference between what a buyer is willing to pay and her valuation
of the future payoffs of the asset - or equivalently the value of the resale
option - is identified as a bubble.3 An increase in the volatility of beliefs
increases the value of the resale option, thus increasing the divergence be-
tween asset prices and fundamental valuation, and also increases the volume
of trade. Hence, in the model, bubble episodes are associated with increases
in trading volume. As we argue in Section 3.1 below, the connection be-
tween high trading volume and bubbles is a well established stylized fact.
This relationship between bubbles and trading distinguishes models of bub-
bles based on heterogeneous beliefs and cost asymmetries from “rational
bubble” theories.4 A rational bubble is characterized by a continuous rise in
an asset’s price. Investors are content to hold the asset at the current price,
because they believe that they are compensated for any risk of the bubble
bursting by a suitable expected rate of price increase. In contrast to models
based on heterogeneous beliefs and costly short selling, rational bubble theo-
ries fail to explain the association between bubbles and high trading volume
2
page S398
3
A related definition of bubbles that has been used in the literature, is of episodes in
which buyers purchase an asset not on the basis of payoffs that the asset would generate,
but because she intends to resell it at a higher price in the future (Brunnermeier (2008)).
4
e.g. Santos and Woodford (1997)

3
and cannot be invoked to explain bubbles in assets that have a known value
at a certain future date T , such as many credit instruments.5
Market prices are determined at each point in time by the amount that
the marginal buyer is willing to pay for the asset. When beliefs are not
homogeneous, this marginal buyer is the least optimist investor that is still
a buyer of the asset. An increase in the capacity of individual investors to
buy the asset, perhaps through increased leverage, allows for more extreme
optimists to acquire the full supply of the asset at any point, and thus in-
creases the value of the resale option. In a similar manner, when investors
have limited capital and restricted access to leverage or limited capacity to
bear risk, an increase in the supply of the asset is accompanied by a less
optimistic marginal buyer. Thus the valuation that the marginal buyer has
of future payoffs declines as supply increases, because the marginal buyer
attributes a smaller fundamental value to the asset. However, a buyer also
knows today that because of the larger supply that needs to be absorbed,
future marginal buyers are likely to be relatively less optimistic and thus the
value of the resale option also declines. Hence an increase in the supply of
the asset that is unexpected by current holders of the asset diminishes the
difference between the price and the fundamental valuation of the marginal
buyer - that is it diminishes the size of the bubble. In Section 3.2 below I
argue that increases in asset supply helped implode some well known bub-
bles.
Robert Shiller’s influential Irrational Exuberance 6 postulates that bub-
bles result from feedback mechanisms in prices that amplify some initial
“precipitating factors”. The model in this lecture ignores the effect of this
endogenous price dynamics just as it ignores the learning from prices used
by rational theorists to dismiss the possibility of disagreement. It does how-
ever depend on precipitating factors that would generate optimism at least
among some investors. Asset price bubbles often coincide with (over) ex-
citement about a recent real or fake innovation7 and for the purpose of this
lecture one may think of “technological innovations”, broadly construed, as
the precipitating factors generating bubbles.
This lecture is organized as follows: In Section 2, I summarize some
relevant facts concerning the South Sea Bubble, one of the earliest well
documented occurrences of a bubble. In Section 3, I present some evidence
5
For if everyone agrees on the value at T, rational investors would refuse to pay at time
T − 1 any price above the discounted value at T. Thus there would be no bubble at time
T − 1. Repeating this reasoning one concludes that a bubble never arises.
6
Shiller (2006).
7
See Section 3.3 for some examples.

4
on the three stylized facts that inspire the model in this lecture - that asset
price bubbles coincide with increases in trading volume, that asset price
bubble deflation seem to match with increases in an asset’s supply and
that asset price bubbles often occur in times of financial or technological
innovation. In Section 4, I discuss some evidence for the assumption of costly
short selling and for the role of overconfidence in generating differences in
beliefs. Section 5 presents an informal sketch of the model and a discussion
of related issues such as the effect of leverage, the origin of optimism and the
role of corporations in sustaining bubbles. I summarize some empirical work
that provide evidence for the model in Section 6 and present some concluding
thoughts in Section 7. A formal model is exposited in the Appendix.

2 An example: The South Sea bubble

One of the earliest well documented occurrences of a bubble was the ex-
traordinary rise and fall of the prices of shares of the South Sea Company
and other similar joint-stock companies in Great Britain in 1720. At its
origins in 1710, the South Sea Company had been granted a monopoly to
trade with Spain’s South American colonies. However, during most of the
early 18th century Great Britain was at war against Spain’s Phillip V and
the South Sea Company never did much goods trading with South America,
although it did achieve limited success as a slave trader. The real business
of the South Sea Company was to exchange its stock for British government
debt. The new equity owners would receive a liquid share with the right to
perpetual annual interest payments in exchange for government debt that
paid a higher interest rate but was difficult to trade. In the first months of
1720, the Company and its rival the Bank of England engaged on a com-
petition for the right to acquire the debt of the British government. After
deliberating for more than two months, the House of Commons passed the
bill favoring the South Sea Company, a bill that was then “hurried through
all its stages with unexampled rapidity”8 and received royal assent on the
same day, April 7th, 1720, that it passed the House of Lords. The stock
of the Company that had traded for £120 in early January was now worth
more than £300. However this was just the beginning and share prices
approached £1,000 that summer.9
In Famous First Bubbles, Peter Garber argues that the prices attained
by the South Sea Company shares in the summer of 1720 were justified
8
Mackay (1932), page 51.
9
Neal (1990).

5
by the belief on “[John] Law’s prediction of a commercial expansion asso-
ciated with the accumulation of a fund of credit.”10 Garber’s monograph
deals mostly with the Dutch Tulipmania, and Garber presents no original
calculations on the South Sea bubble, but cites Scott (1910-12) who wrote
“[The] investor who in 1720 bought stock at 300 or even 400, may have
been unduly optimistic, but there was still a possibility that his confidence
would be rewarded in the future.” (pages 313-314). Scott is commenting on
prices of shares of the South Sea Company that prevailed until May 18th,
before share-prices doubled in a fortnight and continued to go up. In fact,
in a passage a few pages later, Scott writes that by August 11 “unless the
price of the stock in future issues had been set far above 1,000, the mar-
ket quotations were unjustifiable...Further, it would have been impossible
to have floated the surplus stock at 1,000, much less at an increased issue-
price. This must have been apparent to anyone, who considered the position
calmly.”11 This seems hardly an endorsement of the view that “[The South
Sea] episode is readily understandable as a case of speculators working on
the basis of the best economic analysis available and pushing prices along
with their changing view of market fundamentals”.12
The South Sea Bubble involved much more than the company that names
it. Other chartered companies holding British government debt such as
the Bank of England and the East India Company also experienced rapid
share-price appreciation, albeit in a less dramatic form than the South Sea
Company. In addition, numerous other joint-stock companies, nicknamed
“bubble companies”, were founded. Mackay (1932) list of 86 bubble com-
panies that were declared illegal by the “Bubble Act” of July 1720 is often
quoted, but Mackay is writing 120 years after the fact. A similar enumera-
tion of bubble companies is in Anderson (1787), pp. 104-112.13 Anderson’s
list gives the impression that many, though certainly not all, bubble schemes
were fraudulent.
The speculation mechanism that we propose in this lecture was well
understood by contemporary observers of the South Sea Bubble. The pio-
neering French-Irish economist Richard Cantillon, who was also a successful
banker and merchant, wrote to Lady Mary Herbert, on April 29, 1720 when
shares of the South Sea Company reached £400 “People are madder than
ever to run into the [South Sea Company] stock and don’t so much as pretend
10
Garber (1980), pages 121-122.
11
Scott (1910-12) page 323.
12
Garber (1980), page 122.
13
Anderson was a clerk at the South Sea Company during the bubble (Harris (1994),
page 615)

6
to go in to remain in the stock but sell out again to profit”.14 Similarly, in
his monumental history of British commerce, Anderson (1787) commented
on the initial buyers of bubble-companies stocks: “Yet many of those very
subscribers were far from believing those projects feasible: it was enough
for their purpose that there would very soon be a premium on the receipts
for those subscriptions; when they generally got rid of them in the crowded
alley to others more credulous than themselves”.15
By offering to replace illiquid British national debt by liquid shares,
The Lord Treasurer Robert Harley and the other founders of the South Sea
Company were pioneers of a “business model” that created value by allowing
investors to exercise the option to resell to a future optimist.

3 Three stylized facts

In this Section, I present some evidence on three stylized facts that inspire
my modeling choices: (i) asset price bubbles coincide with increases in trad-
ing volume (ii) asset price bubble implosions seem to coincide with increases
in an asset’s supply and (iii) asset price bubbles often coincide with financial
or technological innovation. The evidence presented here is not meant to
replace systematic empirical analysis - some of which we will discuss later -
but simply to motivate the modeling that follows. To bring these stylized
facts into focus, I will make references to aspects of four remarkable histor-
ical episodes of financial bubbles: The South Sea bubble, the extraordinary
rise of stock prices during the roaring twenties, the internet bubble, and
the recent credit bubble. I have already provided a short description of the
South Sea Bubble and will assume that readers are familiar with a basic
outline of the latter three episodes.

3.1 Bubbles and trading volume

Carlos et al. (2006) document that trading on Bank of England stock rose
from 2,000 transactions per year in 1717-1719 to 6,846 transactions in the
bubble year of 1720. They also estimate that 150% of the outstanding stocks
of the East India Company and of the Royal African Company turned over
in 1720.
Accounts of the stock-market boom 1928-1929 also emphasize overtrad-
ing. In fact, the annual turnover (value of shares traded as a percentage of
14
Cited in Murphy (1986), Chapter 9.
15
Anderson (1787), pages 102-103.

7
the value of outstanding shares) at NYSE climbed from 100% per annum
in 1925-27 to over 140% in 1928 and 1929.16 Daily share trading volume
reached new all time records 10 times in 1928 and 3 times in 1929. No
similar trading-volume record was set for nearly forty years, until April 1,
1968, when President Johnson announced he would not seek re-election.17
At the peak of the dotcom bubble, internet stocks had 3 times the
turnover of similar non-dotcom stocks.18 Lamont and Thaler (2003) study
six cases of spinoffs during that bubble - episodes when publicly traded com-
panies did an equity carve out, by selling a fraction of a subsidiary to the
market via an IPO, and announced a plan to spin off the remaining shares of
the subsidiary to the parent-company shareholders. A well-known example
was Palm and 3Com. Palm, which made hand-held personal organizers, was
owned by 3Com, which produced network systems and services. On March
2, 2000, 3Com sold 5% of its stake in Palm via an Initial Public Offering
(IPO). 3Com also announced that it would deliver the remaining shares of
Palm to 3Com shareholders before the end of that year. Lamont and Thaler
(2003) document that prior to the spin off, shares in these six carve-outs,
including Palm, sold for substantially more than the value of the shares
embedded in the original-company’s shares. Since shares of the parent com-
pany would necessarily sell for a non-negative price after the spin-off, the
observed relationship between the price of carve-outs and original-companies
shares indicates a violation of the law of one price, one of the fundamental
postulates of textbook finance theory. In addition, the trading volume of
the shares in the carve-outs was astonishing - the daily turnover in the 6
cases studied by Lamont and Thaler (2003) average 38%,19 a signal that
buyers of the carve-outs, just as the buyers of bubble-companies stocks in
1720, were looking for others more credulous than themselves.
It is frequently argued that excessive trading causes asset-prices to ex-
ceed fundamental valuations. We will not be making this argument here -
in our model excessive trading and prices that exceed fundamentals have a
common cause, but the often observed correlation between asset-price bub-
bles and high trading volume is one of the most intriguing pieces of empirical
evidence concerning bubbles and must be accounted in any theoretical at-
tempt to understand these speculative episodes.
16
Davis et al. (2005)
17
Hong and Stein (2007).
18
For instance, in February 2000, internet firms represented 6% of the public equity
market but 19% of the trading volume (Ofek and Richardson (2003))
19
This must be compared with an annual turnover of 100% for the typical NYSE stock
at that time.

8
3.2 Asset price bubbles implosion and increases in asset sup-
ply
The South Sea Bubble lasted less than a year, but in that short period there
was a huge increase in the supply of joint-stock companies shares. New
issues doubled the amount of shares outstanding of the South Sea Company
and more than tripled those of the Royal African Company. Numerous
other joint-stock companies were started during that year. The directors of
the South Sea Company seem to have understood that the increase in the
supply of shares of joint-stock companies threatened their own capacity to
sell stock at inflated prices. Harris (1994) thoroughly examined the wording
of the Bubble Act of 1720, in which Parliament banned joint-stock companies
not authorized by Royal Charter or the extension of corporate charters into
new ventures, and the historical evidence on interests and discourses, and
concluded that “the [Bubble Act] was a special-interest legislation for the
[South-Sea Company], which controlled its framing and its passage”. In
any case, the South-Sea Company directors used the Bubble Act to sue old
chartered companies that had moved into “financial” activities and were
competing with the South Sea Company for speculators’ capital.
As the dotcom bubble inflated there were numerous IPOs, but in each
of these only a fraction of the shares were effectively sold. The remaining
shares were assigned to insiders, venture capital funds, institutions, and so-
phisticated investors, who had agreements to hold their shares for a “lockup”
period, often 6 months. An extraordinary number of lock up expirations for
dotcom companies occurred during the first half of 2000, vastly increasing
the supply of shares.20 Venture capital firms that had distributed $3.9 bil-
lion to limited partners in the third quarter of 1999, distributed $21 billion
during the first quarter of 2000, either by giving the newly unlocked shares
to the limited partners or by selling these shares and distributing cash.21
The bursting of the bubble in early 2000 coincided with this dramatic in-
crease in the float (total number of shares available to the public) of firms
in the internet sector.
The recent credit bubble was characterized by an inordinate demand for
liquid “safe assets” usually displaying a AAA rating from one or more of the
major credit ratings agencies. Financial engineering and rosy assumptions
concerning housing-prices growth and correlations of defaults allowed issuers
to transform a large fraction of subprime mortgages22 into AAA credit.
20
Ofek and Richardson (2003)
21
Janeway (2012)
22
Including so called Alt-A mortgage loans.

9
Subprime mortgage loans were pooled to serve as collateral to a Mortgage
Backed Security (MBS), a collection of securities (tranches) that may have
different priorities on the cash flows generated by the collateral. The senior
tranche typically received a AAA rating. Lower rated tranches of MBSs in
turn could be pooled as collateral for a credit default obligations (CDO).
The senior tranches of the CDO would again have a AAA rating. Lower
rated tranches of CDOs could then be combined to serve as collateral for
the tranches of a CDO2 . . .
The high prices commanded by the instruments resulting from this secu-
ritization process increased the demand by issuers for residential mortgage
loans and lowered the cost of taking a mortgage thus facilitating housing pur-
chases. In 2000, issues of private label mortgage backed securities (PLS),
that is mortgage backed securities that were not issued by Government Spon-
sored Enterprises, financed $572 billion in US residential mortgages. By the
end of 2006, the volume of outstanding mortgages financed by PLSs had
reached $2.6 trillion. Many of these PLSs used less-than-prime mortgage
loans and the combined annual subprime and Alt-A origination grew from
an estimated $171 billion in 2002 to $877 billion in 2005, an annualized
growth rate of 72%.23
Several developments added dramatically to the effective supply of se-
curities backed by housing related assets. In the summer of 2005, the Inter-
national Swaps and Derivatives Association (ISDA) created a standardized
credit default swap (CDS), or insurance against default, for mortgage-backed
securities. These contracts allowed a pessimist to buy insurance on a sub-
prime MBS he did not own. Early in 2006, Markit launched ABX.HE,
subprime mortgage backed credit derivative indexes. Each ABX index was
based on 20 MBSs with the same credit rating and issued within a six-month
window. The level of the index reflected the price at which a CDS on this
set of MBSs was trading. Investors that had optimistic views concerning
the risks in subprime MBS could now acquire a short position in a AAA
series of the ABX index. If the market became more positive about these
securities in the future, the cost of the corresponding CDS would drop and
the shorts would make a profit. In the summer of 2006, ISDA went further
and created a standard CDS contract on CDO tranches allowing investors
who had a pessimistic view of say AAA tranches of subprimes to effectively
take short exposures to the subprime market - a market in which, for insti-
tutional reasons, it was often difficult to short individual securities. In this
way, the supply of AAA tranches of CDOs was effectively increased.
23
Thomas and Van Order (2010)

10
 None of these developments however were fully adequate to satisfy the
demand for AAA paper by institutions that, often for regulatory reasons,
found it necessary to buy highly rated securities. Synthetic CDOs were a
perfect supply response to this demand. These were CDOs that did not
contain any actual MBSs but instead consisted of a portfolio of CDSs on
MBSs and some high-quality liquid assets. The creation of a standard CDS
for MBSs, and the consequent increase in supply of these insurance con-
tracts, allowed Goldman Sachs, Deutsche Bank and other Wall Street pow-
erhouses, but also smaller firms such as Tricardia to create an enormous
supply of synthetic CDOs. Wall Street could now satisfy the demands of a
German Landesbank for additional US AAA mortgage bonds without any
new houses being built in Arizona.24 The associated increase in the supply
of assets carrying housing risk seems to have been enough to satisfy not
only optimistic German Landesbanks but also every Lehman trader or Citi
SIV portfolio manager that wanted to hold housing risk. In this way, the
implosion of the credit bubble parallels the implosion of the South Sea and
dotcom bubbles.25

3.3 Asset price bubbles and the arrival of “new technologies”

Asset price bubbles tend to appear in periods of excitement about innova-
tions. The stock market bubble of the 1920s was driven primarily by the
new technology stocks of the time, namely the automobile, aircraft, mo-
tion picture, and radio industries, and the dotcom bubble has an obvious
connection to internet technology.
In the US there has been notable attention to the recent housing bubble.
However the housing bubble was simply one manifestation of an enormous
credit bubble that took place in the early part of this century. In April 2006,
while the Case-Schiller housing index reached its peak, you could buy a 5
year CDS on Greek debt for less than 15 bp (.15%) per year.26 Similarly, in
24
Synthetic CDOs have been blamed for the inordinate damage created by the subprime
implosion (See e.g. Nocera (2010) for a non-technical indictment of synthetics), because
it allowed optimistic financial institutions to take even more subprime risks. However it
is not obvious what would have happened if synthetics had not existed. First, the price of
“safe” subprime based securities would have been higher, causing bigger losses per security
albeit on a smaller number of securities. Second, and more scary, is that we would have
ended up with an even larger number of unfinished houses in the Southwest.
25
Glaeser et al. (2006) argues that real estate bubbles are also deflated by increases in
housing supply.
26
The lowest spreads for a 5 year CDS on Greek debt, in the single digits, were reached
later, in January 2007.

11
April 2006, the average spread for a CDS on debt from Argentina, a country
that had defaulted repeatedly and as recently as 2002, was less than 3% per
year.
This credit bubble coincided with advances in financial engineering -
the introduction of new financial instruments and hedging techniques and
advances in risk measurement that promised better risk management and
“justified” lower risk-premia.

4 Evidence for costly short selling and overconfi-

dence
Economists typically treat short sales of an asset as the purchase of a neg-
ative amount of that asset, and assume that short-sales generate just as
much transaction costs as purchases. Although there are exceptions - such
as future markets - legal and institutional constraints make this assump-
tion problematic in almost all cases. To short an asset requires finding a
lender for that asset and, because often there are no organized market for
borrowing an asset, finding a lender can be difficult. In addition, securities
are often loaned on call and borrowers face the risk of replacing the bor-
rowed securities or being forced to cover their short position.27 Securities
loans are often collateralized with cash. The security lender pays interest
on the collateral, but the lender pays the borrower of the security a rebate
rate that is less than the market rate for cash funds. Rebate rates may be
negative and thus the fee effectively paid by the borrower of the security can
exceed market interest rates. Among other factors, the rebate rate reflects
the supply and demand for a particular securities’ loan and the likelihood
that the lender recalls the security. D’avolio (2002) documents that rebate
rates are negatively while recalls are positively correlated with measures of
divergence of opinions. The possibility of recall makes shorting securities
with a small float and/or little liquidity especially risky. Individual MBS
securities or certain tranches of CDOs had relatively small face values.
Diether et al. (2002) provide evidence that stocks with higher disper-
sion in analysts earnings forecasts earn lower future returns than otherwise
similar stocks. It is reasonable to take the dispersion in analysts forecasts
as a proxy for differences in opinion about a stock, and the observation of
lower returns for stocks with more difference in opinions is consistent with
the hypothesis that prices will reflect a relatively optimistic view whenever
27
In fact, U.S. regulations often require many institutional lenders to maintain the right
to terminate a stock loan at any time. (D’avolio (2002))

12
going long is cheaper than going short. In contrast, the evidence reported by
Diether et al. (2002) is inconsistent with a view that dispersion in analysts
forecasts proxies for risk, since in this case stocks with higher dispersion
should not exhibit lower returns.
There are of course many possible ways in which differences in beliefs
may arise. In this lecture I will assume that differences in beliefs are related
to overconfidence - the tendency of individuals to exaggerate the precision of
their knowledge. The original paper documenting overconfidence is Alpert
and Raiffa (1982). Since then overconfidence has been documented in a va-
riety of groups of decision makers, including engineers (Kidd (1970)) and
entrepreneurs (Cooper et al. (1988)). Tetlock (2005) discusses overconfi-
dence in a group of professional experts who earn a living commenting or
advising on political and economic trends, such as journalists, foreign policy
specialists, economists and intelligence analysts. The vast majority of these
pundits’ predictions seem to be no better than random chance.
Even more directly relevant to the topic of this lecture is the paper by
Ben-David et al. (2010). Between June 2001 and September 2010, Duke Uni-
versity collected quarterly surveys of senior finance executives, the majority
of whom were CFOs and financial vice-presidents. Among other questions,
the respondents were asked to give two numbers: The first was a number
that they believed there was only a one-in-ten chance that the actual S&P
annual return over the next year would be below that number. The second
was a number that they believed there was only a one-in-ten chance that the
actual S&P annual return over the next year would be above that number.
These two numbers form the 10-90 interval, that is the interval of numbers
for which a respondent believes there is a 10% chance that the actual S&P
returns would fall to the left of that interval and a 10% chance that the
actual returns would fall to the right of that interval. The 10-90 interval
should cover 80% of the realizations. In total, the surveys collected over
12,500 of these intervals and the realized returns in the S&P over each year
following a survey fell within the executives 10-90 intervals only 33% of the
time. Evidently, these senior finance executives grossly overestimated the
precision of their knowledge concerning future stock returns.

5 Sketch of a model
The appendix contains a model connecting difference of opinions and costly
shorting to speculation and trading. The model in the appendix is a sim-
plified version of an already stylized model developed in Scheinkman and

13
Xiong (2003), who in turn were inspired by a pioneering paper by Harrison
and Kreps (1978).
In the model in the appendix, there are two types of investors that for
simplicity are assumed to be risk neutral - that is, they are willing to pay for
an asset that they are forced to hold to maturity that asset’s (discounted)
expected payoff. Differences of opinions arise because investors estimate fu-
ture payoffs of a risky asset using signals they believe are useful to predict
payoffs. Some investors are “rational” and use signals in an optimal fash-
ion. Others attribute value to information they should ignore - perhaps a
cable-TV host named JC recommending a “buy” or a “sell.” In the model
“irrational” investors are on average right, but depending on the particular
value of the useless information that they observe, they can be excessively
optimistic or excessively pessimistic.28 Thus, on average, opinions of in-
vestors are unbiased. I also assume for simplicity that short sales are not
allowed, although it would suffice to assume costly short-selling.
Suppose that an asset will have a payoff two years from now which may
be high or low with equal probability. Suppose further that one year from
now, JC may voice an opinion on which of the two payoffs is likely to occur.
The TV host’s opinion is totally unfounded, but there is a large group of
investors that believe that JC’s views are valuable. Since there are no short
sales allowed, if each group of agents has more than enough capital to acquire
the whole float of the asset at their own valuation, then once JC’s opinion
is known, members of the most optimistic group would acquire the whole
supply and, because they compete with others of the same group, buyers
would end up paying their expected payoff. If JC claims the higher payoff
is likely to obtain, the irrational agents would pay a price that reflects an
optimistic view of the asset payoff. If JC claims the lower payoff is likely
to occur, then the irrational agents would be pessimists, but rational agents
would still be willing to buy the asset paying a price equal to the rational-
agents expected payoff. And if JC is silent, both agents agree that the asset
is worth the rational-agents’ expected payoff. Now suppose a market where
the asset is traded opens today. A rational investor knows that if a year
from now JC screams “high dividend”, she would have the option to sell
the asset at that moment to an irrational investor at a price higher than
her own valuation would be at that point. Otherwise, if JC stays silent or
utters a pessimistic opinion, the investor would be happy to hold the asset.
Thus a rational buyer would be willing to pay today in excess of her own
valuation of future payoffs, because she acquires an option to resell the asset
28
That is, if JC emits an opinion, it is equally likely to be a “buy” or a “sell”.

14
one year from now if JC screams “high dividend”. The more likely it is
that in one year from now JC would claim that a high dividend will obtain,
the larger would be the amount that a rational investor would pay for the
asset today. Because of the symmetry we assumed between the probability
that JC claims that a high payoff will occur and the probability that JC
claims that a low payoff will occur, the rational investor would pay more
for the resale option when there is a higher probability that JC would emit
any opinion. Similarly, an irrational investor would pay more than his own
valuation for the asset today, because he knows that if JC claims next year
that a low payoff will occur, he would be able to sell the asset to someone
that he would judge to be overoptimistic.
In the context of the model, I define a bubble as the value that a buyer
pays for this option to resell. Thus a bubble occurs when a buyer pays in
excess of her valuation of future dividends, because she values the opportu-
nity to resell to a more optimistic buyer in the future. Since buyers would
tend to be among the most optimistic agents, it would be natural to call the
difference between buyer’s valuation and a “rational” valuation also a bub-
ble. Here, I do not include buyers’ excessive optimism as part of the bubble,
and thus the definition of a bubble used in this lecture is somewhat conser-
vative. Although bubbles certainly coincide with periods in which excessive
optimism prevails among many investors, our definition emphasizes the role
of the existence of divergent opinions as opposed to the actual differences in
opinions that occur during these episodes.
If the ownership of the asset is equally distributed initially between ra-
tional and irrational agents, trades will occur whenever JC emits an opinion.
On average we would get a higher volume of trade whenever there is a larger
probability that JC would give an opinion. Thus the same cause - the fre-
quency of JC opinions - creates differences in opinion, a bubble and trading.
In the appendix we show that this difference in opinions can be identified
with overconfidence.
The value of the resale option is naturally a function of the costs of
funds. The higher the interest rate faced by investors, the less they are
willing to pay for the resale option. The model in the appendix thus gives
a simple theoretical justification for the argument that lower interest rates
are conducive to bubbles. In a similar manner, shorter horizons yield fewer
opportunities to resell, making the resale option less valuable.
The model in this lecture ignores two forces that have been invoked to
dismiss the importance of differences in beliefs. The first is learning - the
irrational agents should eventually learn that the signal they are using is
useless. Learning no doubt plays a role in diminishing differences in be-

15
liefs over long horizons, but bubbles last for a relatively short period when
learning must have a limited effect. The second argument brought against
the importance of irrational beliefs is survivorship. As argued by Fried-
man (1966), irrational agents should loose wealth on average and thus have
a vanishing influence on market outcomes. However,Yan (2008) performed
calibration exercises on Friedman’s argument and concluded that for reason-
able parameter values, it may take hundreds of years for irrational investors
to loose even half their wealth.Because bubbles are relatively short-lived, I
will ignore learning and survivorship and emphasize other forces that create
and deflate bubbles.

5.1 Limited capital

If irrational investors have limited access to capital and the supply of the
asset increases, perhaps as a result of sales by insiders, then even when JC
emits an optimistic opinion, irrational investors may not be able to buy
the full asset float while paying their own valuation. When the capital
constraint of irrational investors is severe enough, even when irrational opti-
mism occurs, the marginal buyer may be a rational investor who has a lower
valuation of the asset. Hence when irrational agents have limited capital,
the size of the bubble depends on the asset supply. For the same reason,
if the asset’s float is large enough, some of the asset supply may end up in
the hands of rational investors even though irrational investors are optimists
and have a higher valuation for the asset. As a consequence, the turnover
(volume traded as a fraction of the float) of an asset is smaller for assets
with a larger float.
In the model developed in the appendix, in the presence of capital con-
straints, an increase in supply that is not fully expected leads to a deflation
of the bubble. This was one of the main insights in Hong et al. (2006). In
the internet bubble, increases in supply were often the result of sales by
insiders. Hong et al. (2006) observed that it is reasonable to assume that
unexpected sales by insiders lead to a revision of forecasts by current in-
vestors and potential buyers and that this revision in beliefs must reinforce
the tendency of supply increases to produce bubble’s implosion.

5.2 Leverage
The model in the appendix does not explicitly treat leverage, but the ob-
servations on limited capital also provide insights on the role of leverage.
Investors can often access capital using their purchases of assets as collat-

16
eral for loans. The amount loaned to finance the purchase of one unit of a
risky asset would typically be less than the price of the asset. The difference
between the price of the asset and the value of the loan is the margin and its
reciprocal the leverage. A homeowner that acquired a house in 2004 with a
5% downpayment would have thus a leverage of 20. Higher leverage would
increase the access to capital by optimists and thus help to augment and
sustain bubbles.
In the presence of belief disagreements pessimists should be willing to
make loans collateralized by the risky-asset to optimists. Market conditions
determine the leverage and interest rates charged on these loans, but one
should expect that pessimists would demand relatively low leverage and/or
high interest rates. Pioneered by Geanokoplos,29 a literature which studies
the equilibrium determination of leverage and interest rates for loans from
pessimists to optimists has developed.
In reality however, because of tax or regulatory reasons, not all optimists
are adequate holders of certain risky assets. For instance, homeowners ben-
efit from the absence of taxation on imputed rent and the unique treatment
of capital gains in owner-occupied homes. Although they were not the most
appropriate direct investors in houses, optimistic banks had another way
to benefit from the housing price increases that they anticipated in 2004.
They could make loans charging more than prime rates to subprime buy-
ers that the banks believed would be capable of repaying their loans in the
very likely event that house prices continued to behave as they had in the
previous ten years. Contemporary analysts’ reports from major financial in-
stitutions recognized the potentially negative impact of house-prices decline
on the value of mortgage related securities, but underestimated the prob-
ability of occurrence of these adverse events and thus as argued by Foote
et al. (2012), it is reasonable to conclude that some investors in mortgage
related securities were simply excessively optimistic about the possibility of
house-price declines.30 Money market funds, which by their nature must
invest in short-term “safe” securities, participate heavily in repo markets -
essentially loans collateralized by securities. A money market fund that was
willing to finance 98.4% of the purchase-price of a AAA mortgage security
to an investor in 2006 probably thought that these securities were actually
nearly risk-free, warranting a leverage of 60. In this way, a chain of opti-
29
See Geanakoplos (2010) for a recent summary.
30
“the average coupon on subprime adjustable-rate mortgages was several hundred basis
points above the comparable prime loans. And yet, if investors think that house prices
can rise 11 percent per year, expected losses are minimal.” (Foote et al. (2012) pp 32,33)

17
mists provided leverage to optimistic investors in the housing market.31 This
chain was reinforced by US regulation that placed low capital-risk weights on
securities deemed AAA by “Nationally Recognized Statistical Rating Orga-
nizations” and by similar regulations in other countries, and was amplified
by innovations in finance, such as the MBS based CDO. It is ironic that
this same process of innovation in financial engineering eventually allowed
pessimists to express their negative views on these markets and speeded up
the implosion of the bubble.
Compared to pessimists, optimists without direct access to a risky asset
are bound to accept terms more favorable to the borrower on loans backed
by that asset. Thus it is reasonable to argue that during the credit bubble,
leverage from optimists was a more important source of capital for mortgage-
securities investors than leverage from pessimists.

5.3 Origins of optimism

The formal model exposited in the appendix is silent concerning the precipi-
tating factors that generate optimism among investors. In practice, investors
rely on advice from friends, acquaintances and “experts.” Some advice is
without doubt biased because of the financial incentives faced by experts,
but it has been documented that during speculative episodes, apparently
unbiased advisors also issue overoptimistic forecasts.32
Motivated by the coincidence of bubbles and periods of excitement about
new technologies Hong et al. (2008) proposed a theory based on the role of
formal or informal advisors. In the model in Hong et al. (2008), there are
two types of advisors. Tech-savvy advisors understand the new technologies
- think of a finance quant during the credit bubble - while old-fogies are
uniformly pessimistic concerning the new technologies. Tech-savvies may be
well intentioned advisors, but worry about being confused with old-foggies.
As the art critic in Tom Wolfe’s The Painted Word, tech-savvies worry that
“To be against what is new is not to be modern. Not to be modern is to
write yourself out of the scene. Not to be in the scene is to be nowhere”.33
To insure that their advisees do not confuse them with old-foggies, tech-
savvies issue over-optimistic forecasts concerning assets related to the new
31
Optimistic investors also obtained leverage from financial market participants that
understood the risks involved, but benefitted from skewed incentives.
32
During the dotcom period, so-called objective research firms with no investment bank-
ing business, such as Sanford and Bernstein, issued recommendations every bit as opti-
mistic as investment banks (e.g. Cowen et al. (2006)).
33
Wolfe (1975), page 84.

18
technologies. Rational investors understand the advisors motivations and
“debias” the advice, but naive investors take advisors recommendations at
face value. Although the presence of old-foggies tends to depress prices of
the assets related to the new technology, when there is a sufficient number
of naive investors guided by tech-savvies, the biased advice overcomes the
effect of old-foggies and induces over-optimism among investors.

5.4 Executive compensation, risk-taking and speculation

Although our discussion until now has been mainly concerned with the be-
havior of individual investors, corporations have played a central role in
recent bubbles. The implosion of the credit bubble and consequent Wall
Street bailout brought deserved attention to the risk-taking behavior of fi-
nancial firms during that episode and lead to calls for compensation reforms
that would eliminate excessive incentives for managers to take risks.
The standard economists’ approach to compensation uses the “Principal-
Agent” framework which emphasizes how managerial contracts are set by
boards as shareholders’ representatives to solve the misalignment of interests
between managers and stockholders. Bebchuk and Fried (2006) and other
critics of this approach contend that CEOs have been able to essentially set
their own contracts through captured boards and remuneration committees
and that major reforms in corporate governance to increase shareholder
power are necessary to remediate the current state of affairs.
The critics of the standard approach to compensation are no doubt cor-
rect in pointing out important ways in which the selection of board members
and executive pay negotiations depart from the idealized “arms-length” bar-
gaining of the Principal-Agent paradigm. However the critics have more dif-
ficulties explaining how the relatively recent phenomenon of rise in pay and
stock-based compensation has coincided with the dramatic rise in share-
holder influence that began in the 1980s.34 In fact, we have observed a
tendency towards greater board independence, a higher proportion of exter-
nally recruited CEOs, a decrease in the average tenure of CEOs, and higher
forced CEO turnover during this period.
Bolton et al. (2006) point out that a speculative market creates a di-
vergence between the interests of short-term versus long-term stockholders
and between the interests of current versus future stockholders. Short-term
stockholders would like managers to take actions that increase the specula-
tive value of shares, even if at the cost of the fundamental value of the firm.
34
See Holmström (2006) and Holmström and Kaplan (2001).

19
If stockholders with a short-term horizon dominate a board, they would se-
lect contracts for managers that emphasize stock-price based compensation
that vests early, to align the interests of managers with their own interests.35
Examining a panel of US financial firms during the period of 1992-2008,
Cheng et al. (2010) found substantial cross-firm differences in total exec-
utive compensation even after controlling for firm size. Top management
level of pay is positively correlated with price-based risk-taking measures
including firm beta, return volatility, the sensitivity of firm stock price to
the ABX subprime index, and tail cumulative return performance. Man-
agers’ compensation and firm risk-taking are not related to governance vari-
ables but covary with ownership by institutional investors who tend to have
short-termist preferences and the power to influence a firm’s management
policy.36
The empirical results in Cheng et al. (2010) indicate that governance
reforms are hardly the solution for excessive risk-taking by financial firms.

6 Some additional evidence

Two data sets, both coincidently from China, provide additional evidence
to support the mechanisms I proposed in this lecture. These data sets have
been used in research that was motivated by the bubble models discussed
in this lecture.
Between 1993 and 2000, 73 Chinese firms offered two classes of shares,
A and B, with identical rights. Until 2001, domestic Chinese investors could
only buy A shares while foreign investors could only hold B shares. Mei et al.
(2009) used these data to test implications of models of heterogenous beliefs
and short-sale constraints. This is particularly appropriate, because at that
time Chinese buyers of A shares faced very stringent short-sale constraints,
and IPOs and SEO’s (Seasoned Equity Offerings) were tightly controlled by
the central government.
Despite their identical rights to dividends and voting rights, A shares
traded on average at a premium of 420% relative to B shares. The annual
turnover of B shares, around 100%, was similar to the turnover of NYSE
35
Or as Lewis (2004) wrote: “The investor cares about short-term gains in stock prices
a lot more than he does about the long-term viability of a company. .... The investor, of
course, likes to think of himself as a force for honesty and transparency, but he has proved,
in recent years, that he prefers a lucrative lie to an expensive truth. And he’s very good
at letting corporate management know it”.
36
Froot et al. (1991) point out that the horizon of many institutional investors is around
1 year.

20
shares at the time, while A shares traded at 500% a year. The relatively
large panel of 73 stocks allows Mei et al. (2009) to control for cross-sectional
differences in risk and liquidity and time variations in China’s risk and risk-
premium. They find that A-share turnover is significantly positively corre-
lated with the A-B share premium, and in fact 20% of that premium can
be “explained” by turnover variation. On the other hand, B-share turnover
had a positive association with the A-B premium, albeit not statistically sig-
nificant. This supports the hypothesis that A-share prices (but not B-share
prices) were driven by speculation. Mei et al. (2009) also show that the A-
B share premium and A-share turnover increase with a firm’s idiosyncratic
return volatility, a proxy for uncertainty. This is consistent with the bubble
theory based on heterogeneous beliefs if one believes that more fundamental
uncertainty would increase fluctuations in heterogenous beliefs.
Furthermore, Mei et al. (2009) show that controlling for the proportion
of days in which a share did not display a price change - a proxy for liquidity
that has been used in the finance literature - does not significantly change the
association between A-share turnover and the A-B premium. To determine
whether trading in A and B shares was driven by speculation or liquidity,
they examined the cross-sectional correlation between share turnover and
asset float of A and B shares. Liquidity typically increases with asset float,
since as float increases, it is easier for buyers to match up with sellers. On
the other hand, as we argued above, in the presence of speculation and
limited capital, a larger float is associated with a smaller turnover. Mei
et al. (2009) finds a significant negative relationship between share turnover
and float in A shares markets in 1993-2000, suggesting that the large trading
volume in A shares was not a result of liquidity. However for B shares, which
were held by more sophisticated foreign investors, Mei et al. (2009) found
that turnover was positively associated with float - suggesting that liquidity
played a role in attracting trading in B shares.
On February 28, 2001, the Chinese government allowed domestic in-
vestors to buy B shares provided they used foreign currency. The A-B
premium decreased but almost exclusively because B share prices went up.
Monthly B-share turnover in the six months following this event averaged
44%, almost 4 times the monthly turnover of these shares in the 6 months
preceding the liberalization. Moreover, the coefficient of the A-B premium
on B-share turnover becomes significantly negative after the liberalization.
This contrasts with the results for the earlier period (positive and insignifi-
cant) and indicates that speculation became a relevant component of B-share
price formation. In addition, after Chinese investors were allowed to buy B
shares using foreign currency, the coefficient of a regression of turnover of B

21
shares on B-shares float turned from positive to negative, suggesting again
that trading in B shares may have become more driven by speculation.
The Chinese warrant bubble of 2005-2008 was used by Xiong and Yu
(2011) to test predictions of heterogeneous beliefs cum short-sale constraints
theories of bubbles. In 2005-2008, 18 Chinese companies issued put warrants
on their stock with maturities ranging from six months to two years. These
warrants gave the holder the right to sell the issuing companies’ stocks at
predetermined prices.
The extraordinary rise of prices in Chinese stocks between 2005 and
2007 made it almost certain that these warrants would expire without being
exercised. In fact, using the familiar Black-Scholes option-pricing formula
Xiong and Yu (2011) calculated that close to their expiration date, these
warrants often were worth less than .05 hundredth of a yuan. However
prices of these virtually worthless warrants varied substantially and averaged
.948 yuan during the days in which their Black-Scholes values fell below .05
hundredth of a yuan. Xiong and Yu (2011) also found that these warrants
sometimes traded for more than the strike price, what could only be justified
if investors were counting on a negative stock price in the future. Xiong and
Yu (2011) provide other bounds on the value of warrants that are violated
in this sample of warrants.
One security Xiong and Yu (2011) describe in detail is the put warrant on
the stock of WuLiangYe Corporation, a liquor producer.37 The put warrant
was issued on April 3, 2006 in-the-money with an exercise price of 7.95 yuan
while WuLiangYe’s stock traded at 7.11 yuan. Initially, the warrant was
valued close to 1 yuan, but in two weeks, WuLiangYe’s stock price exceeded
the strike and the warrant never returned in-the-money. On October 15th
2007, the stock reached a peak of 71.56 yuan and then drifted down to close
at 26 yuan at the expiration of the warrant on April 2, 2008. The calculation
by Xiong and Yu (2011) is that from July 07, the Black-Scholes price of this
put was below .05 hundredth of a yuan, but the warrant traded for a few
yuans, and only dropped below its initial price of .99 yuan in the last few
trading days. Xiong and Yu (2011) also notes that on at least two trading
days, June 13 and 14, 2007, the put warrant traded for more than the strike
price.
As in other episodes discussed in this lecture, these unjustifiably high
prices were accompanied by trading frenzies. The warrants with a Black-
Scholes value of less than .05 hundredth of a yuan had an average daily
turnover rate of 328 percent. On their last trading day, when they were
37
c.f. Figure 1 in Xiong and Yu (2011)

22
all virtually worthless, these 18 warrants, on average, turned over 100% of
their float every 20 minutes! The trading volume on the warrant on the
stock of WuLiangYe Corporation reached 1,841% of that warrant’s float
in the last trading day. Xiong and Yu (2011) show that as predicted by
the models discussed here, the size of the price bubble on a warrant was
positively correlated with the trading volume of that warrant or the time to
expiration, and negatively correlated with the warrant’s float.

7 Some final observations

One question left unanswered in this lecture is whether one could use the
signals associated with bubbles such as inordinate trading volume or high
leverage, to detect and perhaps stop bubbles. One of the difficulties in
using these signals is that we know next to nothing about false positives.
For instance, the typical empirical paper studying the association between
volume of trade and bubbles examines data during a bubble episode.
Even if we could effectively detect bubbles, it is not obvious that we
should try to stop all bubbles. Although credit bubbles have proven to have
devastating consequences, the relationship between bubbles and technologi-
cal innovation suggests that some of these episodes may play a positive role
in economic growth. The increase in the price of assets during a bubble
makes it easier to finance investments related to the new technologies.
The most straightforward policy recommendations that arise rom the
arguments advanced in this lecture is that to avoid bubbles, policy makers
should consider limiting leverage and facilitating, instead of impeding, short-
selling. In the panic that followed the implosion of the credit bubble, the
SEC banned short-sales of financial stocks. In August 2011, as the markets
questioned the health and funding needs of European financial institutions,
France, Italy, Spain and Belgium imposed bans on short-sales of financial
stocks. Each of these interventions may have have given a temporary respite
to the markets for these assets, but caused losses to investors that were short
these assets and had to cover their positions. Investors learned one more
time that it is dangerous to bet against overvalued assets - a lesson that
they will surely keep in mind in the next bubble.

23
 A Formal model

A.1 A Simple model

I first exposit a simple model to illustrate the role of costly shorting and dif-
ferences in beliefs in generating bubbles and the association between bubbles
and trading. Consider 4 periods t = 1, 2, 3, 4, a single good, and a single
risky asset in finite supply S. In addition to the risky-asset there exits also
a risk-free technology. An investment of δ units of the good in the risk-free
technology at t yields one unit in period t + 1. Assume there is a large num-
ber of risk-neutral investors, that only value consumption in the final period
t = 4. Each investor is endowed with an amount W0 of the good.
The risky asset produces dividends at times t = 2, 3, 4. At each t = 2, 3, 4
each unit of the risky asset pays a dividend θt ∈ {θ` , θh } with θh > θ` .
In what follows, I will refer to θh (θ` ) as the high (resp. low) dividend.
Dividends at any t are independent of past and future dividends.38 The
probability that θt = θ` is .5, and we write

θ̄ = E(θt ) = .5θ` + .5θh .

Assets are traded at t = 1, 2, 3, 4. If a dividend is paid in period t, trading

occurs after the dividend is distributed - that is the asset trades ex-dividend
and the buyer of the asset in period t has the rights to all dividends from
time t + 1 on. Thus in the final period t = 4 the price of the asset p4 = 0,
since there are no dividends paid after period 4. The price at time t = 1, 2, 3
depends on the expectations of investors regarding the dividends to be paid
in the future. We first calculate the willingness to pay of a “rational” risk-
neutral investor that is not allowed to resell the asset after she buys it. Since
the investor is risk-neutral, at time t = 3 she is willing to pay δ θ̄ for a unit
of the asset. In the absence of resale opportunities, at t = 2, the rational
investor is willing to pay δ[θ̄ + δ θ̄] = (δ + δ 2 )θ̄. Finally at t = 1 that same
rational investor with no resale opportunities would be willing to pay

(δ + δ 2 + δ 3 )θ̄ (1)

In addition, we suppose that at each t = 1, 2, 3 a signal st is observed

after the dividend at t (if t > 1) is observed but before trading occurs at t.
Each signal st assumes one of three values {0, 1, 2}, is independent of past
realizations of the signal and of the dividends and has no predictive power
38
It would be easy to accommodate a non-stationary dividend

24
for future dividends. Thus the signal st is pure noise. There are however two
sets of investors, A and B. Each set has many investors. Agents in group
A are rational and understand that the distribution of future dividends is
independent of st . Agents in B actually believe that st predicts θt+1 and
that the probability that θt+1 = θh given st is:

Probability[θt+1 = θh |st ] = .5 + .25(st − 1).

Thus agents in group B believe that the probability of a high dividend at

t + 1 increases with the observed st and that when st = 0 (st = 2), θt+1 = θ`
(resp. θt+1 = θh ) is more probable. All agents agree that st does not help
predict θt+j for j ≥ 2, and thus the only disagreement among investors is
whether st can predict θt+1 . To make agents in set B correct on average,
and thus assure that ex-ante there are no optimistic or pessimistic investors,
assume that the probability that st = 0 equals the probability that st = 2.
Write q < .5 for this common probability, and observe that the probability
that st = 1 is 1 − 2q > 0.
In the case of binary random variables, forecasts have minimal preci-
sion39 when the probability of each realization is 1/2. This is exactly the
forecast of rational agents here. Agents in group B after observing st = 1
also have the same minimal forecast precision. However if they observe
st = 0 or st = 2 they employ forecasts that have higher precision, since they
(mistakenly) believe that one of the two possible events has a probability
of 3/4. In this sense agents in group B have an exaggerated view of the
precision of their beliefs.
At time t = 3, rational agents in group A are willing to bid up to δ θ̄
for a unit of the asset. However, if s3 = 2, agents in set B believe that
the probability of a high dividend in period 4 is .75. Since these agents are
risk-neutral and the risk-free technology transforms δ units at 2 into 1 unit
at time 3, when s3 = 2 members of group B are willing to pay up to

δ(.75θh + .25θ` ) = δ[θ̄ + .25(θh − θ` )] > δ θ̄.

I assume that there are no short sales. Section A.2 below treats the case
when there is limited capital in the hands of a group of investors, but I will
initially assume that W0 is large enough so that each group of investors has
sufficient aggregate wealth to acquire the full supply of the asset at their
own valuation. Suppose s3 = 2. Since there are many agents in group B,
no short sales, and agents in group B have sufficient wealth to acquire the
39
That is, maximal variance.

25
full supply, all risky assets end up in the hand of some agents in group B
and competition among agents of group B guarantees that when s3 = 2 the
price
p3 (2) = δ[θ̄ + .25(θh − θ` )].
If s3 = 1, then all agents value the asset at δ θ̄, and thus

p3 (1) = δ θ̄.

If s3 = 0, agents in set B are unduly pessimistic and the asset ends up in

the hands of some rational agents that pay

p3 (0) = δ θ̄.

Since the probability that s3 = 2 is q, the price p3 equals δ θ̄ with probability

1 − q and equals δ[θ̄ + .25(θh − θ` )] > δ θ̄ with probability q. Hence, before
s3 is observed, all agents anticipate that the price of the asset in period 3
will on average equal

Ep3 = δ[θ̄ + .25q(θh − θ` )]. (2)

Every agent in group A and in group B expects the payoff of the asset in
period 4 to be on average exactly θ̄, but the average price in period 3 exceeds
the discounted value of this expected payoff, δ θ̄, by

δ[.25q(θh − θ` )],

reflecting the fact that for each realization of the signal s3 a member of the
most optimistic group would acquire the asset. When s3 = 0 (s3 = 2) agents
in group A (resp. B) are the most optimistic and end up holding the total
supply of the asset. When s3 = 1 both groups are equally optimistic and
any distribution of asset holdings across the two groups is compatible with
equilibrium.
Although the price paid in period t sometimes reflects the excessively
optimistic views of group B agents, our definition of a bubble - that is a
bubble occurs when buyers pay more than they think the future dividends
are worth - implies that there is no bubble in period 3. As I will show next,
this is not the case for any t < 3.
At time t = 2 if s2 = 2, agents in set B assume that the probability of
a high dividend θh in period 3 is .75. Since s3 has not yet been observed,
agents in set B forecast the price in period 3 to be on average Ep3 . Thus,
when s2 = 2 agents in group B are willing to pay,

p2 (2) = δ(θ̄ + .25(θh − θl ) + Ep3 ). (3)

26
Similarly, if s2 = 1 (s2 = 0) agents of both types (resp. agents of type A)
are willing to pay
p2 (0) = p2 (1) = δ(θ̄ + Ep3 ).
Hence, before s2 is observed, agents anticipate an average price for the asset:

Ep2 = δ(θ̄ + .25q(θh − θ` ) + Ep3 ) = (δ + δ 2 )[θ̄ + .25q(θh − θ` )]. (4)

A buyer of the asset in period 2 acquires the right to dividends in period 3

and 4. Before s2 is observed, agents of both types agree that these dividends
are worth (in period 2) exactly (δ +δ 2 )θ̄. However the average price in period
2 exceeds this fundamental value by

(δ + δ 2 ) × .25q(θh − θ` ).

This difference is a consequence of (i) the fact that in period 2 for each
realization of the signal s2 , the asset will be sold to the highest bidder and (ii)
any buyer of the asset in period 2 acquires the right to resell it in period 3 to
a buyer that is more optimistic than she is. (i) is worth δ×.25q(θh −θ` ), while
the option to resell in period 3 (ii) is worth δ 2 × .25q(θh − θ` ). Furthermore
even in the event s2 = 0 when “rational” (group A) agents acquire the full
supply of the risky asset, these agents pay

δ(θ̄ + Ep3 ) = δ θ̄ + δ 2 [θ̄ + .25q(θh − θ` )] (5)

This value exceeds the fundamental value, since buyers of the asset in period
2 will benefit from the resale option in period 3. When s2 = 1, holders of the
asset (in group A or B) are also willing to pay this same amount. Whereas in
period 3, prices exceed fundamentals only if group B agents are optimistic, in
period 2, prices exceed fundamentals even when group B agents are unduly
pessimistic.
More importantly, for every realization of the signal s2 , the buyer of
the asset is willing to pay in excess of her estimate of the value of future
dividends, an amount that represents the option to resell in period 3 and
that equals
b2 := .25δ 2 q(θh − θ` ). (6)
This difference can be naturally called a (period 2) bubble and is a result of
fluctuating differences of opinion and future opportunities to trade.
In order to model a bubble resulting from differences in beliefs and re-
strictions to short selling, it suffices to consider three periods. However, to
examine bubble implosions as in Section A.2, we need two periods in which

27
an asset pricing bubble can potentially occur. For this reason, I consider
four periods, but as I hope it becomes transparent, the reasoning in period
t = 1 duplicates exactly the argument we used for period 2 above.
In fact, buyers at t = 1 are willing to bid for the asset an amount that
reflects the sum of the dividends they expect the asset to pay and the value of
the option of reselling the asset in future periods. For instance, the valuation
of future dividends at t = 1 for a rational buyer is given by expression (1).
However the rational buyer is willing to pay at t = 1

δ(θ̄ + Ep2 ) (7)

The difference between (7) and (1) is

b1 := .25(δ 2 + δ 3 )q(θh − θ` ). (8)

It is easy to check that (8) also expresses the difference between the
reservation value of a B agent and her valuation of future dividends for
every value of the signal s1 . Thus for every realization of the signal s1 , b1
represents the amount that the buyer of the asset is willing to pay in excess
of her estimate of the value of future dividends.
Since there is no bubble in period 3, we may set b3 = 0 and a comparison
of (6) and (8) establishes that:

b1 > b2 > b3 .

Bubbles decline over time because there are fewer opportunities to resell.
In this very simple model, we may think of the parameter q as a measure
of differences in beliefs. After all, in periods t = 1, 2, 3 , with probability 2q
there are differences in beliefs once the signals st are observed. In addition,
as argued above, agents in group B exaggerate the precision of their beliefs
whenever st 6= 1. The event st 6= 1 has probability 2q. Thus an increase in
q also corresponds to an increase in the probability that agents in group B
exhibit overconfidence. The size of the bubble in periods 1 or 2 is increasing
as a function of q. Further, if we write
1
r := −1
δ
for the (risk-free) interest rate implicit in the risk-free technology, then in
periods 1 and 2 the bubble decreases with the risk-free interest rate.
The model discussed here has predictions on the effect of the difference
in beliefs q on the volume of trade. For symmetry, suppose both groups are

28
of the same size and at time 0 each group owns half of the supply of the
risky asset. In period 1, with probability 2q, all assets are bought by agents
in one of the two groups, and with probability 1 − 2q, s1 = 1 and all agents
agree on the distribution of dividends in the future, and thus there is no
reason to trade. Hence, since S is the total supply of the asset, the average
volume of trade in period 1 is
1
EV1 = × 2q × S = qS. (9)
2
If s1 = 2 and s2 = 0 or if s1 = 0 and s2 = 2 the group holding the risky-asset
in period 1 would sell it in period 2. The probability that s1 = 2 and s2 = 0
equals q 2 , which is also the probability that s1 = 0 and s2 = 2. Also, trade
will occur if s1 = 1, and s2 = 0 or s2 = 2, but in this case, only half the assets
would change hands. The probability of this event is (1 − 2q)2q. Finally, in
all other cases no trade would occur in period 2. Thus the expected volume
of trade in period 2 is,
S
EV2 = 2q 2 S + (1 − 2q)2q × = qS. (10)
2
All shares will change hand in period 3, if s3 = 2 and s2 = 0 or if s3 = 0
and s2 = 2. In addition, all shares will change hand at t = 3 if the history of
signals is (0, 1, 2) or (2, 1, 0). These four events have an aggregate probability
2q 2 + 2q 2 (1 − 2q). In addition, trade will occur if the histories of the signals
is (1, 1, 2) or (1, 1, 0), and each of these events have probability q(1 − 2q)2 ,
but if histories (1, 1, 2) or (1, 1, 0) obtain, only half the shares will exchange
hands at t = 3. Thus the average volume in period 3 is given by

EV3 = 2q 2 S + 2q 2 (1 − 2q)S + q(1 − 2q)2 S = qS. (11)

An examination of expressions (9) - (11) makes it evident that an increase in

the parameter q increases the expected volume of trade in every period, just
as it increases the value of the bubble. An increase in q raises the probability
that the option to resell will be exercised (volume) and the value of that
resale option (size of bubble).
I summarize these results in a Proposition:
Proposition 1. In the presence of fluctuating differences in beliefs and
short-sale constraints, bubbles exist - investors are willing to pay for an
asset in excess of their own valuation of future dividends. In addition,
(i) The size of the bubble increases when the probability of disagreement
increases.

29
 (ii) The volume of trade also increases with the probability of disagreement.

(iii) The size of the bubble decreases with the risk-free interest rate.

(iv) The bubble declines as the time of maturity of the asset approaches,
because there are fewer opportunities to trade.

A.2 Limited capital

It is straightforward to incorporate more periods into the model or treat
a stationary model with infinite horizon. The introduction of risk-aversion
complicates substantially the computations. However as Hong et al. (2006)
showed, in a model of heterogeneous beliefs and costly short sale with risk-
averse agents, the bubble and asset turnover rate decrease as the supply
of the asset increases and bubbles may implode when float unexpectedly
increases. The intuition is that as their holdings of the asset increases,
risk averse agents have a smaller marginal valuation for the asset and, as
a consequence, it takes a bigger difference in opinions for the whole asset
supply to change hands. An alternative that displays the same relationships
between float, bubbles and turnover as when agents are risk averse, albeit in
a less continuous manner, is to adopt the short-cut proposed by Allen and
Gale (2002) of limited capital and cash-in-the-market-pricing. Each group
of agents will have the same preferences and views on signals as in appendix
A.1, but here I no longer assume that the wealth of agents in both groups is
so large that the they can always acquire the full supply of the risky asset
at their own valuation. The demand for the risky asset by an agent will be
limited by the liquidity in her hands.
Before trading in period t, but after receiving period t dividends, agents
in group C ∈ {A, B} will have an aggregate portfolio (St−1 C , K C ). Here
t
C
St−1 is the amount of the risky-asset they chose to hold in period t − 1
and KtC > 0 is the cash available to them. KtC is a result of investments
in the risk-free technology in period t − 1, period t dividends on risky-asset
holdings and any (net) risk-free borrowing the agents in group C may access.
Below, I will deduce each group’s portfolio holdings before trading in period
t, as a function of their initial holdings (S0C , K0C ), portfolio decisions, and the
realizations of signals and dividends, but at this point I treat these portfolios
as exogenous.
Write vtC for the (marginal) valuation that agents in group C ∈ {A, B}
have for the risky asset at time t. More specifically vtC equals the maximum
price per-unit that an agent in group C that has a positive income is willing
to pay for an infinitesimal amount of the asset. This amount is a function

30
of the expectations that the agent has concerning dividends and prices in
period t+1. The limitation imposed by liquidity constraints on demand may
yield lower equilibrium prices than those derived in appendix A.1 and thus
cause the marginal valuation of some agents to be smaller than the marginal
valuations deduced in appendix A.1.
Write St for the total supply of the asset at t. To simplify matters, I will
assume that there is enough capital in the hands of group A agents so that
at time t they can acquire the whole supply not yet in the hands of group
A , at their valuation v A . That is, I assume that:
A agents, St − St−1 t

KtA
A
> vtA (12)
St − St−1

Notice that inequality (12) involves the portfolios held by group A agents
and the valuation of A agents, vtA . However one may replace vtA in the r.h.s.
of (12) by the (larger) valuation of the risky-asset by A agents that was
already calculated in appendix A.1, and the denominator St − St−1 A by S .
t
This stronger assumption would imply inequality (12).
I will assume however that agents in group B have more limited capital.
Suppose in period t group B agents have a valuation vtB for the risky asset,
while the valuation of A agents is vtA . If vtA ≥ vtB then the price of the asset
pt = vtA since inequality (12) insures that group A agents have sufficient
KB
capital to acquire the total supply St at their valuation vtA . If (S −St B ) ≥
t t−1
vtB > vtA then B agents would bid up the price, until pt = vtB , since in
this case group B agents have sufficient capital to acquire the risky assets
KB
not in their hands at their own valuation. If vtA ≤ (S −St B ) < vtB , then
t t−1
the marginal buyer would belong to group B, but because of the liquidity
constraints on group B agents, the price is then given by

KtB
pt = B )
.
(St − St−1

In the language of Allen and Gale (2002), cash-in-the-market-pricing ob-

tains; the price of the asset is determined by the liquidity available to group
KB
B agents. Finally, if vtA > (S −St B ) then the marginal buyer is necessarily
t t−1
in group A and pt = vtA . Thus:
( ( ))
KtB
pt = max vtA , min vtB , B )
(St − St−1

31
 Other things equal, this price decreases as St increases. For instance,
the larger is the initial float S1 relative to the initial holding of the B group,
the lower would be the initial price p1 . If the initial float is large enough,
then even when B agents are optimists some of the asset supply ends up in
the hands of A agents, because of the limited capital of group B agents. In
fact, when B agents are optimists, but the marginal buyer is an A agent,
the amount that ends in the hands of B agents, S1B , solves:

K1B
v1A = .
(S1B − S0B )

In this way, a larger float lowers the price and the turnover of the asset. If
the supply of the risky asset remains constant through time, an asset with
a larger float will also have a smaller turnover in subsequent periods.
To study the effect of an increase in the supply of the risky asset in period
2, we will first consider the price that would prevail in period 3. When s3 = 2
(group B agents are optimists) v3B > v3A . If K3B ≥ (S3 − S2B )v3B , then group
B agents will bid up the price of the asset to their valuation v3B and acquire
the full supply. However, if s3 = 2 but K3B < (S3 − S2B )v3B , group B agents
do not have enough wealth to buy the full supply of the risky-asset at their
valuation v3B . In this case, if they buy the full supply, the maximum group
KB
B agents can pay for an unit of the asset is S −S3
B . If the price group B
3 2
agents can afford when they acquire the full supply in the hands of group
K3B A
A exceeds the valuation of group A agents, that is, S −S B > v3 = δ θ̄, then
3 2
KB KB A
the equilibrium price when s3 = 2 is S −S 3
B . Finally if S −S B < δ θ̄ = v3 ,
3
3 2 3 2
some of the asset is held by group A agents even though s3 = 2, and thus
the price of the risky asset must equal δ θ̄. The price p3 when s3 = 2 is a
(weakly) decreasing function of S3 , and equals δ θ̄ if the supply of the asset
in period 3 is large enough. Since p3 = δ θ̄ whenever s3 6= 2 the average price
in period 3, Ep3 , is also a (weakly) decreasing function of S3 and equals δ θ̄,
if the supply of the asset in period 3 is large enough, when compared to
the holdings of B agents in period 2. This contrasts with the case without
limited capital when Ep3 would always exceed δ θ̄ (c.f. equation (2) above.)
Buyer’s of the stock of the South Sea Company or buyers of Internet
stocks in the late 90’s did not know with certainty the future supply of
these assets. One can model this uncertainty in a simple way by assuming
that supply may increase in period 2. Suppose S0 = S1 = S, and that
S3 = S2 = S with probability π and equals S + ∆S > S with probability
1 − π. Suppose further that the realization of the supply of the risky asset is

32
independent of the realization of (θ1 , θ2 , s1 , s2 ), and that this realization is
observed in period 2 before the signal s2 is observed. Since trading in period
2 occurs after s2 is observed, investors know the actual supply when they
trade in period 2. While the amount S is held from the beginning by agents
in groups A and B any increase in supply will come from sales of the asset
by “insiders”, as in Hong et al. (2006). To simplify matters, I will assume
that insiders only wish to sell.40
I will show that for certain parameter values, when the supply does not
increase (S2 = S) the bubble will persist but when the supply increases the
bubble will deflate. The deflation of the bubble in period 2 will be a conse-
quence of the realization by all investors that, because of the supply increase
and the limited aggregate wealth of group B agents, the marginal buyer in
period 3 will necessarily be rational even when s3 = 2. The occurrence of
this scenario will depend on the wealth of group B agents in period 3 and
the supply they would have to absorb in that same period to guarantee that
the price exceeds the valuation of the rational buyers. I will start by im-
posing some bounds on the aggregate portfolio of group B agents that will
insure that the bubble persists when the supply of the asset is unchanged
and the bubble deflates when the supply increases, and will later show that
these bounds would hold when the initial supply of the risky asset, S, is
sufficiently small and the potential increase in the supply in period 2, ∆S,
is sufficiently large.
Suppose that the capital constraints of group B only bind in period 3
when the supply increases:

K3B
≥ δ[θ̄ + .25(θh − θ` )], (13)
S − S2B
and that when the capital constraints bind, they are strong enough that is:

K3B
< δ θ̄. (14)
S + ∆S − S2B

Equation (13) guarantees that if S2 = S and s3 = 2, the demand by group

B agents will drive the price of the asset in period 2 to v2B and thus exceed
the expected “rational” payoff in period 3. Inequality (14) insures that if
S2 increases to S + ∆S the price of the asset in period 3 will equal the
(discounted) average payoff expected by rational agents, since a member of
group A would necessarily be the marginal buyer of the asset.41
40
However see footnote 41 below, for an assumption that implies that insiders only sell.
41
Equivalently we could have assumed instead that the reservation price of insiders

33
 Thus when S2 = S the price in period 3 matches exactly the price when
there are no capital constraints whereas when S2 = S + ∆S the price in
period 3 equals the valuation of the rational agents, independent of the
signal. In particular, if the supply of the asset increases in period 2 because
of sales by insiders, that is S2 = S + ∆S, there is no period 2 bubble, since
it is known that period 3 prices are independent of the signal s3 that is
observed. On the other hand, if supply of the asset does not increase in
period 2, then, as in the case of no capital constraints,

Ep3 = δ[θ̄ + .25q(θh − θ` )],

reflecting the fact that B agents would acquire the whole float if s3 = 2.
In order to guarantee that if S2 = S prices in period 2 would be exactly
as in the case of no capital constraints we need to insure that

K2B
≥ (δ + δ 2 )[θ̄ + .25q(θh − θ` )]. (15)
S − S1B

Inequality (15) guarantees that when s2 = 2 and S2 = S, B agents will

acquire the full float of the asset. To insure that when S2 = S + ∆S the
marginal buyer of the asset always belongs to group A, it suffices that

K2B
≤ (δ + δ 2 )θ̄. (16)
S + ∆S − S1B

If inequalities (13) and (15) hold then before S2 (and hence s2 ) is ob-
served

Ep2 = (δ + δ 2 )[θ̄ + .25qπ(θh − θ` )]

reflecting the fact that a bubble would occur in period 2 if and only if the
supply stays constant. In addition when inequalities (13), (14) and (15)
hold, rational buyers in period 1 will always be willing to pay in excess
of their own valuations of future dividends of the risky asset, because if
supply does not increase they may have an opportunity to sell the asset
to over-optimistic B agents in the future. Since we have assumed that
group A agents have sufficient capital to buy the total supply of the asset at
these higher prices, the price of the asset in period 1 exceeds the expected
is greater that v2B , with probability π, and less than δ θ̄, with probability 1 − π. For in
this case, given that equation (13) and inequality (14) hold, insiders sell no shares with
probability π and sell all their shares with probability 1 − π.

34
discounted dividends independently of the realized signal in period 1 (s1 )
and the capital constraints of group B agents.
When s1 = 2, group B valuation at time 1 is

v1B (2) = δ θ̄ + .25q(θh − θ` ) + (δ + δ 2 )[θ̄ + .25qπ(θh − θ` )] ,

which again exceeds the present value of dividends expected by group B

agents by (δ + δ 2 )[.25qπ(θh − θ` )]. To guarantee that the price in period 1
when s1 = 2 is observed equals v1B (2) one must assume that:

K1B
≥ v1B (2) (17)
S − S0B
In this case, there will be a bubble in period 1, independently of the
signal s1 observed and b1 = .25(δ + δ 2 )qπ(θh − θ` ), which is smaller than the
bubble that obtains when liquidity constraints are not binding - equation
(8) - reflecting the possibility that future supplies may increase. However
as π → 1 the value of the bubbles under the two scenarios converge to each
other.
Until now I took the portfolio held by group B agents after the payment
of dividends in periods 1 to 3 as given, and assumed that inequalities (13)-
(17) hold. Agents in group B start with a non-negative initial endowment
of S0B < S units of the risky asset and K0B units of the good and the values
B ≤ S and K B for t = 2, 3 are consequences of their actions, realizations
St−1 t
of the random shocks and equilibrium prices in period 1 and 2. Furthermore,
for a given K0B it is straightforward to show that inequalities (13), (15) and
(17) hold, whenever S is small enough. The intuition underlying this result
is that if S is small enough, even if B agents acquire the full supply of the
risky asset and have no possibility of borrowing, they would have a minimum
amount left over to invest in the risk-free technology. This delivers a lower
bound on K1B , K2B and K3B , the amounts available to agents in group B to
acquire additional shares in periods 1 to 3 . By assuming an even smaller
value for S, if necessary, we can thus guarantee that inequalities (13), (15)
and (17) hold.
I now show that inequalities (14) and (16) hold provided ∆S is large
enough and agents in group B have no access to borrowing in addition to the
amount that is already reflected in their initial liquidity K0B . To accomplish
this, we have to study the dynamics of the evolution of the aggregate wealth
of B agents. Given K0B and S0B , the evolution of the aggregate wealth of
group B in equilibrium depends on the realizations of the dividends, signals
and supply and on the way assets are allocated between the two groups when

35
their valuations are identical. I assume that when the two groups have
identical valuations for the risky asset, group B agents get all the shares
they want.42 Write WtB (K0B , S0B ) for the maximum wealth that agents in
group B can have after dividend payments in period t, where the maximum
is taken over all possible realizations of signals, dividends and all portfolio
choices. Notice that that an increase in ∆S cannot increase W1B or W2B
because the price of the risky asset can only decrease with an increase in
∆S. The price of the asset in period 2 is not less than (δ +δ 2 )θ̄, the expected
(by A agents) discounted dividends of the asset. If ∆S is large enough,

S + ∆S S + ∆S
W2B < δ + δ 2 θ̄ ≤ p2 .
2 2
Even if B agents use all their wealth in period 2 to buy the asset they
cannot acquire more then half the total (larger) supply, and thus inequality
(16) holds and,
S + ∆S
S2B < . (18)
2
Hence the marginal buyer in period 2 when supply increases always belongs
to group A. Further, the ex-post rate of return of the risky asset between
periods 2 and 3 will depend on the price of the asset that prevails in periods
2 and 3 and the realization of the dividend θ3 . Since the price of the risky
asset in period 3 is at most the expected (by B agents) discounted dividends
of the asset when s3 = 2, and the dividend paid is at most θh , the rate of
return is at most

θh + δ θ̄ + .25(θh − θ` )
R̄ :=
(δ + δ 2 )θ̄
Notice that this bound exceeds 1 + r and thus is also a bound for the growth
in wealth. Hence,
W3B ≤ R̄W2B
and by choosing if necessary a larger ∆S we can insure that
∆S
K3B ≤ W3B ≤ R̄W2B ≤ δ θ̄ < δ θ̄ S + ∆S − S2B ,


2
where the last inequality follows from equation (18). Hence inequality (14)
holds.
Thus in the model in this section, a bubble arises in period 1 provided
that “irrational” agents have enough initial wealth relative to the initial
42
A different rule, such as giving priority to agents in group A, would change the details
of the computations that follow, but would not alter the result of interest

36
supply of the risky asset and the bubble implodes in period 2 if and only
if the supply of the risky asset increases by a sufficiently large amount in
period 2.

References
Allen, F. and D. Gale. 2002. Optimal financial crises. The Journal of Finance
53 (4):1245–1284.

Alpert, M. and H. Raiffa. 1982. A progress report on the training of probabil-

ity assessors. Judgment under uncertainty: Heuristics and biases 294–305.

Anderson, A. 1787. An historical; and chronological deduction of the origin

of commerce, vol. 3. London, UK.

Arrow, K.J. 1986. Rationality of self and others in an economic system.

Journal of Business 385–399.

Bebchuk, L.A. and J.M. Fried. 2006. Pay without performance: The unful-
filled promise of executive compensation. Harvard University Press.

Ben-David, I., J.R. Graham, and C.R. Harvey. 2010. Managerial miscali-
bration. Tech. rep., National Bureau of Economic Research.

Bolton, P., J. Scheinkman, and W. Xiong. 2006. Executive compensation

and short-termist behaviour in speculative markets. The Review of Eco-
nomic Studies 73 (3):577–610.

Brunnermeier, M. K. 2008. bubbles. In The New Palgrave Dictionary of

Economics, edited by S. Durlauf and L. Blume. Basingstoke: Palgrave
Macmillan.

Carlos, A., L. Neal, and K. Wandschneider. 2006. Dissecting the anatomy of

exchange alley: The dealings of stockjobbers during and after the South
Sea bubble. Unpublished paper, University of Illinois .

Cassidy, J. 2010. Rational irrationality: An interview with Eugene Fama.

The New Yorker November 1.

Cheng, I.H., H. Hong, and J. Scheinkman. 2010. Yesterday’s Heroes:

Compensation and Creative Risk-Taking. ECGI-Finance Working Paper
(285).

37
Cooper, A.C., C.Y. Woo, and W.C. Dunkelberg. 1988. Entrepreneurs’ per-
ceived chances for success. Journal of business venturing 3 (2):97–108.

Cowen, A., B. Groysberg, and P. Healy. 2006. Which types of analyst firms
are more optimistic? Journal of Accounting and Economics 41 (1):119–
146.

Davis, L.E., L.D. Neal, and E.N. White. 2005. The Highest price ever: The
great NYSE seat sale of 1928-1929 and capacity constraints. Tech. rep.,
National Bureau of Economic Research Cambridge, Mass., USA.

D’avolio, G. 2002. The market for borrowing stock. Journal of Financial

Economics 66 (2):271–306.

Diether, K.B., C.J. Malloy, and A. Scherbina. 2002. Differences of opinion

and the cross section of stock returns. The Journal of Finance 57 (5):2113–
2141.

Foote, C.L., K.S. Gerardi, and P.S. Willen. 2012. Why did so many people
make so many ex post bad decisions? the causes of the foreclosure crisis.
Tech. rep., National Bureau of Economic Research.

Friedman, M. 1966. Essays in positive economics, vol. 231. University of

Chicago Press.

Froot, K.A., A.F. Perold, and J.C. Stein. 1991. Shareholder trading prac-
tices and corporate investment horizons. Tech. rep., National Bureau of
Economic Research.

Garber, P. 1980. Famous first bubbles: The fundamentals of early mania.

Cambridge, Ma: MIT Press, 1st ed.

Geanakoplos, J. 2010. The leverage cycle. NBER Macroeconomics Annual

24 (1):1–66.

Glaeser, Edward L, Joseph Gyourko, and Raven E Saks. 2006. Urban growth
and housing supply. Journal of Economic Geography 6 (1):71–89.

Harris, Ron. 1994. The Bubble Act: Its passage and its effects on business
organization. The Journal of Economic History 54 (3):pp. 610–627.

Harrison, J.M. and D.M. Kreps. 1978. Speculative investor behavior in a

stock market with heterogeneous expectations. The Quarterly Journal of
Economics 92 (2):323–336.

38
Holmström, B. 2006. Pay without performance and the managerial power
hypothesis: A comment. Available at SSRN 899096 .

Holmström, B. and S.N. Kaplan. 2001. Corporate Governance and Merger

Activity in the US: Making Sense of the 1980s and 1990s. Tech. rep.,
National Bureau of Economic Research.

Hong, H. and J.C. Stein. 2007. Disagreement and the stock market. The
Journal of Economic Perspectives 21 (2):109–128.

Hong, H., J. Scheinkman, and W. Xiong. 2006. Asset float and speculative
bubbles. The Journal of Finance 61 (3):1073–1117.

———. 2008. Advisors and asset prices: A model of the origins of bubbles.
Journal of Financial Economics 89 (2):268–287.

Janeway, W. 2012. Doing capitalism in the innovation economy. Cambridge,

UK: Cambridge University Press, 1st ed.

Kidd, J.B. 1970. The utilization of subjective probabilities in production

planning. Acta Psychologica 34:338–347.

Lamont, O.A. and R.H. Thaler. 2003. Can the market add and subtract?
mispricing in stock markets carve-outs. Journal of Political Economy
111 (2):227–268.

Lewis, M. 2004. The Irresponsible Investor. New York Times Magazine

June 6.

Mackay, C. 1932. Extraordinary popular delusions and the madness of

crowds. New York: Farrar, Straus and Giroux.

Mei, J., J.A. Scheinkman, and W. Xiong. 2009. Speculative trading and
stock prices: Evidence from Chinese AB share premia. Annals of Eco-
nomics and Finance 10 (2):225–255.

Murphy, A.E. 1986. Richard Cantillon: entrepreneur and economist. Claren-

don Press Oxford.

Neal, L. 1990. The rise of financial capitalism: international capital markets

in the age of reason. Cambridge, UK: Cambridge University Press.

Nocera, J. 2010. A Wall street invention let the crisis mutate. New York
Times April 17:B1.

39
Ofek, E. and M. Richardson. 2003. Dotcom mania: The rise and fall of
internet stock prices. The Journal of Finance 58 (3):1113–1138.

Santos, M.S. and M. Woodford. 1997. Rational asset pricing bubbles. Econo-
metrica 19–57.

Scheinkman, J.A. and W. Xiong. 2003. Overconfidence and Speculative

Bubbles. Journal of Political Economy 111 (6).

Scott, W. R. 1910-12. The constitution and finance of English, Scottish and

Irish joint-stock companies by 1720, vol. 3. Cambridge, UK: Cambridge
University Press.

Shiller, R.J. 2006. Irrational exuberance. Crown Business.

Tetlock, P.E. 2005. Expert political judgment: How good is it? How can we
know? Princeton University Press.

Thomas, J. and R. Van Order. 2010. Housing policy, subprime markets and
Fannie Mae and Freddie Mac: What we know, what we think we know
and what we dont know. In St. Louis FRB Conference.

Wolfe, Tom. 1975. The painted word. New York: Farrar, Straus and Giroux.

Xiong, W. and J. Yu. 2011. The Chinese warrants bubble. American Eco-
nomic Review 101:2723–2753.

Yan, Hongjun. 2008. Natural Selection in Financial Markets: Does It Work?

Management Science 54 (11):1935–1950.

40