Durham E-Theses

Stock return, risk and asset pricing

Abu Ghunmi, Diana Nawwash Abed El-Hafeth

How to cite:

Abu Ghunmi, Diana Nawwash Abed El-Hafeth (2008)

Stock return, risk and asset pricing,

Durham theses,

Durham University. Available at Durham E-Theses Online: http://etheses.dur.ac.uk/1921/

Use policy

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-prot purposes provided that:

a full bibliographic reference is made to the original source a link is made to the metadata record in Durham E-Theses the full-text is not changed in any way

The full-text must not be sold in any format or medium without the formal permission of the copyright holders. Please consult the full Durham E-Theses policy for further details.

Academic Support Oce, Durham University, University Oce, Old Elvet, Durham DH1 3HP e-mail: [email protected] Tel: +44 0191 334 6107 http://etheses.dur.ac.uk

Stock Return, Risk and Asset Pricing

By

Diana Nawwash Abed El-Hafeth Abu Ghunmi

Supervisors Professor Krishna Paudyal Dr Keith Anderson

Submitted for the Degree of Doctor of Philosophy (PhD) in Finance

Durham Business School University of Durham

September2008
a15

JAN2009

Author's Declaration

I declare that no part of the material contained in this thesis has been previously submitted, either in whole or in part, for a degree in this or in any other university. The copyright of this thesis lies with its author, and no quotation from it and no information derived from it may be published without her prior permission in

writing. In addition, any quotation or information taken from this thesis must be
properly acknowledged.

Diana Nawwash Abed El-Hafeth Abu Ghunmi

Abstract

Stock Return, Risk and Asset Pricing

By Diana Nawwash Abed El-Hafeth Abu Ghunmi

This thesis attempts to address a number of issues that have been identified

in the asset

pricing literature as essential for shaping stock returns. These issues include the need to uncover the link between the macroeconomic variables and stock returns. In addition to of the literature, whether to advise

this, is the need to decide, in light of the findings investors to include idiosyncratic

risk and downside risk as risk factors in their asset

pricing models. The results presented here suggest, consistent with other previous studies, that stock returns are a function of a number of previously with the wider set of macroeconomic variables. identified risk factors along variables could

These macroeconomic

be represented by a number of estimated macro factors. However, estimated factors emerged as significant

only one of these

in explaining the cross-section of stock returns. factors remain

Nevertheless, it is important to note that the size (SMB) and value (HML) important existence factors in explaining of the other risk

the cross sectional returns on UK stocks, even with the factors. This finding of inability of the examined may

macroeconomic

variables to capture the pricing power of the SMB and the HML of finding more macroeconomic

cast doubt on the possibility

factors that are able to this

account for these two factors in the cross section of returns in the UK. Interestingly, conclusion seems to contradict

in links UK findings the of potential previous authors'

important find is downside The that that an risk results also support market. past studies risk factor and by allowing conditions, Nevertheless, economic downside risk the downside might risk premium to vary with business cycle risk than market risk.

be a better

measure of this finding

this thesis shows that although during

is applicable

in times of between

expansion,

recession, there is no conclusive

relationship

downside risk and stock returns. Furthermore, this thesis supports the studies which find that idiosyncratic risk is not significant in pricing stocks. However in contrast to other

studies, it reveals this by showing that time-varying

risk could be the reason behind the

nn

z -

, c ' ._.

for findings idiosyncratic This illusive that, thesis confirms of risk effect. potentially London Stock Exchange investors, macroeconomic variables should never be overlooked when estimating stock returns and downside risk could be an influential risk factor.

Acknowledgment

Thank you Allah for giving me a strong faith in you, and the patience and the will to get this PhD thesis completed. I know that you will never ever let me down.

It had been long journey and there came to many points when I could not see an end
to it but I am finally there and this could not have happened without the help of the

people around me. I am deeply indebted to my supervisor Professor Krishna Paudyal, for his deep insights and suggestions as well as his comments, support and advice. I am also deeply indebted to my supervisor Dr. Keith Anderson for his constructive comments,

help and suggestions. I would also like to thank Professor Antonios Antoniou for his help as a supervisor through the first period of my PhD.

Now I turn to thank the best family on earth. No one can ask for better parents than my parents; Baba Nawwash and Mama Salimeh, I am so grateful to you, I am so sorry for all the anxiety and concern that I have caused to you. Thank you so much for being always there for me, with your warm hearts and encouragement I have managed my way through this long and exhausting journey. Special thanks also to my sister Lina, brother Mohammad, and brother Obaida, I am blessed for having you as my siblings. You have been so patient and understanding, I do not know how to thank you. My family, I cannot ask you for more, as there is nothing anyone can give more than you give to me. Uncle Salim, I will never forget your support, your frequent phone calls and encouragement,
thank you so much.

Diana, 24 - September -2008

Table of Contents

Chapter 1 Introduction
1.1 1.2

1.3 Idiosyncratic Risk and Time Varying Betas 15 ...................................................... 1.4 Downside Risk and Business Cycle 16 .................................................................. 1.5 Issues and Contribution 17 ..................................................................................... Chapter 2 Macroeconomic Factors and Fama and French Asset Pricing Model 22 ..... 2.1 23 Introduction ....................................................................................................... 2.2 31 Literature Review .............................................................................................. 33 2.2.1 Fama and French's (1993) Three-Factor Model ........................................ 2.2.2 Intertemporal Capital Asset Pricing Model 39 ............................................... 2.2.3 Dynamic Factors Models 44 ........................................................................... 2.2.4 48 Factor Models and Asset Pricing Models .................................................. 2.3 Hypotheses 50 ......................................................................................................... 2.3.1. Fama and French's (1993) Model and UK Stock Market 50 ......................... 2.3.2 State Variables and Stock Returns 51 ............................................................. 2.3.3 Are Fama and French (1993) Factors tracked by macroeconomic forces in UK Stock Market? 52 ...................................................................... Stock Returns
2.3.4 Large Panel of Macroeconomic Variables and

10 Introduction ....................................................................................................... Macroeconomic Factors and Fama and French Asset Pricing Model 11 ...............

9 .....................................................................................................

2.3.5 Other Estimation Methods 56 ......................................................................... Data and Methodology 57 2.4 ...................................................................................... 2.4.1. 57 Data ............................................................................................................ 2.4.1.1 Stock returns 57 .......................................................................................... 62 2.4.1.2 State Variables - Petkova's (2006) Chosen Variables ........................... 64 2.4.1.3 State Variables - Large Panel of Macroeconomic Variables ................. 66 2.4.2 Methodology .............................................................................................. 66 2.4.2.1 Petkova's (2006) Model of the ICAPM ................................................ 2.4.2.2 VAR Innovations 67 ................................................................................... 69 2.4.2.3 Kalman Filter Innovations ..................................................................... 2.4.2.4 Dynamic Factor Model of Stock and Watson (2002a, b): Their Static
71 Representation ........................................................................................... 73 2.4.2.5 Fama and MacBeth's (1973) Cross Sectional Regressions ...................

..................

54

76 2.4.2.6 Generalized Methods of Moments (GMM) ........................................... 80 2.5 Results ............................................................................................................... 80 2.5.1 Fama-French's (1993) Three-Factor Model in the UK-LSE .................... 2.5.2 Petkova's (2006) Model that includes Excess Market Return and 82 Innovations to Four Variables ................................................................... Fama and French's (1993) Factors and State Variables; Do they 2.5.3 86 Relate? ....................................................................................................... 89 The Common Macro Factors and the Stock Returns 2.5.4 ................................. 90 2.5.4.1 Are the Common Macro Factors Pricing Factors? ................................ 2.5.4.2 Do the Priced Common Macro Factors Relate to HML and SMB?...... 92

2.5.5 2.5.6
2.5.7

The Macro Common Factors; Do they contain different news to stock prices? ..............................................................................................95 97 What is the Priced Estimated Macro Factor? ............................................ 99
Visual Representation .............................................................................. ..

101 2.6 Conclusion ....................................................................................................... 122 Chapter 3 Idiosyncratic Risk and Time-Varying Betas ............................................. 123 3.1 Introduction ..................................................................................................... 128 3.2 Literature Review ............................................................................................ 128 3.2.1 Idiosyncratic Volatility and Stocks Returns ............................................ 136 3.2.2 Time-Varying Betas ................................................................................ 139 3.3 Hypotheses Development ................................................................................ 139 3.3.1 Idiosyncratic Risk in UK . ......................................................................... 141 3.3.2 Time-Varying Beta and Idiosyncratic Risk ............................................. 144 3.4 Data and Methodology .................................................................................... 144 3.4.1 Data . .......................................................................................................... 3.4.2 146 Methodology ............................................................................................ 146 3.4.2.1 Idiosyncratic Risk Measures ................................................................ 3.4.2.2 The Cross-Sectional Regression 149 .......................................................... 3.4.2.3 Portfolio Formation 151 ............................................................................. 3.4.2.4 The Time-Varying Beta Model of Avramov and Chordia (2006)....... 155 157 3.5 Results ............................................................................................................. 157 3.5.1 Idiosyncratic Risk and Stock Returns ...................................................... 158 3.5.1 OLS Daily - Idiosyncratic risk / Cross Sectional Regression ........... .1 158 3.5.1 OLS Daily - Idiosyncratic risk / Portfolios ....................................... .2 162 3.5.1 OLS Monthly Idiosyncratic Risk / Cross Sectional Regression .......... .3 163 3.5.1 OLS Monthly Idiosyncratic Risk / Portfolios ...................................... .4 166 3.5.1 EGARCH / Cross Sectional Regression ............................................. .5 167 3.5.1.6 EGARCH / Portfolios ......................................................................... 169 3.5.2 Time Varying Beta and Idiosyncratic Risk ............................................. 175 3.6 Findings Summary ........................................................................................... 175 3.7 Conclusion ....................................................................................................... 176 3.7.1 OLS Daily Frequency .............................................................................. 177 3.7.2 OLS Monthly Frequency ......................................................................... 178 3.7.3 EGARCH Monthly Frequency ................................................................
3.7.4 Time-Varying Betas and Idiosyncratic Risk

3.7.5 Concluding Remark 179 ................................................................................. 196 Chapter 4 Downside Risk and Business Cycle ............................................................
4.1 Introduction ..................................................................................................... Literature Review 4.2 ............................................................................................ Downside Beta 4.2.1 ......................................................................................... Downside Risk in UK 4.2.2 .............................................................................. Business Cycle and Time-Varying Risk 4.2.3 .................................................. 197 201 201 207 209

...........................................

178

211 Hypotheses Development 4.3 ................................................................................ Is Downside Risk a Significant Beta Factor in the UK? 212 4.3.1 ......................... 214 Downside Risk and Firm Characteristics 4.3.2 ................................................ 215 Downside Risk and Business Cycle 4.3.3 ........................................................

217 Downside Risk and Industry.... 4.3.4 .................................................. 218 Data and Methodology 4.4 .................... ................................................... 218 Data 4.4.1 ......................................... .............................................. 221 Methodology 4.4.2 ............................................................................................ 221 4.4.2.1 The Downside Risk Model of Ang, Chen and Xing (2006) ................ 4.4.2.2 Risk Measures 223 ...................................................................................... 224 4.4.2.3 Portfolio Formation ............................................................................. 4.4.2.3.1 Risk Sorted Portfolios 224 .................................................................... 4.4.2.3.2 Accounting for Size and Book-to-Market Value Effects 227 .............. 4.4.2.4 Fama and MacBeth's (1973) Cross-Sectional Regression 228 ..................
4.4.2.5

Chapter

230 4.5 Results ............................................................................................................. 230 4.5.1 Downside Beta and Realized Returns ..................................................... 236 4.5.2 Downside Beta and Size and Book-to-Market Value .............................. 4.5.2.1 Downside Beta and Size 237 ...................................................................... 4.5.2.2 Downside Beta and Book-to-Market 240 ................................................... 4.5.3 242 Downside Risk and Economic Conditions .............................................. 4.5.4 Downside Risk and Industry 246 .................................................................... 4.5.5 Is Past Downside Beta a Good Proxy for Future Downside Beta? 250 ......... 4.5.6 253 Monthly Return and Downside Risk ....................................................... 4.6 Conclusion 254 .......................................................................................................
5 Conclusion .................................................................................................... 276

Downside Risk Premium and Business Cycle 229 .....................................

Conclusion 277 5.1 ....................................................................................................... 277 5.2 Macroeconomic Factors and Fama and French Asset Pricing Model ............. 280 5.3 Idiosyncratic Risk and Time-Varying Betas ................................................... 283 5.4 Downside Risk and Business Cycle ................................................................ 285 5.5 Implications ..................................................................................................... 286 5.6 Further Area for Research ............................................................................... 286 5.6.1 Idiosyncratic Risk and Time-Varying Betas ........................................... 5.6.2 287 Downside Risk and Business Cycle ........................................................ 119 Appendix (A) .................................................................................................................. 290 Bibliograpy ....................................................................................................................

List of Tables

Table 2.1 Table 2.2 Table 2.3 Table 2.4 Table 2.5 Table 2.6 Table 2.7
Table 2.8

Number of Stock in the Sample 103 ...................................................................... Average Returns on UK Fama and French's (1993) Factors 103 ......................... Fama and French's (1993) Three-Factor Model 104 ........................................... Petkova's (2006) Model of ICAPM : Five Risk Factors VAR 105 ................... Petkova's (2006) Model of ICAPM: Five Risk Factors - Kalman Filter 106 ...... Petkova's (2006) Model of ICAPM: Seven Risk Factors - VAR 107 ................. Petkova's (2006) Model of ICAPM: Seven Risk Factors - Kalman Filter... 108
Estimated Factors as Risk Factors - VAR ..................................................... 109

Table 2.9 Estimated Factors as Risk Factors - Kalman filter 110 ....................................... Table 2.10 Estimated Factors as Risk Factors and Fama and French's (1993) HML and SMB .................................................................................................. 111 Table 2.11 Estimated Factors as Risk Factors and Fama and French's (1993) HML and SMB-Kalman filter ........................................................................... 113 Table 2.12 Estimated Factors as Risk Factors and HML and SMB-Kalman filter........ 114 Table 2.13 Estimated Factors as Risk Factors as Augmenting Factors 115 ........................... Table 3.1 Cross-Sectional Regression with OLS Daily Idiosyncratic Risk 180 .................... Table 3.2 OLS Daily Idiosyncratic Risk Portfolios 181 ......................................................... Table 3.3 OLS
Daily Idiosyncratic

Table Table Table Table Table Table Table

Table 3.12 Avramov and Chordia's (2006) Time Varying Beta Fama and French's

3.4 Cross-Sectional Regression with OLS Monthly Idiosyncratic Risk 184 ............... 3.5 OLS Monthly Idiosyncratic Risk Portfolios 185 .................................................... 3.6 OLS Monthly Idiosyncratic Risk Portfolios - Double Sorting 186 ....................... 3.7 Cross-Sectional Regression with EGARCH (1,1) Idiosyncratic Risk 188 ............ 3.8 EGARCH (1,1) Idiosyncratic Risk Portfolios 189 ................................................ 3.9 EGARCH (1,1) Idiosyncratic Risk Portfolios - Double Sorting 190 ................... 3.10 Avramov and Chordia's (2006) Time Varying Beta Fama and French's (1993) Model with OLS Monthly Idiosyncratic Risk 192 ............................. Table 3.11 Avramov and Chordia's (2006) Time Varying Beta Fama and French's (1993) Model with EGARCH (1,1) Idiosyncratic Risk 193 ..........................

Risk Portfolios -Double

Sorting

.............................

182

(1993) Model with Idiosyncratic Risk Portfolios 194 .................................. Table 4.1 Descriptive Statistics 257 ....................................................................................... Table 4.2 Annual Excess Returns on Risk Characteristic Based Portfolios 258 .................... Cross-Sectional
Table 4.3 Regressions

Table 4.6 Annual Excess Return on Risk Characteristics Based Portfolios in Expansio n Table 4.7 Cross -Sectional

261 Table 4.4 25 Size and CAPM or Downside Beta Portfolios ........................................... 262 Table 4.5 25 Book-to-Market and CAPM or Downside Beta Portfolios ........................ and Recession..........................................................................................................263
Regression in Expansion and Recession 265 .............................

...........................................................................

260

Table 4.8 Annual Excess Return on Risk Characteristics Based Portfolios within each 266 Industry .................................................................................................................... 268 Table 4.9 Cross -Sectional Regression within each Industry .........................................

Table 4.10 Future Annual Excess Return on Past Risk Characteristics Based Portfolios 271
272 Table 4.11 Does Past Information Predict Downside Beta? ........................................... Table 4.12 Annual Excess Return on Risk Characteristics Based Portfolios - Monthly Frequency 273 ................................................................................................................

List of Figures

Figure 2.1 Average Returns on HML and SMB 116 .............................................................. Figure 2.2 Fitted Versus Actual Returns on the UK Fama and French (1993) 25 Portfolios

.................................................................................................................................

117

Figure 4.1 Changes in UK Coincident Index 274 .................................................................. Figure 4.2 Relationship Between Risk and Returns 275 ........................................................ Figure 4.3 Relationship Between Risk and Returns Over the Business Cycle 275 ................

Chapter 1 Introduction

1.1

Introduction

"Theorists

develop models with

testable predictions;

empirical

researchers this

document "puzzles"-stylized

facts that fail to fit established theories-and

stimulates the development of new theories." (Campbell, 2000, p. 1515)

This relationship between the theories and the empirical data is what this thesis attempts to study. It examines, in the light of the recent issues and developments in the area of asset pricing to what extent stock returns are explained by the theory. It also aims to study what findings are spurious and what puzzles remain and therefore how much support can be offered to the current literature. The relationship between stock returns and risk is of particular interest not only for researchersbut also because it is at the very heart of all investment decisions. Therefore, this research is not only important from a theoretical point of view but it also matters to investment decision makers.

No researcher in finance can deny that the theoretical

models of Sharpe (1964),

Merton (1973), Ross (1976) and Breeden (1979) are important. However, Schwert (2003, p. 964) notes "Many people have developed extensions of theoretical asset-pricing models that include multiple factors ......, empirical Fama-French although none of these models match closely with the

model". Despite the fact that Schwert's (2003) statement looks (1993) three-factor model, the asset pricing

like good news for Fama and French's

determinants Campbell describes being (2000) the with of as concerned models, which the risk premium, are facing challenges ahead. These challenges are posited by Campbell

10

(2000) and Cochrane (2006) as being the requirement to unearth the association between the macroeconomic variables and stock returns.

1.2

Macroeconomic

Factors and Fama and French Asset Pricing

Model

The capital asset pricing model (CAPM) of Sharpe (1964) has been questioned in terms of its underlying assumptions as well as its empirical application as pointed out by Merton (1973) and Fama and French (2004). Roll (1977) points out that providing that the theoretical market portfolio has not been identified the model could not be assessed empirically. Fortunately, APT of Ross (1976) overcomes this problem as indicated by Roll and Ross (1980). Roll and Ross (1980) point out that the APT does not need the market portfolio and is empirically testable. Burmeister and McElroy (1988) state that the risk factors in the APT could be either statistically based or macroeconomic factors. They also point out that the latter has the benefit of relating stock returns to the wider economy. Furthermore, Priestley (1996) argues that the first route lacks economic meaning'.

As mentioned earlier, Merton (1973) refers to the theoretical

shorcomings

of the

CAPM and points out that its static nature is not realistic. Merton (1973) developed an intertemporal capital asset pricing (ICAPM) model and indicates that it is important as it shifts that are ignored by the CAPM.

accounts for the future investment opportunities'

For the statistically based factors (see (Roll and Ross (1980), Connor and Korajczyk (1986,1988) and Jones (2001)) and for the macroeconomic factors (see Chen et at., (1986), Priestley (1996) and Antoniou et at, (1998)).

However, Breeden (1979) points out that while this intertemporal aspect of the ICAPM is crucial, the model's applicability is questionable as it measures risk with multi-betas that are associated with unknown state variables. He developed the consumption asset pricing model. And he argues that this model overcomes the ambiguity regarding the risk factors as it replaces the multi-betas with a single consumption beta. The standing of this model is described by Cochrane (2001, Ch (2)) who points out that although the consumption asset pricing model is perfect theoretically, its poor empirical performance prompts the
need for other models. Nevertheless, Cochrane (1996) states that the consumption model performs unsatisfactorily based

in the empirical applications and this could be caused by, data. Campbell (1993) also points to

among other things, problems with consumption consumption data problem.

On the empirical side, Campbell (2000) indicates that the empirical financial anomalies that defy the CAPM include the small size, value and momentum effects. He points out that that potential causes for such findings that have been put forward in the literature to date include the failure of market proxy, spurious findings, mistakes and in be findings biases. further He that these explained also comments could psychological a rational multifactor model such Fama and French (1993) three-factor model or the ICAPM of Merton (1973).

Fama and French (1993) developed a three- factor model that includes in addition to fact factors. SMB Despite HML that they report a the the and as risk market portfolio, in for in describing their the stock price they admit model changes strong performance

12

that there are problems.

Fama and French (1993) state that as these two factors are

lack the and empirically with of a theory to support them, any explanation will selected never be ultimate.

Furthermore, Fama and French (1995) point out that they support the rational story behind the size and value factors as these factors are associated with profitability. However, they indicate that the state variables behind these factors still need to be established. Furthermore, Fama and French (1996) point out that they support that Fama but factors line French's factors in ICAPM (1993) the state variables the model of and are
need to be known concretely. Cochrane (2006) states that the economic drivers of the

factors French Fama the the size and and value on marginal value of wealth and returns need to be known. He further points out that this could be achieved only through macroeconomic models. Among the studies that meet such a challenge is Petkova (2006) dividend innovations includes to that market return and excess who suggests a model her interest default She that term reports one month rate. yield, spread, spread and selected state variables capture Fama and French's (1993) SMB and HML in the context of the ICAPM and her model outperforms the Fama and French's (1993) three-factor disappear influence SMB HML the when she adds them on stock returns and and model in her model.

However, Petkova (2006) points out that the variables she chooses are not the only factors. be investors information in there the other useful risk and could set of variables Therefore her model leaves open the possibility that other macroeconomic factors are

13

ignored in the analysis even though they are important. Fortunately, the dynamic factor model method of Stock and Watson (2002a, b) opens the opportunity for asset pricing to exploit many macroeconomic variables to help answer the challenges that are posited by Campbell (2000) and Cochrane (2006) who stress that those macroeconomic variables which are responsible for the behavior of stock returns must be understood. Furthermore, as cited earlier, Cochrane (2006) points out nothing other than macroeconomic models can give understanding to the performance of Fama and French's (1993) factors. Studying this relationship between the economic forces and Fama and French's (1993) factors is the first objective of this thesis.

On the dynamic factor approach applications in asset pricing side, Mnch (2004) points out that he uses the dynamic factor model method of Stock and Watson (1998, 2002a) on large macroeconomic variables to employ the resulting estimated factors as risk factors which he describes as a new approach for uncovering the risk factors. He indicates that he compares his model with Fama and French's (1993) three-factor and Campbell's (1996) models and also augments the latter model with the estimated factors to test if the estimated factor adds additional information. However he does not provide
an answer to whether these estimated factors capture Fama and French's (1993) factors.

In addition he uses the estimated factors while this thesis employs the innovations in the estimated factors. Furthermore, Ludvigson and Ng (2007) points out they employ the dynamic factor models with large dataset to provide a solution for the omitted risk factors
in the conditional information set.

14

1.3

Idiosyncratic Risk and Time Varying Betas

Another important recent topic in asset pricing is idiosyncratic his CAPM model Sharpe (1964) states that as a result of diversification

developing In risk. the unsystematic

part of the total risk does not affect the asset's return. In spite of this, Merton (1987) develops a model in which stock return depends positively as its specific risk in a market where information on its systematic risk as well

is incomplete. Furthermore, Malkiel and idiosyncratic risk

Xu (2006) point out they derive a model that includes the undiversified as a potential idiosyncratic risk factor because in a world

of no market portfolio,

undiversified

risk is a priced factor.

Empirically,

Malkiel

and Xu (2006), Spiegel and Wang (2005), Chua, Goh and relationship between stock returns and report a negative

Zhang (2007) and Fu (2007) report a positive idiosyncratic idiosyncratic risk, while Ang, Hodrick,

Xing and Zhang (2006,2008)

risk effect on stock returns. Ang, Hodrick, Xing and Zhang (2008) point out

that this is a puzzling global finding whose sources need to be studied. On the other hand, Huang et al. (2006) indicate that there is no cross sectional idiosyncratic risk influence on the stock returns and the negative effect arises from the return reversal. Furthermore, Bali and Cakici (2008) state that becauseof these confusing findings their objective is to study whether idiosyncratic risk effect is genuine. They report that it is not robust and it is a matter of differences in applied methods used by these studies.

From another perspective, Chen and Keown (1981) point out that when timevariation in beta is not accounted for then residual risk from OLS will not be pure as it

15

will reflect this variability and will be heteroskedastic. Ang and Chen (2007) also refer to this issue and they call the part that is due to time-varying beta and appears in the residuals an omitted variable. Furthermore, Malkiel and Xu (2006) point out that the residuals from any pricing model could be representative of omitted factors and other effects. Therefore, the second objective of this thesis is to attempt to contribute to idiosyncratic risk literature by examining whether accounting for time-variation in betas by applying Avramov and Chordia's (2006) conditional model and methodology could explain the mystery of idiosyncratic risk in the cross section of returns and hence support
Bali and Cakici (2008) and others who provide evidence that idiosyncratic risk is not

significant.

1.4

Downside

Risk and Business Cycle

Sharpe (1964) points out that the mean variance model might be inadequate for some circumstances and he argues that Markowitz indicates that semivariance based model could be favoured. However, Bawa and Lindenberg (1977) develop a model

which they call the mean - lower partial CAPM that uses the lower partial moment instead of the variance. They point out that their model implies the CAPM when stock returns are normally distributed.

More recently, Ang Chen and Xang (2006) propose a downside risk model with
disappointment aversion utility maximizers. They point out that these investors are

downside the changes of stock prices. They report a significant crossabout anxious

16

sectional price for downside risk. In addition, Post and Vliet (2005) report that they find downside beta, conditionally and unconditionally, is a superior cross sectional risk factor to market beta. They report that they find the superiority of mean-semivariance CAPM occurs, in particular, when the economy in its bad states but it is not that strong over Ang Chen and Xing's (published later as Ang Chen and Xing (2006)) sample period. Post and Vliet (2005) point out that the latter study's downside risk measure is questionable and it does not use conditional tests of downside risk. Therefore, the third objective of this thesis is to examine the effect of a time-varying (conditional) risk premium in Ang Chen and Xing's (2006) downside risk model, over the business cycle on the relationship between downside risk and stock returns.

1.5

Issues and Contribution

The above cited literature demonstrates the current standing of asset pricing models and highlights the unresolved issues that remain to be addressed. Based on this, the objectives of the chapters of this thesis are outlined below.

Chapter

(2) starts by reviewing

the relevant

literature

related to asset pricing

models and dynamic factor models. Based on this and motivated by Campbell's (2000) between demand for (2006) Cochrane's the the understanding of relationship and SMB French's (1993) Fama and and and and stock returns macroeconomic variables HML, the second chapter examines if the macroeconomic variables are able to explain the cross section of UK stock returns and whether they can capture Fama and French's

17

(1993) SMB and HML. Therefore, this chapter applies Petkova's (2006) model that includes excess market returns, innovations to dividend yield, term spread, default spread and one month interest as well her model that includes, in addition to these five variables, the HML and SMB. Then instead of confining the analysis to a small number of macro factors, it moves to augment Petkova's (2006) model (keeping only factors that are found significant in the UK) with innovations to factors estimated from a large panel of macroeconomic variables by the dynamic factor model method of Stock and Watson (2002a, b).

The results show that innovations in dividend yield, term spread and one estimated factor (relates to unemployment and term spread) and the HML and SMB are significant factors in the UK market. However, this chapter finds no association between the first three factors and the Fama and French's (1993) SMB and HML. This chapter contributes to the literature by using innovations in estimated factors from large macro variables to
examine whether they can capture the HML possibility of finding and SMB. The results cast doubt on the and SMB in the

further macro factors that can replace the HML

UK. As mentioned earlier, Mnch (2004) uses estimated factors from a large set of macro variables and augments Campbell's (1996) model with these estimated factors, but this

link factors differs in innovations to them to the to estimated and attempts chapter using the HML and SMB.

Chapter (3) starts by reviewing the relevant literature related to idiosyncratic risk importance betas time-varying to the of studying and establish and stock returns

18

idiosyncratic risk. Based on this and motivated by the indecisive findings regarding idiosyncratic risk ability to explain stock returns. This chapter examines idiosyncratic risk in the cross-section of UK stock returns. Bali and Cakici's (2008) attribute idiosyncratic risk effect to methodological issues and conclude that idiosyncratic risk is insignificant. Furthermore, there are findings that idiosyncratic risk effect is significant in the UK as found by studies such as Angelidis and Tessaromatis (2008a), Ang, Hodrick, Xing and Zhang (2008) and Fletcher (2007). This chapter examines idiosyncratic risk in the UK market by following mainly Ang, Hodrick, Xing and Zhang (2006,2008) and Spiegel and Wang (2005). Then it examines whether accounting for time-variation in beta by applying Avramov and Chordia's (2006) conditional model and methodology can capture the effect of idiosyncratic risk. Avramov and Chordia (2006) point out that what distinguishes their methodology is that they model beta explicitly.

These results show, consistent with the literature, confusing findings regarding the significance of the relationship between idiosyncratic risk and stock returns. However when Avramov and Chordia's (2006) conditional model and methodology are applied, the results show idiosyncratic risk is insignificant in explaining the cross sectional returns in the UK market. This conclusion is consistent with Bali and Cakici (2008). This chapter literature by idiosyncratic the to supporting those who stress the risk of contributes importance of time-varying beta as a potential explanation to the confusing evidence be It idiosyncratic to this that could attributed evidence confusing risk effect. shows about Chordia's (2006) in Avramov beta and time-varying as explicitly modelled not using beta does however This time-varying the of effect not study chapter conditional model.

19

on idiosyncratic risk which is calculated using daily returns following Ang, Hodrick, Xing and Zhang (2006,2008).

Chapter (4) similarly commences by reviewing the relevant literature related to downside risk and time varying risk over the business cycle. The fourth chapter is motivated by the encouraging downside risk findings of Ang Chen and Xing (2006) and Post and Vliet (2005) and the criticisms the latter study made concerning Ang Chen and Xing' (later published as Ang Chen and Xing (2006)) methodology, which is in line with the importance of time-varying risk. Added to this, are the findings of Pedersen and Hwang (2007) and Olmo (2007) regarding the potential importance of downside risk in UK. Therefore, the fourth chapter examines if the significant role of downside risk of Ang Chen and Xing (2006) is also important in the UK cross section of returns and whether these findings are strengthened or weakened by allowing for downside risk return relationship to vary over the business cycle. More specifically, it studies whether their findings hold only over particular phases of the business cycle, as argued by Post and Vliet (2005), rather than all over the business cycle. Consequently, this chapter first
applies Ang Chen and Xing's Vliet's (2005) conditional (2006) study to the UK market. Then it follows Post and

test approach which includes dividing

the full sample period (2006) model

into recession and expansion and then reapplying and methodology

Ang Chen and Xing's

over the recession and expansion periods one by one. The Economic dates of the business cycle are used to divide the sample into Antoniou et al., (2007).

Research Cycle Institute's

recession and expansion following

20

The results show that Ang Chen and Xing's (2006) downside beta is priced in the
UK market, although it has a problem pricing the riskiest stocks. Furthermore, downside beta and CAPM beta show, to a large extent, similar performance unconditionally. are

However, when Ang Chen and Xing's (2006) downside risk model and methodology

applied separately over the expansion and recession periods rather than over the full sample period as before, the results show downside beta excels over CAPM beta in

explaining the cross sectional changes in stock prices over expansion, but both measures of risk have inconclusive results over the recession period. Also the findings show some
support for Pedersen and Hwang's (2007) and Olmo's (2007) studies.

21

Chapter 2 Macroeconomic Factors and Fama and French Asset Pricing Model

22

2.1

Introduction

"In sum, the program of understanding the real, macroeconomic risks that drive asset prices (or the proof that they do not do so at all) is not some weird branch of finance; it is the trunk of the tree. As frustratingly slow as progress is, this is the only way to answer the central questions of financial economics"(Cochrane, 2006, p.6).

This chapter addresses this issue by examining the link between macroeconomic factors and stocks prices, the importance of which is stressed by Cochrane (2006) and Campbell (2000). The keystone asset pricing models in the asset pricing literature are the capital asset pricing model (CAPM) of Sharpe (1964), the arbitrate pricing theory (APT) of Ross (1976), the intertemporal capital asset pricing model (ICAPM) of Merton (1973), the consumption-based asset pricing model (C-CAPM) of Breeden (1979), and then Fama and French (1993) developed their three factor model.

Unfortunately the situation of the latter model compared with first four models is summarized by Cochrane's (1999, p.40) statement "In general, empirical success varies inversely with theoretical purity". Fama and French (1993) admit that as the additional
two factors in their model; the size (SMB) and the book-to-market empirically, with no theoretical support, the interpretation (HML) were chosen

of such factors is debatable. In

fact connecting SMB and HML to macroeconomy

is one of the most important issues in behind this

current research in this area (Cochrane, 2006) and one of the motivations chapter.

23

In an attempt to legitimatize their model as a rational asset pricing model, Fama and French (1996) argue that the SMB and HML factors could be interpreted as proxies for multifactor-minimum-variance portfolios in a two-state ICAPM of Merton (1973), that

could be related to relative distress. In addition, they reported that the three-factor model is able to explain variation in portfolios' returns constructed from sorting stocks on a number of firm characteristics, but it fails the Jegadeeshand Titman's (1993) momentum effect. Despite this success they point out that they have not yet discovered, unquestionably, the ICAPM's state variables or the APT's factors that would support their findings in the context of Merton's (1973) model or Ross's (1976) model, and hence a number of explanations is still feasible. This is also stressed by Lewellen (1999) who points out that the risks factors that underlie the size and book-to-market portfolios must be known for a complete rational explanation to be attained and Hahn and Lee (2006) indicate that the above statement of Lewellen (1999) is still valid. Furthermore, Lettau and Ludvigson (2001) argue that Fama and French's (1993,1995) interpretation that the three-factor model captures common risk factors, is debatable as the association between the macroeconomic risk factors and the size and book-to-market factors is still to be clarified.

A way to resolve the controversy surrounding the HML and SMB was suggested by Cochrane (2006). He points out that asset pricing models that use portfolios as risk

factors can do the job of pricing stocks but cannot explain why portfolios HML are so priced and only macroeconomic Liew and Vassalou (2000)

of SMB and

models can come up with the explanation. and SMB are related to future

reported that the HML

French's findings Fama (1992,1993, that these and and argue support growth economic

24

1995,1998) Merton's

risk story that these variables proxy for the ICAPM's

state variables of

(1973). Lettau and Ludvigson

(2001) reported that the higher nondiversifiable

risk of value portfolios

can, in part, explain the value premium. Recently, Petkova (2006) and SMB

reported that these debatable Fama and French's (1993) HML (book-to-market) (size) portfolios are proxies for risk factors which are the innovations

to aggregate She points set.

dividend yield, term spread, default spread and the one-month Treasury-Bill. out that these four2 variables predict future changes in the investment Furthermore, Hahn and Lee (2006) report support to the risk story as well.

opportunity

However, these studies have used only a subset of the potential actual information set that is available to investors. Therefore it keeps open the question of whether the employed risk factors are the only priced factors or if there are missing macroeconomic risk variables that are able to capture the variation in stock returns that need to be considered and whether these missing variables are able to capture the HML and SMB performance. In fact, Lettau and Ludvigson (2001) address this issue of potential missing variables. They point out that the information set observed by the researcher is only part of the set used by the investor. They indicate that they overcome this problem by
choosing a conditioning factor that captures the expectations in the market. Furthermore, in the

Ludvigson and Ng (2007) point to what they call a problem of omitted-information context of the conditioning employing information,

and indicate that they overcome this problem by Mnch

a large panel of data using the dynamic factor models. Additionally,

(2006) employs the dynamic

factor models with large set of data and reports that the

2 Petkova

(2006)

suggests a model

that includes

excess market

returns

and innovations

to aggregate

default dividend Treasury-Bill. spread andtheone-month yield,termspread,

25

for factors the information used by investors. Mnch (2004) points out account estimated that he uses the factors estimated using the dynamic factor model method of Stock and Watson (1998,2002a) on large set of macro variables as potential risk factors and also points out that he augments Campbell's (1996) model with the priced estimated factors to emphasize the added information by these factors.

This chapter contributes to the literature of asset pricing by providing another investigation of this potential problem. This problem has arisen because the potential existence of risk factors that are important for explaining the returns on stocks are missed when only a few sets of standard risk factors that are used extensively in the literature are factors) Chen (2003) (i. that this e. missing risk could potential problem chosen. points out be behind his model's rejection to the rational explanation for the book-to-market effect. This is important as the objective of this chapter is first to respond to Campbell's (2000) and Cochrane's (2006) demand to find the macroeconomic variables that are responsible for stock's risk premia. The second aim is to respond to Cochrane's (2006) challenge to find the macroeconomic factors that are tracked by Fama and French's (1993) HML and SMB factors.

A recent paper that takes on these challenges is Petkova (2006). She suggests a

model that includes market excess returns and innovations to variables that are state if Fama in Merton (1973) ICAPM the the examines and and context of of variables French's (1993) factors fall within ICAPM interpretation by relating them to the surprises to the four state variables. Furthermore, Petkova (2006) mentions that in response to the fishing license criticism of the ICAPM that is made by Cochrane (2001) and Fama

26

(1991), she selects as state variables those that have the ability to predict the future investment opportunity set. Also she points out that Cochrane (2001) criticize studies that do not check the predictive ability of their factors. Therefore, this chapter starts by applying Petkova's (2006) study and methodology to the UK market. This provides an out-of-sample test for her study.

However, Petkova's (2006) study does not rule out the potential existence of other macroeconomic risk factors that are valuable for pricing stocks and these need to be taken into account in order to respond satisfactorily to the challenges put forward by Cochrane (2006) and Campbell (2000). Indeed Petkova (2006) acknowledges this shortcoming and states that there could be other useful information that is used by market participants to predict the changes in the investment opportunity set. But she concludes that based on her findings the state variables that she has employed appear to be appropriate pricing variables.

This chapter overcomes this shortcoming and augments her model with other possible predictive macroeconomic variables to examine if there are missing risk factors that are not captured by her chosen state variables and/ or if other factors are required to
capture the returns on the HML and SMB. In extending the possible set of state variables, care needs be taken regarding the variables chosen as additional potential risk factors.

Fama (1991) accuses the asset pricing models with multiple factors of being licenses to look for factors explaining the stock returns that already have been found. However, Cochrane (2001, Ch (9)) points out that ICAPM is not exactly such a license as state have be factors. have He the those that they to predictive variables are states variables

27

ability to predict stock returns or economic variables. In addition, Cochrane (2006) indicates that for the state variables in the ICAPM model to affect stock's return, they have to be able to forecast the return on the market and consumption.

Therefore,

in order not to fulfill

Fama's (1991) fears regarding

the multifactor

models, the dynamic factors model of Stock and Watson (2002a, 2002b) is utilized in this chapter. The relevance of this model to the objective of this chapter comes clear from understanding what it does. Stock and Watson (2002a) employ the method to forecast macroeconomic series. They point out rather than choosing a few potential predictive

variables to forecast macroeconomic variables, the approximate factor model summarizes the information contained in a large set of variables in a few factors that are estimated by component analysis. They reported that a great deal of the

method of the principal variability factors.

in the examined large set of macro variables is captured by few estimated Therefore this chapter examines if such estimated factors that are found to activity can be additional risk factors in the context of the

forecast the macroeconomic

ICAPM and if they bear any relation to the Fama and French's (1993) HML and SMB. In sum, this chapter aims at addressing the challenges to asst pricing models, that have been stressed and discussed by Campbell (2000) and Cochrane (2001,1999,2006) macroeconomic to relate the

forces to stock returns and Fama and French's (1993) factors in the UK.

On the estimated macro factors as potential risk factors front, Mnch (2004) pointed out that he studied whether a few factors estimated from a large dataset of macroeconomic variables could explain the cross-sectional returns on Fama and French's (1993) 25 size and book-to-market portfolios. He indicated that he developed a model,

28

that he calls the "diffusion index pricing model", which employs the common factors estimated from a large macroeconomic dataset, using the method of Stock and Watson (1998,2002a) as risk factors. He states that his motivation is examining whether the common factors that drive the macroeconomy also drive stock returns. Mnch (2004) points out that he compares his model with a number of models including Campbell (1996) and Fama and French (1993) three-factor model. Furthermore, he points out that to test for additional information provided by these estimated factors, he augments Campbell's (1996) model with the estimated factors. However this chapter differs from his in number of ways. First, he developed a theoretical framework for his model, while this chapter makes no assumptions but rather uses the innovations to the estimated common factors as potential risk factors. In contrast he uses the estimated common factors themselves as risk factors. This is important as Merton (1973) states that in the ICAPM world, stock return is determined by market risk and the risk that arises from the changes in the investment opportunities where the latter could be described by at least interest in is the the rate. Furthermore, Brennan et al one state variable which changes (2004) emphasize this nature of the ICAPM and indicate that the risk factors in the
ICAPM are innovations to state variables that describe the investments set and not only

factors that are related to stock returns. Second, this chapter examines whether the innovations to these estimated factors are able to absorb the explanatory power of the

Fama and French's (1993) HML and SMB, while he merely compared the performances HML SMB is despite fact the the the the two that and one of models, capturing effect of of the main issues in asset pricing (2001,1999,2006) that needs to be solved as discussed by Cochrane

among others. Third, this chapter is applied to the UK whereas he

factors (2004) USA. Furthermore, Mnch his treated the to the as risk while study applied

29

possible state variables, this chapter treats the innovations in the estimated factors as potential state variables, and as in his study it uses Stock and Watson's (2002a, 2002b) dynamic factor model and methodology to estimate the common factors.

The second contribution of this chapter is providing an out-of-sample test of Petkova's3 (2006) study. Fletcher (2007) applies Petkova's (2006) model to the UK. He points out that he uses Petkova's (2006) and Fama and French's (1993) three-factor model among other models to compare their pricing ability of idiosyncratic and systematic risks on two set of UK stocks return; industry portfolios and cluster portfolios. However, the aim of this chapter is to examine whether the risk factors in Petkova's (2006) model that she found to be priced risk factors in the US market and are able to
capture the value and size premium of the Fama and French's (1993) HML and SMB

factors, deliver similar results in the UK market; i.e. to be priced as risk factors and capture the Fama and French's (1993) HML and SMB in the UK. This is important as Griffin (2002) reported that a version of the Fama and French's (1993) three-factor model that includes country-specific factors is better at explaining the cross section of returns

than its global version counterpart.In addition he reportedthat the correlation between
the US market excess returns and the UK market excess returns is 0.68 and the

correlation between the US SMB (US HML) and UK SMB (UK HML) are 0.15 and 0.27
respectively. Griffin (2002) pointed out these low correlations opposed with what was and SMB

expected under the assumption of similar

state variables underlie the HML

portfolios across integrated markets. Therefore it is important for the objective of this

3 Cochrane (2006) pointed out that Campbell (1996) is among few studies that ensured that their selected risk variables are along the line of the ICAPM of Merton (1973). Also Cochrane (2006) pointed in this regard to Petkova (2006) as well.

30

chapter to examine the wider set of macroeconomic variables and not restrict the analysis to a few selected potential risk factors.

The rest of the chapter is organized as follows: section 2.2 includes a review of the relevant literature, section 2.3 states the hypotheses, section 2.4 discussesthe data and the methodology, section 2.5 presents the results and discusses the empirical findings and finally section 2.6 concludes.

2.2

Literature

Review

At the millennium, Campbell (2000) argued that understanding what economic forces drive the reward for risk is the challenge that is facing asset pricing. Also Cochrane (2006) pointed out that understanding those macroeconomic factors that drive the price of risk and the premiums on the size and value portfolios is the challenge. Inspired by these, the current chapter studies the macroeconomic determinants of stocks prices in general and the Fama -French HML and SMB in particular.

Stochastic discount factor, multifactor models and the macroeconomic variables are linked together as shown by Campbell (2000) and Cochrane (2001,1999). Campbell

(2000) indicates that the stochastic discount factor (SDF) is useful for understanding the multifactor pricing models. He writes the asset pricing equation as (Campbell, 2000, p. 1517, Eq. 1): PI, = E, [m,+1X1,1

31

Equation (1) above is in Campbell's (2000) notations. He defines the notations in the above equation as follows; p,, is stock's i price at t time, E, is the operator for

conditional expectations, m,,, is the SDF4 and X,.,,, is stock's i random payoff at t+1.

He derives the risk premium on stock i as (Campbell, 2000, p. 1520, Eq.5): RJ., (m, R,., cov, +1) +,, +, E, m, +, (2)

Equation (2) above is in Campbell's (2000) notations. He points out that this equation means that the risk premium on risky stock is calculated by the negative cothe m, divided by risk-free asset's variance of its excess returns (R,,,,, - R, with +) +, price (E, m,+,). He explained this as stocks that produce low returns in bad states of the economy, when investors have high marginal utility must be rewarded with high risk premiums.

Cochrane (2001, Ch (9)) argues that the consumption-based model is a theoretically sound asset pricing model but does not produce good empirical performance, hence, searches have been made to link the stochastic discount factor with other variables, such
as in the factor pricing models. He shows clearly the link between the SDF equation and

the CAPM, APT, and ICAPM. Furthermore, Cochrane (2001, Ch (9)) points out that the variables that should be employed in the SDF as risk factors need to be searched for. He further indicates that asset pricing models seek risk factors that signal the incidence of the bad times of the economy. Cochrane (1999) identified such factors as those that are
' Cochrane (2001, Ch(1)) pointed out that other names for the SDF includes the marginal rate of substitution and pricing kernel.

32

associated with consumption which include; market return (CAPM), state variables and events of eroded noninvestment-based income, and fourth group is portfolio returns which could be seen as proxies for one of these three groups. Nevertheless, he stated that theory and empirics are in disagreement. He indicated that the Fama and French's (1993) SMB and HML factors are examples of such portfolios, but he acknowledges that it is not entirely obvious what macroeconomic risks are behind these two factors.

In summary, Cochrane (2001,1999,2006)

and Campbell (2000) have summarized

the challenges that asset pricing faces which are establishing the link between the macroeconomic risks and stock returns and Fama and French's (1993) SMB and HML. This chapter attempts to address these issues, but first its reviews the relevant literature that matters to this study.

2.2.1

Fama and French's

(1993) Three-Factor

Model

Fama and French (1992) examined the ability of a number of variables along with
the CAPM beta to explain the cross sectional returns on stocks. They pointed out that

these chosen variables constitute a challenge to the CAPM and include the size, the bookto-market ratio, the leverage, and the earnings to price ratio. They reported that their

findings could be interpreted as size and book-to-market 5 are potential proxies for risk factors. More specifically, variation they reported that beta fails to explain the cross-sectional succeed in this task and even

in stock returns while size and book-to-market

5 Chen (2003) points out that there is a recent support to the risk story. However, he reports, based on his model, a rejection for such an explanation to be underlying the book-to-market value effect. Nevertheless, he points out that other state variables that may be used by investors need to be considered.

33

leverage be both found drive the to they to out and more earnings price ratio which are individually significant. Fama and French (1993) developed a three-factor model which includes market portfolio, size portfolio - SMB (the difference in return between small and large stocks' portfolios) and book-to-market portfolio - HML (the difference in return between value and growth stocks' portfolios). They pointed out because they aim to examine if the SMB and HML are able to explain the cross sectional returns on stocks that are associated with size and book-to-market characteristics, therefore, they

constructed 25 portfolios of stocks according to their size and book-to-market values to be the testing assets. They reported that their three-factor model is able to explain the variation in stock returns and pointed out that these two factors are proxies for nondiversifiable risk factors.

Fama and French (1996) reported that the three-factor model is able to explain the variation in the cross section of portfolios' average returns that are constructed by sorting stocks on other firm characteristics, but it does not succeed in capturing the momentum effect of Jegadeeshand Titman (1993). They state that their results regarding the ability of HML and SMB to explain returns on stocks, are in line with a rational asset pricing's story such as the ICAPM of Merton (1973) or the APT of Ross (1976), however they acknowledged that other stories are still possible.

Ferson, et al., (1999) questioned the reliability

of pricing factors that were

constructed on the basis of attributes that are found to explain the cross sectional HML Fama SMB in the of and and French on an empirical as ground returns variation

34

(1993,1996)6.

They pointed out that such portfolios would have power to explain

changes in stock prices even if they might not be connected to risk. Ferson and Harvey (1999) reported that, using lagged predictive variables, a conditional Fama-French's (1993) three-factor model is rejected and argued that their findings send a strong warning to the use of this model in the risk-returns calculations. Brennan, et al., (1998) studied the individual stocks rather than portfolios and stressed that this was vital in light of Roll's (1977) and Lo and MacKinlay's (1990) criticism of the use of portfolios. They reported that non-risk variables are able to explain stock returns, even though the Fama and French's (1993) HML and SMB factors are accounted for, these include the trading volume (a liquidity proxy) and momentum, while the book-to-market and size variables are weakened.

Davis, et al., (2000) studied the value premium and states that four widespread interpretations were put forward to this phenomenon?including: (1) it is a reward for risk in agreement with the ICAPM of Merton (1973) or APT of Ross (1976), (2) it is a chance and possibly to disappear in other samples, (3) it is a result of overreaction by investors, or (4) it captures the characteristics rather than the risk aspect of the value effect. They studied the last explanation and reported that no matter what the book-to-market value characteristics is, the loading on the HML is what explains returns, and Fama and French (1993) model performs better in capturing the value premium than the characteristics model. They explained that the reason for Daniel and Titman's (1997) findings that

6 Ferson, et al., (1999), in their footnote (2), summarize the potential explanations for the HML and SMB performance. In addition, Fama and French (1998) provide summery of the potential explanations for the value premium. ' See Davis, et al., (2000), for more details on the relevant literature relate to these issues.

35

support the characteristics model is that these are peculiar to the short sample period of their study.

Liew and Vassalou (2000) pointed out that their findings, based on data of ten countries, is consistent with the risk-explanation story of the HML and SMB in line with ICAPM of Merton (1973). They reported that these two factors link positively to the future economic growth, a finding that is sustained in the existence of business cycle variables (short-term interest rate, dividend yield, term spread and industrial production). They argued that this finding could be explained as when high growth in the economy is in large term of stocks and growth small and outperform predicted, value stocks prosperity. In addition they reported the HML and SMB carry information that is different from that of the market portfolio. However, they commented that the results

SMB between HML in deciding the the the and relationship suggest country plays a role in due differences future justified to the the they size, accounting as and economy which standard and market capitalization between these countries.

Kelly (2003) reported that Fama and French's (1993) factors are related to in line ICAPM. hence (shocks) the are with state variables and macroeconomic variables He pointed out that he builds on Liew and Vassalou's (2000) study by decomposing the indicated inflation. into He that this permits and growth real growth economy's nominal inflation between SMB HML two these the the parts; and and and relationship studying

8 Liew and Vassalou's (2000) study sample's includes United States, United Kingdom, Netherlands, Germany, France, Switzerland, Italy, Canada, Australia and Japan.

36

real growth. He reported that generally, using data from 18 countries9, there is a negative (positive) relationship between the SMB and unexpected inflation (real growth) and positive relationship between the HML and the future real growth. Hanhardt and Ansotegui (2008) also reported that their findings lend support to the HML and SMB as risk-based factors consistent with the ICAPM for the Eurozone area. Furthermore, they pointed out this consistent with Liew and Vassalou (2000), the Fama and French (1993) factors are predictors of future growth, in particular, the SMB factor.

Other evidence that supports the risk story behind the HML and SMB is provided by Lettau and Ludvigson which the consumption -wealth (2001). They reported a conditional risk factor is scaled by the consumption CAPM, in (log

conditioning

variable

consumption

ratio) capture the value premium.

They pointed out that value

stocks are riskier in times of high risk premium (bad states of the economy) and therefore have higher returns. They indicated that for Fama and French's (1993) factors to be

interpreted as capturing macroeconomic risk then the average cross sectional returns of these factors should be accounted for by macroeconomic factors. They stated that their findings propose that the Fama and French's (1993) SMB and HML factors could be proxies for risk variables that have risk premia that are time-varying.

In addition, Petkova and Zhang (2005) reported a positive (negative) covariation between betas of value (growth) stocks and expected excess return on the market which they interpreted as helping toward explaining the value premium. They stated that their

9 Kelly's (2003) study sample includes, United States, Canada, United Kingdom, Netherlands, Germany, France, Switzerland, Austria, Italy, Sweden, Spain, Denmark, Finland, Belgium, Norway, South Korea, Japan and Australia.

37

ability to find such results when earlier studies failed the task is that they use the expected rather than realized returns. They pointed out that the value premium seems to be explained by the time variation in risk but nevertheless, the conditional CAPM is not able to completely capture the value premium's magnitude.

Furthermore, Petkova (2006) pointed out that the explanation of Fama and French's (1993) SMB and HML are in line with risk story of the ICAPM of Merton's (1973). She reported the HML portfolio is positively related to the innovations in term spread and innovations in default spread while it is negatively related to the innovations in the dividend yield, she stated that, hence HML could be a proxy for duration risk.

Furthermore, she reported that the SMB portfolio is negatively related to the innovations in the default spread and pointed out that SMB could be a proxy for distress risk.

Although the above studies examine the possible explanation of the HML and SMB in the ICAPM of Merton (1973), by attempting to connect them with the macroeconomic variables, the risk story behind the performance of these factors could be supported using other sources as Anderson and Garcia-Feijo (2006) do. They explained that the investment activities of the firm that precedes forming the portfolios, can explain the returns of value stocks and growth stocks. They pointed out that before portfolio formation, growth (value) stocks increase (decrease) their investments, lower (increase) their book-to-market value and experience low (high) return in the following period. They pointed out that their findings support the risk explanation behind the size and book-to-market characteristics.

38

Fama and French (1996) described their interpretation of the HML and SMB portfolios as state variables in the ICAPM of Merton (1973) or risk factors in the APT of Ross (1976) as aggressive. However, in light of the above mentioned promising literature, this chapter is interested in this interpretation of the Fama and French's (1993)
HML and SMB as state variables in the ICAPM of Merton (1973). To shed more light on

the ICAPM of Merton (1973) as an asset pricing model the chapter will proceed to discuss the literature related to this model in the next sub-section.

2.2.2

Intertemporal

Capital Asset Pricing Model

Merton (1973) developed the intertemporal capital asset pricing model (ICAPM) and explained that the single-period capital asset pricing model is a special case of the
ICAPM when the investment opportunities are assumed constant. He pointed out that,

however, the interest rate is stochastic, which is a component of the investment opportunities, and hence the constant set assumption is implausible. He developed an equilibrium model in which the expected return is a function of the exposure to the market risk and the other risks that arise from the changes in the future investment opportunities and stressedthat an important feature of this model relative to the CAPM is that an asset's expected excess return will not be zero if it has zero market risk.

Cochrane (2001, Ch (9))10 wrote the ICAPM equation in the SDF framework where he indicates that the state variables proxy for the consumption. However, Cochrane (2001, Ch (9)) points out becausethe ICAPM model does not specify what are these state
10This i; from Cochrane (2001) chapter 9.

39

variables, researchers exploit this model as a justification to the ad hoc factors that they employ in their studies as Fama (1991) accused them. He comments further that ICAPM is not an open tool as the factors in the context of the ICAPM should be forecasting the future investment opportunity set. Cochrane' 1 (2006) points out that Campbell (1996) (and those who built on his work including Petkova (2006)), Ferson and Harvey (1999) and Brennan et al, (2005) are the only papers that have ensured that their employed factors predict returns on the market.

Chen, Roll and Ross (1986) examined if shocks (surprise) to a number of macroeconomic variables (state variables) are priced in the cross-section of returns. They pointed out that the use of macro variables as potential risk factors is consistent with ICAPM of Merton (1973) and APT of Ross (1976). They reported shocks (innovations) to the risk premia (difference in return on low-grade and high-grade (governmental) bonds), industrial production, term structure (difference between long-term and shortterm governmental interest rates) and unexpected and expected inflation are risk factors as they are able to capture the returns on the stocks. They acknowledged that they have not studied all the potentially priced macroeconomic variables but they argued that their variables seem to be important compared with others. Shanken and Weinstein (2006) challenged the findings of Chen Roll and Ross (1986). They examined the relation between these same five variables and stock returns. They reported that only one factor

11Cochrane (2006) cite Brennan et at (2005) and in his reference appears "Brennan, Michael J., Yihong Xia, and AshleyWang 2005, "Estimation and Test of a Simple Model of Intertemporal Asset Pricing, " Journal of Finance 59,1743-1776". 1 found Brennan, M., Wang, A., and Xia, Y., (2004), "Estimation and Test of a Simple Model of Intertemporal Capital Asset Pricing", Journal of Finance, 59, pp. 1743-1775. This is obviously the same paper that Cochrane refers to.

40

remains significant, which is the industrial production while the other variables are not. They indicated that this change in the significance of the macro variables compared with Chen, Roll and Ross's (1986) findings is a result of the macro variables' factor loadings in their study are estimated using portfolio's returns after ranking dates whilst Chen, Roll and Ross (1986) estimated the factor loadings using portfolio's returns before the ranking dates.

Asprem (1989) reported, using data from different European countries12, a number of macro variables are related to market returns and stated that these macro factors could be state variables in the context of the ICAPM. More specifically, he reported that there is, among other variables, a negative reaction of stock prices to employment, interest rates and inflation. Chen (1991) reported an association between the state variables' predictability of the returns on the market and their predictability of the macroeconomy. He pointed out whilst the current growth in the economy is tracked by the dividend yield and the default spread, it relates negatively to expected returns. Furthermore he stated that whereas the future growth is tracked by the term structure, the Treasury-Bill and the past growth of the industrial production, it relates positively to the expected returns.

Campbell

(1996) developed a model in which he described that asset return as

function of innovations in a three set of factors; (1) market return, (2) predictors of future returns on the market and (3) predictors of future human capital's returns. He pointed out that the second set of factors (2 above) is the state variables of Merton. He reported that

'2Asprem's (1989) study sample includes; United Kingdom, Germany, France, Netherlands, Switzerland, Sweden, Denmark, Norway, Finland, and Italy

41

the overriding pricing factor is market risk and stated that the value of the intertemporal for from its the theory to explanation give an comes ability view of asset pricing importance of the market return as a risk factor in asset returns. He clarified that this importance stems form the market return's association with other two factors (2 and 3 above) and not just being part of the investor's wealth.

Furthermore, Ferson and Harvey (1999) pointed out that the ICAPM is one of the likely successors to the empirically failed CAPM, although the empirical findings are disappointing. They reported that a number of lagged macro variables which include the dividend yield, term spread, default spread and a measure related to short-term interest rate, are able to capture the stock returns' variation. Vassalou (2003) reported that the GDP future news is important for pricing stocks and captures the HML and SMB pricing power. She pointed out that a pricing model, which includes proxy for future GDP's news and market returns as risk factors, is consistent with the ICAPM in which investors are hedging against the state variable's risk.

Supporting Protopapadakis

the importance

of the macroeconomy

to stock returns, Flannery and (1973), Ross

(2002) indicated that risk factors, in context of Merton

(1976) and Breeden (1979) could be powerfully

proxied by macro variables, although, ground. They examined the

they have not received the expected support on empirical importance of the announcements on 17 macroeconomic

variables for stock market and

They in is largest this the that this pointed out that they context. set ever used pointed out found six variables are potentially important risk factors. More specifically role of employment, they reported housing starts

that they uncovered new evidence of the influential

42

Furthermore, balance they the trade the stock returns. on of of conditional variance and is influenced PPI, by CPI the and whilst money supply and market returns reported influences the level and the conditional variance of returns, the industrial production and GNP are not significant.

Brennan, Wang and Xia (2004) indicated Merton's (1973) ICAPM has an important feature which is, its state variables are innovations in predictive factors of the investment in developed factor. They ICAPM model which the an opportunities and not merely any interest Sharpe the the are rate. They reported that these state real state variables ratio and ICAPM in their that the model of returns and are priced cross section variables in CAPM Fama French (1993) the three-factor the a model and model outperformed and number of exercises.

Petkova (2006) proposes a model that includes excess market returns and innovation to dividend yield, term spread, default spread and short term interest rate as a four (1973). She Merton these that ICAPM in the the pointed out of model context of potential state variables describe the conditional returns and the yield curve components French's Fama her investment She the that and the model outperforms reported set. of (1993) model. Furthermore, she reported that her ICAPM specification succeeds as fails (1993) French Fama three-factor model the and conditional model whereas includes In that excess market returns and model a addition she examined conditionally. innovation to dividend yield, term spread, default spread, short term interest rate and Fama and French's (1993) HML and SMB. She reported the innovations in the four variables drive out the HML and SMB ability to explain stock returns.

43

The above studies employ a small number of macro variables compared with the large number of macroeconomic variables that are available publicly and may represent factors. This makes them vulnerable to the charge that they leave out other risk potential precious information for pricing stocks that is not accounted for (Lettau and Ludvigson (2001), Petkova (2006), Mnch (2004,2006) and Ludvigson and Ng (2007)). Therefore, the main objective of this chapter is to attempt to fill this gap by augmenting the set of the proposed risk variables by Petkova's (2006) to include the whole set of macroeconomic variables available by utilizing the recent development in the dynamic factor models of Stock and Watson (2002a and 2002b). To shed more light on this topic the chapter present the related literature chapter in the next sub-section.

2.2.3

Dynamic Factors Models

Stock and Watson (1998,2002a, b) introduce the dynamic factor model. Stock and Watson (2002a) point out that instead of choosing a limited set of predictive variables,

the dynamic factors model can be used to reduce the large set of available data into a small set of factors that contain the information related to the common variation. They

use the approximate dynamic factor model and the principal component analysis to model the macro variables to be predicted and to estimate the common factors from a large set of potentially useful predictors, respectively. They explained that the assumption is the

follow (X, ) interest forecasted (y, ) be the to and predictive set a of and macro variable dynamic factor model. Stock and Watson (2002a) stated that assuming that the number of lags is finite representation they can express the dynamic of the dynamic factor model" factor by, what they call "the the principal static

and employ

components

44

technique to estimate the factors. They write the static factor model as following (Stock and Watson, 2002a, p. 148, Eq.2.3 and Eq.2.4. respectively) Y,+, = 'F, +Y(L)Y, +e, +, X, = AF, + e, (3) (4)

Equations (3) and (4) above are in Stock and Watson's (2002a) notations. They define the notations in the above equation as follow; F, as rxI vector of factors to be

estimated and their lags, E, as an errors that they assume is uncorrelated with the +, factors, their lags, the lagged forecasted variable and its lags and the predictive variables and their lags, and e, as NxI idiosyncratic of errors. They used their model to vector

forecast a number of macro variables and reported that a few estimated common factors required for capturing the common variation in the 215 macro variables used as predictive set. They pointed out that this means that the variability of the macroeconomy is driven by a few common factors. Stock and Watson (2002b) point out that they allow in this dynamic model for serial and cross sectional correlation in the idiosyncratic error. In addition they state that when the number of observations in the time series (T) and be factors infinity; (N) the to will estimated common go number of predictive variables consistent. Furthermore, they employ the model to forecast industrial production and report promising results in favor of the factor models against the more traditional forecasting methods (such as AR).

The factor models were also studied intensively by Forni et al., (2000) who developed the generalized dynamic factor models. Forni et al., (2000) point out that their factors lagged the model with of static model permits autoregressive reaction whilst

45

Stock and Watson (1998) does not. Furthermore, they state that on the other hand, the latter model takes account of time variation in the factor loading whereas their model

does not. Indeed Forni et al., (2004) state that Forni et al's (2000) generalized dynamic factor model is a generalization models. of Stock and Watson's (2002a) model as well as other method that they stated

Forni et al., (2005) point out that they suggested a prediction

overcomes the shortcoming of Forni et al's (2000) method. More specifically,

that Forni et al's (2000) method lacks the ability to predict whilst Stock and Watson's (2002a, b) method is appropriate for such task. They mentioned that their suggested

method keeps the benefits of the former approach and reported that their method is superior to that of Stock and Watson.

An application

of the factor model for forecasting in the UK is provided by Artis, (2005). They use the dynamic factor model and the principal to previously of 81

Banerjee and Marcellino component employed

analysis and reported that the factor models are superior prediction methods. They They stated that they half of compile the

a panel in

macroeconomic macroeconomic among

variables. variables

reported

that

variation

these

is captured by six estimated common interest rates, employment,

factors which track, measures.

other major

variables,

and monetary

Furthermore, Artis, Banerjee and Marcellino

(2005) point out that as the factors estimated

from a large set of macro variables are linked to vital macro variables, these estimated factors can be seen as the UK economy' drivers.

Forni et al., (2000) reported that they use generalized dynamic calculate indicator an of the business conditions. Inklaar,

factor model to

Jacobs and Ward (2003)

46

indicated that they employed this latter model along with NBER approach to calculate a business cycle index. They point out that while the first method weights the constituent variables on statistical basis the second uses the judgment. and a limited set of predictive limited They argued that a large set providing the

variables can deliver similar performance, In addition, Gillitzer,

set of variables is chosen carefully.

Kearns and Richards factor models In addition

(2005) reported that the coincident

index calculated using the dynamic

better description of the business cycle in Australia than the GDP. provide they point out that their findings

are consistent with Inklaar et al., (2003), in that the

number of variables used in factor models need not be large as long as the variables are selected carefully.

Boivin

and Ng (2006) examine whether a limited

number of variables compared

with a large dataset usually employed in the factor models hurts or helps. They point out that more variables may reduce the common variability across the error components. and increases the correlation delivered

They reported that a small set of variables

common components that either similar if not superior to their counterparts from a much larger set. They stated that the dataset's quality and not only its size what matter for a good estimation.

The above studies show factor models are useful in economics applications such as forecasting macroeconomic variables as in Stock and Watson (2002a, b) or the construction of a coincident index as in Forni et al (2000). However as this chapter studies the behavior of stock price, it turns attention to the next sub-section which cites the studies that apply factor models to asset pricing, even though they are limited so far.

47

2.2.4

Factor Models and Asset Pricing Models

A number of studies study the dynamic factor models, used for estimating factors from a large set of macroeconomic variables, in asset pricing. Among these is Mnch (2004) who argues that there could be common factors that drive all the macroeconomic variables that have been found by researchers to be priced risk factors in the stock returns. Therefore, he indicates that his goal is to examine if the estimated factors that account for the variation that is common across macroeconomic variables also drive stock
returns. He points out that factors, estimated from a large set of macro variables using the dynamic factor model and the principal components analysis of Stock and Watson (1998, 2002a), are used as state variables in an asset pricing model as "diffusion index pricing model". model. Mnch (2004) calls this

He reported that a pricing model with two explains cross-sectionally portfolios the returns on similarly to the

estimated factors along with the market portfolio

the Fama and French's (1993) 25 size and book-to-market Fama and French's (1993) three-factor

model. He pointed out that one of these factors

associates with business cycle (interest rates spread, unemployment and capacity utilization) and the other relates to the exchange rates. He points out that he augments Campbell (1996) model with these two priced estimated factors and reports this augmented model does much better than Campbell model while a little better than his diffusion index model. He reported results using cross sectional regression of Fama and MacBeth (1973) and GMM of Hansen (1982).

Mnch (2006) indicates that he uses the common factors estimated by applying the dynamic factor model of Stock and Watson (2002a, b) to a large group of macro variables

48

as conditioning

instrument

in a conditional

asset pricing portfolios.

models on the Fama and

French's (1993) 25 size and book-to-market large set of information and replacing

He points out that investors have a small number of predictors is

these with

inadequate. He indicated that he overcame the problem of degrees of freedom, which is associated with the use of a many variables, by employing the dynamic factor models. He reported that the estimated factors, which contain additional information are employed as conditioning instruments, conditioning

to the those contained

in the widespread

instruments (interest rates measures, term spread, dividend income to consumption and log consumption

yield, default spread, labor

to wealth ratios) and are better than them. interest rates and housing

He reported that these common factors are related to inflation, variables.

Ludvigson

and Ng (2007) point out that they use the dynamic of omitted information in studying volatility.

factor model to of market

overcome the problem return's conditional positive relationship conditioning

the relationship

mean with its conditional

Consequently,

they reported a

between risk and return. They pointed out that they augmented the variables,

instruments set that include a number of widespread predictive

such as, among others, dividend

price ratio, term spread and default spread with the

dataset, financial factors from factors. They two that which a estimated reported common they call the risk premium factor and volatility factor, are important for conditioning the

from factor the macroeconomic estimated market returns and one mean of stock important for conditional volatility.

dataset is

They reported that they find this latter factor is a real They pointed out that augmenting the

factor as it relates to output and employment.

49

conditional information set with common factors is important for uncovering the correct link between risk and return.

The above studies show the importance of the dynamic factor models in extracting information that may have not been utilized so far, by depending on using only the variables that have been identified in the literature as useful, for describing the behavior of the stock returns. Therefore, this chapter applies Stoc k and Watson's (2002a, b) dynamic factor model and principal component approach to estimate factors from a large set of macroeconomic variables and use these factors as inputs for possible risk factors.

2.3 Hypotheses

Therefore, building on the findings of the previous studies and the issues that they emphasize are important for asset pricing, this section develops the hypotheses of this chapter

2.3.1. Fama and French's (1993) Model and UK Stock Market

A number of studies have examined the Fama and French's (1993) three-factor
model in the UK. Fletcher (2001) reported that, using a sample periods span January1982 to December -1996, the risk premium is insignificant for HML for the SMB but significant

with positive sign. Fletcher and Kihanda (2005) reported that, for the period and SMB's means are insignificant with

January -1975 to December 2001, the HML's

hand, Al-Horani, On the signs, respectively. negative other and positive

Pope and Stark

50

(2003) reported that, for the period 1990 - 1999, the SMB has significant mean with positive sign, while the HML has an insignificant average returns. Furthermore they reported that the Fama and French's (1993) three factor model has a good explanatory power. Hussain, Toms and Diacon (2002) pointed out that they follow Fama and French's (1996) study as closely as possible when apply it to the UK market. They reported positive means for the UK HML and SMB although the last is insignificant. Furthermore, they reported that these factors are significant factors and the Fama and French's (1993) three-factor model outperforms the CAPM in UK market, although none of them is a perfect model. In addition they pointed out the UK results are in line with that of US.

Therefore, this chapter first examines if Fama and French's (1993) three factor model in the UK stock returns over sample period from July 1981 to December 2005 by following Fama and French (1993). The first hypothesis is stated as follows Hypothesis (1): The Fama-French's (1993) SMB and HML portfolios are priced risk factors in the UK cross sectional returns of the Fama and French's (1993) 25 size and book-to-market portfolios.

2.3.2

State Variables

and Stock Returns

Cochrane (2001,2006) emphasizes that the state variables of Merton's (ICAPM) Campbell (1996) future Cochrane (2006) and Petkova returns. praises must predict (2006), among few other studies, for applying this criterion. Petkova (2006) suggests a dividend innovations includes to that market and yield, term spread, excess return model

51

default spread, and the short term interest rates. She pointed out that these four innovations capture the shifts in the investment opportunities. In light of this, this chapter applies Petkova's (2006) model to examine if the innovations to these state variables are priced in the UK market as she found them priced in the US. This provides an out-ofsample test of Petkova's (2006) study. Nevertheless, this is done not for the sake of replication but in response to Campbell's (2000) challenge. He points out that identifying the economic determinants of the risk premiums is the current challenge. Also Cochrane (2006) argues that it is crucial to identify the macroeconomic drivers of the price of risk. Therefore, the second hypothesis is stated as follows Hypothesis (2): Innovations to dividend yield, term spread, default spread, and the short term interest rates are priced cross sectionally by the UK Fama and French's (1993) 25 size and book-to-market portfolios

2.3.3

Are Fama and French (1993) Factors tracked by macroeconomic forces in UK Stock Market?

As cited in the literature review section, Cochrane (2006) points out that the macroeconomic variables that are behind the size and value factors of Fama and French
(1993) should be identified. Petkova (2006) reports that she finds for the US stock

market, the Fama and French (1993)

SMB

and HML

are driven

by innovations

to

variables that are state variables in Merton's (1973) ICAPM.

In fact her exercise is in line

be (2006) tested to whether Cochrane that the should macro models who points out with they can explain the Fama and French three factors instead of the 25 portfolios.

52

Therefore, (hypothesis

if the Fama and French's (1993) size and book-to-market

portfolios

1) and the innovations in the four variables chosen by Petkova (2006) which

are innovations to dividend yield, term spread, default spread, and the short term interest rates (hypothesis 2) are priced in the UK, then the next hypothesis examines if the Fama and French's (1993) SMB's influence and HML's macroeconomic factors as hypothesized influence on stock returns are lost for

and examined by Petkova (2006) for the US Petkova (2006) is:

market. Therefore, the third hypothesis following

Hypothesis (3):

The innovations to dividend yield, term spread, default spread, and the
short term interest rates drive the Fama and French's (1993) SMB and HML factors.

Liew and Vassalou (2000) reported that the SMB and HML in the UK do relate to the future GDP growth and the significance of this association is preserved for the HML

but is absorbed for the SMB by other variables related to business cycle (such as dividend yield, term spread and T-Bill). Furthermore, they point that these findings support that

these two variables are in line with ICAPM

of Merton (1973). Kelly (2003) reports that

that the SMB and HML are linked to real GDP growth and the HML factor also relates to unexpected inflation interpretation in the UK. He points out that this is in agreement with ICAPM and HML. Furthermore, Kelly (2003) report that Fama and

of SMB

French's (1993) model outperforms the CAPM.

53

2.3.4

Large Panel of Macroeconomic Variables and Stock Returns

So far this chapter follows Petkova (2006) and applies her study and methodology to the UK, as she takes on the challenges of Cochrane (2006) and Campbell (2000) as mentioned earlier. However, Ludvigson and Ng (2007) pointed to an important problem of using a small set of variables. They reported that they augmented the conditional information set, in studying the risk and return relationship, with factors estimated from a large dataset using the dynamic factor model to overcome the problem of the omitted information. They pointed out that this problem arises from the fact that the information used by researchers is short of all the real information used by investors. Although Ludvigson and Ng (2007) use the estimated factors as conditioning variables, their argument applies to this chapter as well.

Furthermore, Mnch (2006) pointed out that the factors estimated using dynamic factor model with large macroeconomic variables captures the information used by investors. He examines whether factors estimated from a large set of economic variables, information in the asset pricing pricing when used as conditioning variables model, carry in information information the to those these contained are additional and whether pricing widespread conditioning variables. He points out that he uses the Fama and French's (1993) 25 size and book-to-market portfolios as his test assets. He reports that the important factors are conditioning estimated information and are better than the

factors he information. Furthermore, that these estimated reports conditioning widespread in information the widespread conditioning those contained over carry additional default dividend includes, term spread, yield, spread, one among others, variables which

54

month interest rate and difference between three and one month rates. Also Chen (2003) refers to potential missing state variables in the empirical research that is used by
investors.

On the other hand, Flannery and Protopapadakis (2002) pointed out that the Chen, Roll, and Ross's (1986) statement of the lack of knowledge about the economic sources of risk is valid. Although Flannery and Protopapadakis (2002) employs 17 macro variables, the dynamic factor models can utilize much larger set of variables.

Given the above, it is natural for the next step in this chapter to be the examination whether a few common factors estimated from a large set of macroeconomic variables following Stock and Watson's (2002a, b) dynamic factor model method are priced in the UK stock market. This is done by replacing the innovations in the state variables in Petkova's (2006) model that includes innovations in dividend yield, term spread, default spread and one month T-Bill with the innovation to the estimated factors. The fourth hypotheses is stated as follows Hypothesis (4)A: Innovations to factors that are estimated from a large panel of macroeconomic variables are priced cross sectionally by the Fama and French's (1993) 25 size and book-to-market portfolios

Mnch (2004) applies the dynamic factor models to estimate the factors, which he then uses as state variables (see the above literature review section (Factor Models and Asset Pricing Models)). However, this chapter differs in that its uses the innovations to the estimated factors, which follows the application of the innovations to the four

55

variables chosen by Petkova (2006)), while Mnch (2004) uses the estimated factors themselves.

Hypothesis (4)B: The innovations to estimated common components drive the Fama and
French's (1993) SMB and HML factors.

We believe this chapter could be the first study to use the innovations in factors
estimated form a large panel of macro variables as potential drivers of the Fama and French's (1993) SMB and HML. Hypothesis (4B) becomes important, in particular, if

the innovations to the four state variables in Petkova' (2006) model (in Hypothesis (3)) do not capture the effects of the size and value portfolios of Fama and French (1993).

2.3.5

Other Estimation Methods

Priestley (1996) suggested the Kalman filter

based method to produce the

innovations. He pointed out that this method ensures that the learning process is reflected in the investor's expectations, as well as the produced surprises are real innovations. He reported that this method delivered better results for the APT than the autoregressive and
rate of change methods. On the other hand, Shanken (1992) pointed out that the

from in (1973) Fama MacBeth's the the suffers and second step regression coefficients problem of the error-in-variable, becauseof the generated regressors. He pointed out that the standard errors of the risk premiums under the Fama and MacBeth's (1973) method are understated. Jagannathan and Wang (1998) pointed out that when the conditional distribution of returns is not homoskedastic, the standard errors are not inevitably

56

understated by the Fama and MacBeth (1973) method. Jagannathan and Wang (2002) point out that the reason the generalized methods of moments (GMM) of Hansen (1982) gains its status as it overcomes all the problems by allowing for non-normality, heteroskedasticity and serial dependence.

Therefore as a robustness check for the chapter's results, different innovation estimation techniques and model estimation method are used; VAR and Fama and MacBeth (1973) following Petkova (2006), then the VAR based innovations are replaced by Priestley's (1996) Kalman Filter based innovations. In addition this chapter replaces the Fama and MacBeth (1973) method with the GMM on beta representation following Jagannathan and Wang (2002). Petkova (2006) noted that she also used the GMM method. However, she applied the GMM to the stochastic discount factor representation, while this chapter follows Jagannathan and Wang (2002) in applying the GMM to the beta representation.

2.4

Data and Methodology

2.4.1.

Data

2.4.1.1 Stock returns

The sample consists of all the UK common stocks traded on the London Stock Exchange (LSE), excluding foreign stocks, for the period of June 1981 to December 2005 obtained from the Datastream. This includes both active and dead (de-listed) stocks.

57

Actives stocks are those that were still being traded until December, 2005 while the delisted stocks are those that happened to be traded some time between 1981 and 2005 and then de-listed. Financial firms are excluded from the dataset. Fama and French (1992) pointed out that they excluded them as they have different leverage than non financial stocks. In addition, Griffin (2002) indicated that he used non financial firms and marketto-book ratio's inverse downloaded from Datastream. Additionally, Fletcher (2001)

pointed out that he formed the Fama and French's (1993) six portfolios for UK using non-financial stocks. Following them, this chapter excludes financial stocks from the analysis. The final database that is available for analysis before applying any more criteria constitutes of 3706 non-financial stocks. The database has not been checked for the illiquid and very small stocks and therefore it could contain a large number of highly
illiquid and very small stocks that are not frequently traded. These could severely affect

the results and some of the findings may be affected by the presence of such stocks. The

beginning of the period is determined based on the availability of book value data from
the Datastream. This chapter uses the monthly frequency.

Professor Krishna Paudyal has supplied this chapter with the Fama and French's (1993) SMB and HML for the UK from July - 1981 up to December - 2003 which are
constructed following the methods explained in Fama and French (1993,1996). For the

SMB 2004 (1993) French's 2005, Fama the the and HML sample period; rest of and and for the UK are constructed in this chapter following Fama and French (1993) closely, French (1993) by Fama is Professor Paudyal. the and same method used pointed which fiscal book (t-1)'s (t-1) June's (t) December's year end price and year value price, out describe be French (1993) how for Fama be to they and used. each stock available should

58

divided follows; (t), (1) SMB, in HML June, the are common stocks and as year construct into small size portfolio and big size portfolio, using the median size of NYSE as breakpoints, (2) stocks are divided independently into three; low (30%), medium (40%) and high (30%) book-to-market portfolios, using NYSE book-to-market values after by breakpoints. They that they measure size as shares multiplied pointed out ranking as their individual price, market value as its value at end of December of year (1-1), and book-value as of its value at the end of the stock's fiscal year occurred in the previous
year (t-1). They pointed out that they exclude stocks with negative book values from the breakpoints' computation and portfolios' constructing. They pointed out that they form, six portfolios and compute

by intersecting the above size and book-to-market these portfolios' monthly

portfolios,

returns from July (t) to June (t+1)

as the value-weighted

is book is They the to that this already publicly ensure used value returns. pointed out available. They pointed out that this process is repeated every June. They pointed out that they construct, for each month, the SMB as the spread between the three small portfolios' average returns and the three large portfolios' between the two high portfolios' average return, and the HML as the spread average

average returns and the two low portfolios' values.

returns, where high and low in terms of their book-to-market

Furthermore, Fama

and French (1993) pointed out that they use the constituent stocks of the six portfolios form book the market portfolios. to the value and stocks with negative that return on this portfolio is the value-weighted They pointed out

free-rate. the risk returns minus

This chapter replicates Fama and French's (1993) procedure for constructing SMB an dH ML for 2004 and 2005 for the UK. In addition

the

it replicates Fama and for the UK for the whole

French's (1993) procedure for constructing the market portfolio

59

into divide LSE it but LSE the the to the size stocks median of uses size period, sample into divide book-to-market LSE LSE' to the the stocks ranked and values portfolios book-to-market portfolios. Indeed, Hussain, Toms and Diacon (2002) point out that they use the median size of LSE and the LSE (30%, 40% and 30%) ranked book-to-market values to divide LSE stocks into size and book-to-market portfolios, respectively and then to form the HML and SMB for UK market.

This chapter employs as test assets, the UK Fama and French's (1993) 25 size and book-to-market value portfolios. The reason for choosing these portfolios is that Petkova (2006) uses these portfolios as test assets, and as the first objective of this chapter is to provide an out-of-sample test for her study on the UK market, this means that this chapter has to use them as test assets. In addition to this Petkova (2006) points out these test assets are benchmark. Furthermore, Fama and French (1993) point out that in order to examine the ability of the HML and SMB to explain the size and value returns, they employ these 25 portfolios. This chapter forms the Fama and French (1993) 25 portfolios by closely following Fama and French (1993). They point out that they form the 25 portfolios by similar to the six portfolios, with the difference is, five size portfolios are constructed using the breakpoints of NYSE and independently five book-to-market 25 breakpoints NYSE then the the size and and of portfolios also constructed, using book-to-market portfolios is the result of the intersection between these size quintile portfolios and book-to-market quintile portfolios. They pointed out the monthly value from July (t) free 25 (of the are computed portfolios risk rate) returns of weighted excess to June (1-1) to become the test assets.

60

This chapter replicates Fama and French's (1993) method for forming the 25 size and book-to-market portfolios for the UK stocks, but again this chapter use the LSE's breakpoints for the quintiles to divide the stocks of LSE into quintile portfolios. Hussain, Toms and Diacon (2002) also point out that they use LSE's breakpoints for the quintiles to divide the stocks of LSE and form the 25 size and book-to-market portfolios.

Table (2.1) shows the number of firms available after applying the criteria of Fama and French (1993) for stock's selection. Nagel (2001) points out that the data from Datastream has the problem of selection bias that existed until the 1970's end as high book-to-market value stocks as well as small stocks are missed from the Datastream. In agreement, Table (2.1) shows the number of available stocks with book values is small until the late eighties.

Table (2.2) presents the average returns on the UK Fama and French (1993) SMB, HML and market excess return for the period July-1981 to December-2005. average return on the HML portfolio return on the SMB portfolio is consistent with Dimson, is positive and significant It shows the

at 1% while the average

is negative and significant at 10%. This is not surprising as it Nagel and Quigley (2003). They reported that there is a

bigger, in magnitude, and more significant in magnitude, and insignificant

value premium in UK relative to the smaller, However while this chapter finds

size premium.

negatively significant

size premium they reported positively

insignificant

size premium

over the period from 1955-2001. Dimson, Nagel and Quigley returns of each portfolio

(2003) depict the yearly

(SMB and HML) over time. Dimson, Nagel and Quigley (2003)

is find behavior SMB they the that the consistent with Dimson and of premium point out

61

Marsh (1999), it is volatile and exists before 1989, then it reverses before it recovers in 1999. In addition, Dimson, Nagel and Quigley (2003) report the HML portfolio had high and stable returns during the first part of their sample period, but since the nineties it has become more volatile and the highest value premium magnitudes happened in this recent
period. To compare the premiums of Fama and French's (1993) factors used in this (2003)), following them, it to

chapter with theirs (i. e. with Dimson, Nagel and Quigley depicts each portfolio's December-2005 (SMB and HML)

return for the sample period July-1981

in Figure (2.1), however it uses monthly

instead of yearly frequency. monthly as those

Panel (A) shows the HML's return. The figures reported by Dimson,

monthly returns and Panel (B) shows the SMB's behavior for the size and value premium (2003) over the overlapping

show similar

Nagel and Quigley

period, this is

despite the fact that the there are slight differences relation to forming the SMB and HML.

between this chapter and theirs in

2.4.1.2 State Variables Petkova's (2006) Chosen Variables -

Two sets of candidate state variables are employed by this chapter; (1) the state variables selected by Petkova's (2006) and (2) the factors estimated from a large set of macroeconomic variables using the dynamic factor model and principal component to four (1)

analysis of Stock and Watson (2002a, b). Petkova (2006)

uses innovations

variables as candidate states variables. She defines these four variables as follow; dividend yield on the CRSP portfolio

(the ratio of total dividends calculated over the past

12 months to the index level), (2) the term spread (the spread in yields between ten- year default difference bond bond), (3) (the in spread and oneyear government government

62

the yields between corporate-Baa and government bonds, both long term bonds) and (4) one-month T-Bill yield. She states that these variables capture the shifts in the investment opportunities, where the second (2) and fourth (4) variables model the changes in yield curve and the first (1), third (3) and fourth (4) variables stock returns' conditional distribution. As the first step of this chapter is to apply Petkova's (2006) study and methodology to the UK, it follows her and employs the UK counterpart variables. The purpose is to examine whether her chosen set of variables are priced in the UK in a step toward responding to the challenges of Campbell (2000) and Cochrane (2006). These variables are; 1- dividend yield of FTSE all share, 2- term spread; the spread on 5-year central government securities over that of the three-month Treasury bill, 3- default spread; the spread of the UK FTA Debenture and Loan Stock Redemption Yield (25 years) over that of the UK 20-year central government bonds from July - 1981 to October - 1995, then the spread of the Corporate Bond Yield over that of the UK 20-year central government bonds from November- 1995 to December-2005, 4- one-month T-Bill yield

These UK variables were also used by Antoniou momentum in the UK (in p. 959). This chapter follows

et al., (2007) in their study of their definitions of the dividend

bonds default but five-year it the while they use the 20 government yield and spread uses year government bonds in the term spread definition. The five year government bonds is

because bonds 20 in the term the when the term year spread and not measuring used

63

spread is measured based on the 20 year government bonds, the innovations to the term spread is found to be insignificant.

2.4.1.3 State Variables

Large Panel Macroeconomic of -

Variables

As mentioned earlier in the chapter, the problem, in choosing a limited number of potential risk variables, is that they may not be representative to the information set used by the investors, as pointed out by Ludvigson and Ng (2007) and Petkova (2006) among others. To overcome this problem, Ludvigson and Ng (2007) and Mnch (2006) use the dynamic factors models. For the same reason as well as to avoid the pitfall of using any
variable that is pointed out by Fama (1991), this chapter chooses variables that are

suggested by Cochrane (2001). He points out that the factors in the ICAPM are those variables that have predictive ability of stock returns or have predictive ability of macroeconomic variables. Stock and Watson (2002a) used factors estimated from a large dataset of macroeconomic variables to forecast macro variables therefore this chapter estimates factors from a large dataset of macroeconomic variables using Stock and Watson's (2002a, b) method. Then this chapter uses the innovations to these estimated factors as candidate state variables. Therefore, it selects variables in line with the ICAPM
of Merton (1973) as suggest by Cochrane (2001), i. e. those that predict macroeconomic

variables.

Lagan and Mountford (2005) point out that VAR model has the problem that it uses a limited set of variables that may not reflect the reality and hence it could be misspecified. They point out that they use factor -augmented vector autoregression

64

model to study the interest rates in the UK. They point out that they construct a large
balanced panel of macroeconomic variables (105) for the UK for the purpose of using

them to estimate factors using Stock and Watson's (1998,2002a) method and add then them to VAR model. Lagan and Mountford (2005) also point out that Boivin and Ng (later published as Boivin and Ng (2006)) indicate that increasing the number of variables may be harmful. In addition, Lagan and Mountford (2005) point out that they chose variables and categories similar to Bernanke et al., (2005) and Stock and Watson (2002a) from Datastream. Furthermore, Artis, Banerjee and Marcellino (2005) point out that they build a balanced large set of macroeconomic variables (81), from Datastream and the
OECD, for the UK economy which they use with the dynamic factor model to forecast

macro variables.

Therefore this chapter downloads from the Datastream as much as it finds of the variables that are used by Lagan and Mountford (2005) and some of those used by Artis, Banerjee and Marcellino (2005) and Kapetanios, Labhard, and Price (2006). However it does not retrieve any financial variables that are used by these studies as its studies the macroeconomic variables.

In this chapter, a balanced dataset containing 78 monthly macroeconomic

variables

for the UK, from July- 1981 to December - 2005 is collected from the Datastream. These variables are (1) those used by Lagan and Mountford13 (2005) covering their macroeconomic categories. However, this chapter does not manage to retrieve the entire
" Lagan and Mountford (2005) use the following categories "employment; government finance; output; housing starts and vehicles; consumer and retail confidence; prices; money and loans; interest rates; composite leading indicator; and stock prices and exchange rates". See their Appendix (Lagan and Mountford (2005, p.94-97)

65

105 variable used by them for its balanced dataset over the period July - 1981 to December 2005; (2) Additional few variables are obtained similar to those used by Artis,
Banerjee and Marcellino (2005); (3) Additional few variables are obtained similar to

those of Kapetanios, Labhard, and Price (2006); and (4) variables of default spread which are also similar to Ludvigson and Ng (2007) are also included in the chapter dataset. See Appendix (A).

2.4.2

Methodology

2.4.2.1 Petkova's (2006) Model of the ICAPM

This chapter applies Petkova's (2006) model. She states that she assumesa discrete and unconditional version of Merton's (1973) ICAPM. Before proceeding, it is important to understand the context of the unconditional form of the ICAPM model. For this purpose this chapter cites Constantinides's (1989) study. Constantinides (1989) points out that in the unconditional ICAPM that he derives from the conditional ICAPM, stock returns co-vary not just with the state variables but also with the vector of information set (c'-') and the variables that have been found are able to predict stock returns are

candidates for this vector.

Petkova (2006, p. 583, Eq. 1) assumes the following
E(R, ) = YepQ. (Y )Q;. + n L, u

ICAPM for each i: (5)

Equation (5) above is in Petkova's (2006) notations, she defines the notations in the

follow; E(R) as above equation

is stock i excess return yy, and yu, are risk premiums ,

66

8,, 8. in innovations K the are obtained state variable, respectively. market and on ,, K from time series regression, (Petkova, 2006, p.584, Eq.2)
R., =a,
+N, MRAI, +I ., `/ )uf ,, ^ +6rr

(6)

Equation (6) above is in Petkova's (2006) notations. She defines the notations in the above equation as follow; ' 's are the state R,,,.,is the market's excess return and u,

variables' innovation, all measured at the end of time t. To estimate the innovations from the state variables, Petkova (2006) uses the vector autoregressive approach of Campbell (1996). She assumes the following autoregressive model (VAR), (Petkova, 2006, p. 584, Eq.3)
z, =Az, _, +u,

first-order

(7)

Equation (7) above is in Petkova's (2006) notations. She points out that she inserts as the first element in the demeaned14state variables vector (z, ), the excess market return followed by the other state variables. She states that the model presented by the above cross-sectional, time-series and VAR models has an advantage. She points out that Campbell (1996) points such model reduces the possibility of uncovering spurious relationships.

2.4.2.2 VAR Innovations

This chapter uses VAR following Petkova (2006). As mentioned earlier Petkova (2006) estimates the vector of innovations to the state variables (u, ) from the VAR (equation (7) above). She indicates the first variable to enter the VAR is the excess
14Campbell (1996) points out that this assumption is for simplicity.

67

market returns, then it is followed by (in her order) the dividend yield, term spread,
default spread, risk-free rate, Fama and French's (1993) HML and SMB where all the variables are demeaned. She points out that in this specification the HML and SMB are considered as possible state variables.

The VAR lag length is decided as suggestedby Hall (1991). He points out that there are two methods to determine the order of the VAR. However, he points out the preferable approach is the one that begins by a high order and then reduces down the lags' length and then uses the likelihood ratio restriction's test. In additions, he indicates that when the OLS is used to estimate the VAR model, the log likelihood ratio still can be computed.

Petkova (2006) points out that Campbell (1996) indicates for the VAR's estimation results to have a meaning, there is a need to orthogonalize the factors and scale them. Petkova (2006) points out that she follows Campbell in trianagularizing the VAR system and leaves unaffected the innovation to the excess return on the market (the first variable to enter) and calculates the orthogonalized innovations to the second variable in the system as its part which is orthogonal to the first unaffected variables etc. In addition,
Petkova (2006) points out that again she follows Campbell (1996) and scales the

innovations

to the variables so that their variances will

be equal to variance of the

innovation to excess market return. Therefore, this chapter follows Petkova (2006) in 15. in triangularizing and scaling the innovations the state variables

15 See Hamiltcn triangularization.

(1994), Time Series Analysis, for illustration

of the technical procedure of the

68

This chapter follows Petkova (2006) exactly in her application of VAR system (including the order of the variables inserted) to estimate the innovations to the dividend yield, term spread, default spread, one-month interest rate, Fama and French's (1993) HML and SMB. Furthermore, this chapter also applies her VAR system in similar way to the factors estimated from a large set of macroeconomic variables where the latter replace the above Petkova's (2006) chosen state variables. Including factors estimated from a large step of macroeconomic variables into VAR model is similar to Lagan and Mountford (2005) who augment the VAR model with factors estimated using the dynamic factor model. Lagan and Mountford (2005) point out that they follow Bernanke et al. (2005) and include factors estimated from a large set of macroeconomic variables into the VAR model to study the monetary policy, where the factors estimated using the dynamic factor approach of Stock and Watson (1998,2002a).

2.4.2.3 Kalman Filter Innovations

In addition to using the VAR to calculate the innovations to the potential state variables as in Petkova's (2006), this chapter uses the Kalman filter of Priestley (1996) as
an alternative technique to produce these innovations. Petkova (2006) points out that she

uses AR (1) as an alternative method to estimate the innovations and points out that the results are not different from the VAR's. However Priestley (1996) studies the APT of

Ross (1976) and points out that a problem with the autoregressive model is that it does based He learning by investors. that the the out a method points on the process not entail Kalman filter overcomes this problem.

69

Priestley (1996) models the risk factor, as (Priestley, 1996, p.873, Eq.4 and Eq.5, respectively) X, =X*, + u,
X =X i-I +Yr-I +yt

(8)

Y, = 7, + w, -1

(9)

Equations (8) and (9) above are in Priestley's (1996) notations, he defines
Equations (8) and (9) as the measurement and transition equations, respectively. In

addition he defines the notations in the above equations as follow, X',

value of the risk factor (X, ), y,

is the expected

is a parameter, and u, C, w, are white noises. He , _,

indicates if the above model produces non-serially correlated residuals then these will be the innovations to be used as risk factors, otherwise, X, is modeled as (Priestley, 1996, p.873, Eq.6 and Eq.7, respectively p.873) X, =811X, +s, _; 8 =S;, +0) ;, It -1 (10) (11)

Equations (10) and (11) above are in Priestley's (1996) notations. Priestley (1996) defines equations (10) and (11) as the measurement and transition equations respectively. He points out that that X, is here modeled as an autoregressive process in which the parameters are time-varying.

70

2.4.2.4 Dynamic Factor Model of Stock and Watson (2002a, b): Their Static Representation

This chapter follows Stock and Watson (2002a, b) static representation of dynamic factor model and principal components analysis which they use to estimate factors from a large set of macroeconomic variables.

Stock and Watson (2002a, p. 148, Eq.2.3 and Eq.2.4. respectively) assumes X, =AF, + e, (4)

Equations (4) above is in Stock and Watson's (2002a) notations (see section 2.2.3 Dynamic Factors Models in the literature review above). Following Stock and Watson's (2002a, b) this chapter assumes the macroeconomic variables follow equation (4) and uses their principal component approach to estimates the factors (F, ). Then it uses innovations in these estimated factors as potential risk factors. Stock and Watson (2002a, b) include these estimated factors in a second step regression to predict a number of macroeconomic variables (see Equation (3) in section 2.2.3 Dynamic Factors Model in the literature review above.

Stock and Watson (2002a)

point

out16 that as under the dynamic

factor the

macroeconomic series should be 1(0), these series may need to be (1) transformed, (2) first differenced and (3) undergone outliers screening. Artis, Banerjee and Marcellino (2005) point out that they follow Marcellino, Stock and Watson (2003) in this regard. For

16As mentioned earlier in the chapter, Mnch (2004) uses the factors estimated, using the method of Stock and Watson, from a large set of US macro variables, as potential state variable, however, this chapter is different from his study as it uses the innovations in these estimated factors as potential state variables while he used these estimated factors themselves as state variables.

71

convenience, as Marcellino, Stock and Watson (2003) describe these processes in details this chapter also follows Marcellino, Stock and Watson (2003). They point out that, all non-negative, non-rates and non-percentage series, are transformed by taking logarithms and the general rule is to apply same transformation and differencing degree to the group of variables. They indicate that the next step is that all the series under study undergo seasonal adjustment process of two-step which includes Wallis's (1974) adjustment and in the final step all the series are screened for outliers. They define the outliers as those observations that 6 times more than the interquartile range. They point out that the outliers are treated as missing observations and all the series are transformed so that they have zero and one unit of mean and variance, respectively.

This chapter follows Marcellino, Stock and Watson (2003) in these steps. However, firstly, for the first step Stock and Watson (2002a) point out that unit root tests are undertaken as a part of the process to deicide whether to take differencing. Therefore, the Augmented Dickey Fuller test for unit root is performed for all the series as described and explained by Harris and Sollis (2003, Chapter (3))". Furthermore this chapter applies Perron's (1997) test for break points in the series. Marcellino, Stock and Watson (2003)
point out that they use two sets of transformed data, however, this chapter follows Lagan and Mountford (2005) and Stock and Watson (2002a) in that using one set of transformed

variables. Secondly, for seasonal adjustment, this chapter simply uses the X-11 procedure in SAS version (9.0) for seasonal adjustment. Finally, instead of treating the outliers as missing observations, this chapter replaces the outliers as follow; each outlier observation is replaced by the maximum value (after removing all the outliers) if the original

17As preliminary test, this chapter applies Dickey Fuller test as in Harris and Sollis (2003, Chapter (3)).

72

observation is positive and the minimum value if the original observation is negative. Schneider and Spitzer (2004) pointed out that they replaced the outlier by an interpolation, and hence it is not necessarily to treat the outliers as missing observations.

2.4.2.5 Fama and MacBeth's (1973) Cross Sectional Regressions

The next step is estimating Petkova's (2006) models with

her different

specifications following Petkova (2006) as explained below. She uses the Fama and MacBeth (1973) methodology to estimate the time-series and cross-sectional equations of the model as follows: she indicates that in the first she runs for each stock the time-series regressions to estimate the betas as (Petkova (2006, p. 587, Eq.5):
= a, +, 8, ,M RAI., + / ({-' )Z/ A + e., Vin'

(12) (2006) notations, although she

Equation I. specifies

(12)

above

is

in

Petkova's

Then she states the next second step is estimating the monthly

cross-sectional regression as (Petkova, 2006, p.587, Eq.6): E(R) =Y,, +Y, ;. +Z M Equation specifies (13)
i,

(13) is in Petkova's ( ); ` ,, A (2006) notations,

above

although

she

She specifies

1(yu. )/ and 1.

,,

in the above

equations (12 and 13) for her most general model which includes as factors the market excess returns, innovations to dividend yield, innovations to term spread, innovations to default spread, innovations to one-month interest rate, innovations to HML and

innovations to SMB. In addition Petkova (2006) examines the Fama and French's (1993) three-factor model and her other model that includes the market excess returns,

73

innovations to dividend yield, innovations to term spread, innovations to default spread, and innovations to one-month interest rate. Petkova (2006) points out that she proposes this last model (that includes the market excess returns, innovations to dividend yield, innovations to term spread, innovations to default spread, and innovations to one-month interest rate) as a superior ICAPM model.

Petkova (2006) points out that she estimates the factor loadings in the time series regression by running multiple regressions over the full-sample period as in Lettau and Ludvigson (2001) and also by using Fama and MacBeth's (1973) rolling multiple
regressions over five-year window. This chapter follows Petkova (2006) in using Fama

and MacBeth (1973) regressions with multiple regressions over the full sample period for

estimating the betas.

Petkova (2006) calculates Fama and MacBeth's

(1973 t-statistics for the estimated

coefficients but because of the problem of the errors-in-variables, Petkova (2006) points out that she also uses the Shanken (1992) correction to correct for this problem. This chapter does the same and applies Shanken's (1992) correction. Shanken (1992, p. 13,
Eq. 1 1) derives it as:

(t + c)[w - E;; ]+E; -. l.

(14)

Equation (14) above is in Shanken's (1992) notations, he defines the notations in

'I', the above equation as; 0= I'2E: z ,W

is the covariance matrix of r,

where the latter

is a vector of the cross-sectional regression estimates, E;:

is bordered version of the

74

covariance matrix of the factors (E,, ). He points out under the assumption of serially independent factors, then E;: =ij., *

Petkova (2006, p. 599, Eq. 13) computes the cross sectional R2 as:

RZ =

aC

22-

-6`. 9((R)

(15)

Equation (15) above is in Petkova's (2006) notations, she defines the notations in k the above equation as; a( the cross-sectional variance, vectors of average and as ,e residuals and average excess returns, respectively. She points out that Lettau and Ludvigson (2001) as well as Jagannathan and Wang (1996) use this measure. She also calculates the adjusted cross-sectional RZ. Furthermore, she depicts the performance of the models visually, she points that this is useful for comparison. This chapter follows Petkova's (2006) in calculating both cross-sectional R2 and depicting the models' performance visually. However, she points out becausethe RZhas a problem of assigning similar weights to the test assets regardless of how much they are correlated, she calculates the composite pricing errors (test of jointly zero pricing errors). Similarly Lettau and Ludvigson (2001) point out that they test the hypothesis of jointly zero pricing errors using the Wald test (x2 test) as (Lettau and Ludvigson, 200
(1+/' f 2) -'a, Cov(a1 , )-I rn, _XZN-A

footnote 28) (16)

Equation (16) above is in Lettau and Ludvigson's (2001) notations. They define the ;,,, factors' in follows Ef; the the covariance matrix, above equation as ; the notations vector of Fama and MacBeth's pricing errors, K and N; number of factors and

75

portfolios, respectively, and 1+2'Y-j' A is Shanken's (1992) correction. This chapter

follows Petkova (2006) in calculating the test of jointly zero pricing errors. However it

applies as in Lettau and Ludvigson (2001) which differs from Petkova (2006) in that she takes a transformation of the above statistic. Note that I+JA in Equation (16) is equivalent to I+c in Equation (14)

2.4.2.6 Generalized Methods of Moments (GMM)

Jagannathan and Wang (1998) indicate that Fama and MacBeth's (1973) method does not inevitably underestimate the standard errors if the factors and returns are not homoskedastic. In addition, Jagannathan and Wang (2002) point out that the generalized method of moments (GMM) allows for the violations of homoskedasticity and serial independence and because of that they use it on their beta representation asset pricing model. Petkova (2006) points out that she uses GMM estimation but on the stochastic discount factor model form of the ICAPM. However, this chapter adopts Jagannathanand Wang's (2002) method of applying the GMM on the beta representation of the asset pricing model. They call it the beta method.

Jagannathan and Wang (2002) write the beta representation

of an asset pricing

follow Eq. 1) 2002, 2339, (Jagannathan Wang, p. and as model

E[r,I =15

(17)

Equation (17) above is in Jagannathanand Wang's (2002) notations, they define the 8, is in the the excess of stocks returns, are a n-vector above equation as r, notations 8

76

factor price's of risk and factor loading, respectively. They point out that the above beta

2002, Wang, in following factor (Jagannathan the and model representationresults

8 2)' 2340, Eq. p.

,-,=(s-u+ f)Q+C,

(18)

Equation (18) above is in Jagannathanand Wang's (2002) notations, they define the is factor f, is follow; the in the the mean of risk e, as above equation notations , 1u in factor, f, be (1) In this case there traded they that could a point out residuals. addition is a restriction that p=8 its factor's is implies the risk premium's used as mean which

(2002) factor. Jagannathan Wang Furthermore, (2) point out and non-traded estimate, or that in the case of non-traded factors, the moments restrictions of the model (18) above Eq. 3, Eq. 4, Eq. 5 Eq. 6, (Jagannathan Wang, 2002, 2341, respectively) and are and p.

E(r, -(6-p+.

f, ))=Ox,

(19) (20)
(21) (22)

E[(r, - (S -+. f ))]f, = 0, x

E[f -p] =0 E[(.f, -p)Z -6Z] =0

Equations (19), (20), (21) and (22) above are in Jagannathan and Wang's (2002)

is 0x, in define the n-vector of They the the as above equations notations notations. zeros, is factor's be the to o2 estimated, vector of parameters

(1982) Hansen's They they that regulatory conditions are assume point variance. fulfilled. Furthermore, they point out that they test the model specification using
and large number of time series' observation

Hansen's J-statistic which under linearity

18See Jagannathan and Wang (2002), for more details.

77

assumptions converges to x2. Jagannathan and Wang (2002) point out that when the factor is traded then the moments restrictions are (Jagannathan and Wang, 2002, p.2341, footnote 2) E[r, -. f ] = 0., E[r1- f ].f = 0xl (23) (24)

Equations (23) and (24) above are in Jagannathan and Wang's (2002) notations. They point out that in such case 6 can be calculated using (Eq. 23 and Eq. 24) and p and a2 can be estimated separately from (21 and 22), where u is the risk premium. To keep consistency across equations' notations that appear above in the other sections of the
methodology, to u; ` 8 (risk premium) is equivalent to y and f, (risk factor) is equivalent

This chapter follows Jagannathan and Wang (2002) and estimates all various Petkova's (2006) models - which are estimated in the previous section using Fama and
MacBeth (1973) methods following Petkova (2006) - using GMM with the moments

Eq. 19, Eq. 20, Eq. 21 and Eq. 22. In a model which includes traded factors only or traded factors and non-traded factors the model is estimated using the moments from Eq. 19 to Eq. 22 and a test of the equivalency between the estimated risk premium and the mean of the factor is conducted as in Jahankhani (1976). Furthermore, Fama and French's (1993)

Eq. 21 Eq. 22. 24 Eq. Eq. 23 is and and separately and model also estimated using

Fama and MacBeth (1973) points out that in the CAPM the estimated market risk

They to test that the point out return. excess average market premiumshouldequal
78

statistically the difference in magnitude between the two is in fact equivalent to the test of the difference between the estimated intercept and the average risk free-rate. They
calculate the t-statistics for the difference by dividing the difference between the

estimated coefficient and its average by the standard error of the estimated coefficient. Similarly, Jahankhani (1976) points out that he tests if the estimated risk premium is equal to the corresponding factor's average excess return using the following statistics (Jahankhani, 1976, p.520)
(Rrn, RJ!

(25)
s(Yl,

Equation (25) above is in Jahankhani's (1976) notations. Eq.(25) could be written,

using the above Jagannathan and Wang's (2002) notations to keep the consistency, as

t(f, ) =

s tan dard

.5-,u
-

error ((5)

Lewellen, Nagel and Shanken (2006) point out that the fact that many models are

found to have power to capture the value and size effects while they are not related to each other is confusing. They point out that the restriction on the slopes' magnitudes from the cross sectional regression; i. e. the estimated risk premium should be equal to the
average of its risk factor, should be tested. This chapter applies the test as in Jahankhani

(1976) for the equivalence between the estimated risk premium and the average return to traded factors where the Fama and MacBeth (1973) or GMM on the beta representation following Jagannathan and Wang (2002) are used to estimate the model as Jagannathan its factor imposes (2002) Wang that the traded restriction on point out estimated risk and its (1972) be Scholes Black, Jensen to to also point out that the equal mean. and premium

79

pricing theory entails the estimated risk premium in the cross sectional regression equals the average market excess return. They also show and calculate the above t- static for testing the equivalency between the two values. Furthermore, Brennan, Wang and Xia

(2004) also use two-step cross sectional regression and examine if the average return on each of the Fama and French (1993) three factors is equal to its corresponding estimated risk premiums.

2.5

Results

2.5.1

Fama-French's (1993) Three-Factor Model in the UK-LSE

This chapter starts by testing the first hypothesis which examines if the FamaFrench's (1993) SMB and HML portfolios are priced in the LSE using the Fama and French's (1993) 25 portfolios' excess returns. Panel (A) of Table (2.3) shows the

coefficients estimated from the Fama and MacBeth (1973) second-step cross-sectional regression. It is apparent that excess market returns is insignificant, while the SMB and HML are statistically significant at 5% and 1%, respectively using Fama and MacBeth's (1973) t-statistics. However, when Shanken's (1992) corrected standard errors are despite be However, found SMB HML the the to and only significant. calculated, and are fact that the estimated intercept is not significant, the hypothesis of jointly zero-pricing hypothesis is Nevertheless, that the risk premiums on the the as shows. x2 errors rejected is SMB to their average corresponding returns and are equivalent market portfolio accepted while rejected for the HML at 10%.

80

Panel (B) of Table (2.3) shows the results of estimating the model using the GMM on the beta representation following Jagannathanand Wang (2002). As pointed out by the latter study paper when the factors are traded factors then there are restrictions that the average returns on the factors are equal to their corresponding risk premiums. The results in Panel (B) show the risk premiums estimated using the four moments restrictions (Eq. 19-Eq.22), i. e. the factors are treated as non-traded factors and then the final column presents the test of the equivalency between the risk premium and its corresponding factor's mean. When the model is estimated by treating the candidate risk portfolios as traded factors, using the moments (Eq. 23 and Eq.24) and then separately the moments (Eq. 21 and Eq.22) as mentioned above, it produces similar J-statistic for the model restrictions.

The results from the GMM MacBeth's

estimation

are qualitatively

similar

to Fama and

(1993) method results using the Shanken (1992) corrected standard errors. are priced significantly with negative and positive risk premiums, priced. However, while the

The SMB and HML respectively,

while the market factor is not significantly

magnitude of the SMB premium is similar under the two methods, the risk premiums of the market portfolio and the HML are smaller under the GMM and the hypothesis that

factor's mean equal to its estimated risk premium is accepted for the three factors; market excess return, SMB and HML. premium under the GMM This is not surprising for the HML as its estimated risk is much closer to its average return. Finally the

estimation

specification test is accepted as the Hansen's J- statistic shows.

81

The first hypothesis, which states that the Fama-French's (1993) SMB and HML are priced factors in the cross sectional returns on the UK Fama and French's (1993) 25 size and book-to-market portfolios, is accepted. Although the Fama and French's (1993) three-factor model's pricing errors hypothesis is rejected under the Fama and MacBeth's (1973) methodology, the model specification is accepted under the GMM. In addition, while the SMB premium is negative and significant, the HML premium is positive and significant. This finding is expected as it is shown in Table (2.2) that the average return on the SMB portfolio is negative and significant at 10% over the sample period from July-1981 to December -2005. Furthermore the performance of SMB is consistent with Dimson, Nagel and Quigley's (2003) study although the sample of the latter study ends in 2001. Petkova (2006) reported that the application of the Fama and French's (1993) three factor model resulted in a SMB premium that is positive but insignificant and a significant and positive risk premium associated with HML for the US market.

Furthermore she reports that the risk premium of the market portfolio is negative but insignificant and the jointly zero pricing error hypothesis is rejected for the Fama and French's (1993) three-factor model

2.5.2

Petkova's (2006) Model that includes Excess Market to Four Variables

Return and Innovations

This subsection shows the results of the testing of the second hypothesis which includes Petkova's (2006) ICAPM, the excess market return, the which model of applies default T-Bill, dividend innovations to to term spread and one-month yield, spread, and

82

the UK market to examine whether it is priced in the UK cross sectional average returns on stocks.

Panel (A) of Table (2.4) shows the results of estimating the model using FamaMacBeth's (1973) methodology. The Fama and MacBeth's (1973) t-statistic show that all the risk premiums are significant including the intercept. However when Shanken's (1992) corrected standard errors are calculated, it is found that only the innovations to dividend yield is priced with significant negative risk premium. The hypothesis of jointly is pricing errors also rejected. -zero

When the model is estimated using the GMM on the beta representation following Jagannathan and Wang's (2002), the results in Panel (B) reveal that in addition to the innovations to dividend yield, the innovations to term spread also significantly priced are not

with negative risk premium, but the innovations to the default spread and T-Bill significant.

A point to be mentioned here is that although the risk premiums of the market to default spread are not significant, their signs change (1993) estimation.

excess returns and innovations under the GMM estimation

compared with Fama and MacBeth's is accepted by Hansen's J-statistic.

Finally the model specification

The innovations in the potential state variables in Table (2.4) are estimated as Petkova (2006) from a VAR system. To check the robustness of the results to the technique used to produce the innovations of the candidate risk factors, Table (2.5) reports results when the innovations are estimated based on Kalman Filter technique that is suggested by Priestley (1996) to replace the VAR innovations in Table (2.4). Panel (A)

83

is from Fama It MacBeth's (1973) the regression. clear and cross-sectional results shows that the final conclusion is qualitatively close to its VAR counterpart in Panel (A) of

Table (2.4). However, using the Kalman filter-based innovations, it is found that only the innovations to dividend yield and terms spread are priced risk factors under the Fama and MacBeth's (1973) t-statistics and when Shanken's (1992) corrected standard error is to dividend yield is priced. A significant difference filter

calculated, only the innovations

between the results based on VAR innovations innovations,

innovations

and those based on Kalman

is the hypothesis of jointly-zero this may imply that Kalman

pricing filter

errors is accepted for the latter are potentially superior

innovations

estimates of the risk factors than VAR innovations.

Panel (B) of Table (2.5) shows the estimation using the GMM method. The results support those reported in Panel (A) using Fama and MacBeth's regression. Nevertheless, the innovations to the T-Bill (1973) cross-sectional with positive risk to the T-Bill is

is significant

premium. The positive sign of the risk premium of the innovations

surprising given it was negative in all the previous estimations although insignificant. This may cast doubt on the importance of the short-term interest rate as a useful risk factor in the cross section of the UK stock returns.

Given the above, it could be concluded that innovations to the dividend yield and term spread are priced risk factors in the UK, a result that is robust to the innovation estimation technique and the model estimation method. In addition, the Fama and

French's (1993) model, when applied to UK stock return, is rejected under the hypothesis of jointly-zero Fama is the using and MacBeth's error estimated pricing when model

84

(1973) methodology,

innovations Petkova's (2006) and market return while model with is accepted for the UK market filter-based technique. These of the

to dividend yield, term spread, default spread and T-Bill when the innovations are estimated using the Kalman

results may be interpreted to mean that the latter model is a better description cross-sectional

returns on the Fama and French's (1993) 25 size and book-to-market

for fact is higher in is UK This despite RZ the that the the the market. adjusted portfolios Fama and French's (1993) model.

In comparison, innovations to T-Bill

Petkova (2006) reports that the innovations are significantly

to term spread and

priced in the US market. She reports the risk

premium of the innovations to term spread is positive for the US while this chapter finds a negative risk premium for the UK market. Chen et al., (1986) report the risk premium associated with the shocks to the term spread is negative. However Chen et al., (1986) pointed out this is should be read in the light of the fact that inflation is already accounted for. Antoniou but insignificant (1998) a positive et al., report risk premium for the

Kalman filter- based shocks to term spread and a positive and significant risk premium for the market portfolio for the UK in the context of APT of Ross (1976). Furthermore, factors have significant risk

they report that a number of shocks to macroeconomic

premium in the UK market, including, among others, default spread. However they report that its risk premium is not stable. Clare and Thomas (1994) also study the shocks to a number of macroeconomic in UK. They factors the report, risk variables as potential default is, the spread with positive risk variables risk premiums for the unemployment and

among the priced macroeconomic

premium. However they report insignificant

term spread (both non stable sign). In addition, Petkova (2006) reports that the jointly

85

zero pricing error hypothesis is rejected for the Fama and French's (1993) model but accepted for her model innovations to dividend with the following state variables; market excess return and is

yield, term spread, default spread and short term T-Bill

accepted. Note that Petkova (2006) reports the zero error for her model with innovation estimated from the VAR system while in this chapter her model is accepted on this basis, when the Kalman filter innovations replace VAR innovations while the model with the latter innovations is rejected in the UK.

2.5.3

Fama and French's

(1993) Factors and State Variables;

Do they Relate?

Having found (1) that Fama and French's (1993) model is able to explain the cross sectional UK stock returns and (2) Petkova's (2006) model with market excess return and innovations to dividend yield, term spread, default spread and T-Bill does well in pricing the Fama and French's (1993) 25 size and book-to-market portfolios, the next step is to

test hypothesis (3) which also follows Petkova (2006) and examines if the innovations in her selected state variables can drive the Fama and French's (1993) SMB and HML factors. Therefore, this section uses Petkova's (2006) model with market excess return, innovations to dividend yield, term spread, default-spread, one-month interest rate and Fama-French's (1993) HML and SMB as risk factors. Table (2.6) presents the results of Petkova's (2006) model with market excess return, innovations to dividend yield, term spread, default-spread, one-month interest rate and Fama-French's (1993) HML and SMB as risk factors. Panel (A) reports the estimates using Fama-MacBeth's (1973) cross-sectional regression. It is clear that the HML is still In

highly significant with positive risk premium, while the SMB becomes insignificant.

86

addition, it is found that only the innovations to dividend yield is significant while the innovations to term spread is insignificant. the hypothesis of jointly zero pricing errors is rejected.

as before

The adjusted R2 is high although

Panel (B) of Table (2.6) shows the results form estimating the model as has been done for the models in the above sections of the results using the GMM representation following on the beta

Jagannathan and Wang (2002). The results show that Fama and do not lose their influence on stock returns in the with positive risk premium

French's (1993) SMB and HML

existence of the other risk factors. They are still significant

for the HML and negative risk premium for the SMB. In addition, the innovations to term spread and the innovations to dividend yield are also significant, even more the

innovations to T-Bill is significantly

priced with negative sign. The model specification is

accepted as shown by Hansen's J-statistics.

Panel (A) of Table (2.7) presents the results when the Kalman filter innovations replace the VAR innovations in Table (2.6) and with HML and SMB real returns replace the innovations to HML and SMB. Panel (A) shows the results for Fama and MacBeth's (1973) cross-sectional regression. The results are qualitatively similar to those with VAR

innovations in Panel (A) of Table (2.6), although the SMB is found to be negatively priced. The hypothesis of jointly-zero pricing errors is rejected but the test statistic is

slightly smaller than in the case of the VAR innovations. Remember that in Table (2.5), when the model has only the market excess returns with innovations to the four variables, the jointly zero pricing errors is accepted for the Kalman-based innovations. Taken

together, this may suggest that adding the HML and SMB to the model increases the

87

deviations of the 25 Fama and French (1993) size and book-to-market

expected returns

from their actual values, despite the fact that the R2 is much higher when the HML and SMB exist.

Panel (B) of Table (2.7) shows the results from the GMM estimation. They support the results from Panel (A) however, now, as in the case of the VAR innovations when the model is estimated with the GMM, the term spread is significant, even more the T-Bill also significantly is

priced, though it differs from its VAR counterpart in the sign as the risk

premium is now positive. But this positive T-Bill risk premium is consistent with the sign in Table (2.5). It seems that the one-month T-Bill risk premium is not stable; it is

negative under the VAR innovations while positive under the Kalman Filter innovations.

Jagannathan and Wang (2002) point out that when the factor is traded then the restriction of the equivalency of the risk premiums of the factors to their corresponding is it is SMB, for This hypothesis HML the while average returns applies. accepted and marginally rejected for the market portfolio at 10% significance level. For the GMM

estimates the hypothesis that each of the market, HML and SMB risk premium estimate is equal to its corresponding return average, is accepted.

It could be concluded that innovations to dividend yield, term spread, HML and SMB are important risk factors that drive the cross-sectional returns on the UK Fama and French's (1993) 25 size and book-to-market SMB continue to be significant value portfolios. However, the HML and

factors in the existence of the other variables. This captured by the

information in UK, SMB do HML the the that, share and not suggests

88

innovations to the dividend yield, term spread, default spread and one month interest rate. Hence the best description of the UK returns is a model that contains excess market return, innovations to dividend yield, term spread, HML and SMB. This is unlike Petkova's (2006) findings for the US. She reports that her findings accept the hypothesis of the HML and SMB lose their influence to explain stock returns cross sectionally to the innovations to dividend yield, term spread, default spread and T-Bill, where the last four factors capture the information in these two factors.

2.5.4

The Common Macro Factors and the Stock Returns

The results of -the previous section show that the UK Fama and French's (1993) SMB and HML are not explained by the innovations to Petkova's (2006) selected state variables. This does not answer the challenge that is posited by Cochrane (2006) which is to find the economic factors that are tracked by the SMB and HML portfolios. Therefore, these findings point at the possibility of the problem of omitted information as pointed out by Ludvigson and Ng (2007) and they point out that dynamic factor models solve such a problem. Therefore, this section applies the testing procedures, that are applied so far in the chapter, to factors estimated from large set of macroeconomic variables utilizing the dynamic factor models and principal components analysis of Stock and Watson (2002a, b), in an attempt to uncover these potential economic risk factors. Mnch (2004,2006) also apply the dynamic factor models of Stock and Watson (2002a, b) in asset pricing as well.

89

2.5.4.1 Are the Common Macro Factors Pricing Factors?

To test the Hypothesis (4A) that the innovations to factors estimated from a large macro set are potential risk factors, six factors are estimated from a set of 78 UK macroeconomic variables using the dynamic factor models and principal components analysis of Stock and Watson (2002a, b). Petkova (2006) uses VAR to estimate the innovations of the state variables by including the market excess return as the first variable in the VAR model followed by the rest of the state variables. Following her, this section estimates the innovations to the estimated six factors from the VAR by including the market excess return first, followed by the six estimated factors. As mentioned earlier, adding estimated factors into the VAR this is also similar to Lagan and Mountford (2005) who point out that they add factors estimated from a large set of macro variables into the VAR. In addition, innovations in the estimated factors are estimated using the Kalman filter based method following Priestley (1996). The reason for choosing to estimate six factors is based on Artis, Banerjee and Marcellino (2005). They report that they found that half of the variation in their 81 UK macroeconomic variables could be accounted for by six common factors.

Table (2.8) reports the results of estimating the model in which the innovations in the six estimated factors replace the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term (A) interest Panel default shows the results of Fama rate. spread and one month spread, four factors Based (1973) MacBeth's their t-statistics, on cross-sectional regression. and fifth factors. innovations first, found the the the to the the and sixth second, priced; are

90

However, when Shanken's (1992) corrected standard errors are calculated, none of the innovations to these factors are significant. Although R2 is very low, the zero pricing errors are accepted. Panel (B) of Table (2.8) shows the results of the GMM estimation following Jagannathan and Wang (2002). These results support the significance of the first and the sixth factors which are found to be priced by the Fama and French's (1993) 25 size and book-to-market portfolios. In addition, the model specification is accepted.

Table (2.9) reports the results for the same model in Table (2.8) but with Kalman Filter based- innovations substituting VAR-based innovations. The results from both Fama and MacBeth's (1973) cross-sectional regression (Panel (A)) and the GMM estimation (Panel (B)) support the above findings that the innovations to the first and the sixth estimated factors are potentially priced risk factors. Similarly as in the VAR-based innovations the jointly zero-pricing errors and Hansen J-statistic are accepted however RZ is much higher. A point to be mentioned is that innovations based on the Kalman filter seem to be more robust than those from the VAR.

The finding surprising

that not all the common and Ng (2007)

factors are significant

risk factors is not factors are

as Ludvigson

point out not all the estimated

necessarily useful for forecasting the conditional to here as well.

mean of returns. Their argument applies

Indeed this is what should occur as Cochrane (2006) points out that the Cochrane (2006)

degrees of freedom are 3 and not 25 in Fama and French's portfolios. points out further that this is Lewellen, Titman's (2005) essential point.

Nagel, and Shanken's (2006) and Daniel and

91

Furthermore, Mnch (2004) reports that his model which includes market returns and two diffusion indexes (estimated factors) which relate to business cycle and foreign exchange risk does better than Campbell (1996) model and as well as Fama and French's (1993) model on the Fama and French's (1993) 25 portfolios.

2.5.4.2 Do the Priced Common Macro Factors Relate to HML and SMB?

Given that Hypothesis (4A) is accepted as it is found that innovations to at least two
estimated factors are priced by the Fama and French's (1993) 25 size and book-to-market portfolios, this section turns its attention to testing Hypothesis (1993) S MB common and HML factors. (4B) which examines information to the to

whether the Fama-French's innovations

share similar (4B)

in the estimated

Hypothesis

is equivalent

Hypothesis (3). Therefore this section uses Petkova's (2006) model with excess market

returns and innovations to dividend yield, term spread, default spread, one month interest rate, HML and SMB where the innovations in the estimated factors replace the

innovations to dividend yield, term spread, default spread, and one month interest rate. The innovations in the estimated factors that are used include the innovations to the above four potentially priced estimated factors; the first, the second, the fifth and the sixth factors. Furthermore, the HML and SMB real returns are used rather than the innovations. Table (2.10) presentsthe results.

Panel (A) of Table (2.10) reports the results of estimating the model using the Fama and MacBeth's (1973) method. It shows that only the HML is significantly and positively SMB is insignificant levels. In the coefficient at conventional marginally priced and

92

found is it that the market and SMB estimated risk premiums are equal to their addition corresponding averages, while the hypothesis that the HML risk premium is equal to its average is rejected at 5%. Panel (B) reports the results from the GMM on the beta representation following Jagannathanand Wang's (2002). It shows that innovations to the second factor is priced as well as the HML and SMB. Note that the risk premium associated with the innovations in the second factor is negative while previously in Table (2.8) (without HML and SMB) the sign is found to be positive. Again the hypotheses that the market and SMB estimated risk premiums are equal to their corresponding averages are accepted while it is rejected for the HML at 10%.

Panels (C and D) report the results when the insignificant

estimated factors from to the in the

above results; i. e. the innovations to five estimated factors except the innovations second factor, are excluded. The results support the importance of the innovations

second estimated factor along with HML and SMB as potential risk factors. Furthermore, the jointly-zero pricing errors is marginally rejected at 10%, and the model specification

measured by Hansen J-statistics is accepted. In addition, it is found that market and SMB estimated risk premiums are equal to their corresponding for the HML premium is rejected at 5%. However, averages, while the hypothesis estimation, the

under the GMM

hypothesis that the estimated risk premium is equal to its corresponding is accepted for all the factors; the market, HML and SMB risk premiums.

average returns

Table (2.11) presents the results when the VAR innovations in Table (2.10) are filter by Kalman based innovations. It is clear that the only risk factors that replaced continue to be significant are the HML and marginally, based on the Fama and

93

MacBeth's (1973) cross-sectional regression, the SMB. Accordingly, it seems that the significance of the innovations to the second estimated factors is not robust to the innovation estimation technique. The hypothesis that the estimated risk premium of the traded risk factor is equal to its average return is accepted for the SMB while it is rejected for the market and HML, both at 5% significance level. Under the GMM, it is accepted for the market and SMB, and it is rejected for the HML at 10% significance level.

Panels (C and D) of Table (2.10) show the VAR based innovations to the second estimated factor is significant along with the HML and SMB. This finding is robust to the model's estimation method. This motivates replacing the VAR based innovations to the second estimated factor in Panels (C and D) of Table (2.10) with the Kalman filter based innovations to the second estimated factor to examine if this estimated factor is robust to the innovation estimation technique, although it is found to be insignificant in Table (2.11). Panel (A) and Panel (B) of Table (2.12) present the results from the Fama and MacBeth's (973) cross sectional regressions and Jagannathan and Wang's (2002) GMM method on the beta representation, respectively. The innovations to the second estimated
factor is significant hypothesis with a negative risk premium risk premium under the GMM factor method and the is equal to its

that the estimated

of the traded

corresponding

average is accepted for the SMB, but it is rejected for the market and level, respectively.

HML, at 10% and I% significance

This chapter concludes that the innovations to the first and the sixth common factors are potential significant risk factors in the UK stock market and these results are robust to the innovations estimation technique and the model estimation method.

94

However, the results show that in the existence of the HML and SMB, it seems that the innovations in the second estimated factor is a potential significant risk factor in the cross section of UK Fama and French's (1993) 25 portfolios, nevertheless, it does not account for the information in the HML and SMB as they continue to be significant. Therefore Hypothesis (4B) is rejected. This is in contrast with the findings of Liew and Vassalou (2000) and Kelly (2003) who report the SMB and HML are linked to GDP growth.

2.5.5

The Macro Common Factors; Do they contain different news to stock prices?

The question now is whether the estimated common factors that are found priced in
the cross section of the UK stock return in the previous sub-section carry information for

stock returns that is not captured by Petkova's (2006) four selected state variables. In other words, this section examines if there are other variables that are valuable for stock's returns that are missing from Petkova's (2006) set of variables19. The aim of this chapter is to search for the potential macroeconomic factors that explain the changes in stock

prices, which is the challenge that is posited by Campbell (2000) and Cochrane (2006). So in this context, Petkova's (2006) selected variables are just few potential risk factors. Mnch (2004) augments Campbell's diffusion (1996) model with estimated factors which he calls

indexes and reports the estimated macro factors add additional risk information

beyond those of the model of Campbell (1996). Mnch (2006) augments the conditioning variables set with factors estimated from a large set of variables. Furthermore, Ludvigson

19Watson (2001) points out that, in reality more variables need to be used than the limited number that can be handled by the VAR. Lagan and Mountford (2005) point out they follow Bernanke et al. (2005) in augmenting the VAR model with factors estimated from a large set of macro variables to study the monetary policy.

95

factors do Ng (2007) that their the out point estimated carry and study examines whether information that is not captured by the other widespread used predictive variables for predicting the conditional mean and volatility of stock returns. They point out for this purpose they augment the conditional set of variables to predict the conditional mean and volatility of stock returns with factors estimated from a large set of variables. This is section of the chapter follows Ludvigson and Ng's (2007) and Mnch's (2004,2006) procedure in augmenting the potential risk factors with factors estimated from a large set of macro variables. More specifically it augments Petkova's (2006) model that includes excess market return, innovations in dividend yield and term spread, HML and SMB with the innovations to the second factor estimated from a large set of macro variables. Note that the other two variables in Petkova's (2006) model which are the innovations in the default spread and one month T-bill are not included as they are found previously in this chapter are insignificant in the UK market. Table (2.13) presents the results. Panels (A) and (B) show the estimates with VAR innovations using Fama and MacBeth's (1973) method and GMM on beta representation following Jagannathan and Wang (2002), respectively. Panels (C) and (D) show the corresponding results when Kalman filterbased innovations replace VAR innovations.

The results show that the innovations to dividend yield, HML and SMB continue to be significant regardless of the innovation estimation technique or the model estimation under the GMM method regardless to the second macro

method. The innovations in term spread is significant of the way the innovations estimated factor is significant technique

are estimated and the innovations

when the innovations are estimated using the Kalman filter using the GMM method as Panel (D) shows.

and the model is estimated

96

Indeed Panel (D) shows that all the factors are significantly priced, even more, the hypothesis that the factor estimated risk premium equals to its corresponding average return is accepted for the market, HML and SMB. Based on Fama and MacBeth's (1973) methodology, the hypothesis of zero jointly pricing error is rejected but under the GMM estimation the model specification is accepted as before.

In summary, the results from Table (2.13) suggest that the innovations in the term spread and the innovations in the second estimated macro factors share some information about stock prices, evidenced by the fact that in some of the estimations one of them loses power when both exist in the model. To shed more light on this, this chapter follows Stock and Watson (2002a) to examine what the estimated factors do represent.

2.5.6

What is the Priced Estimated

Macro Factor?

Stock and Watson (2002a) point out that in order to characterize the estimated factors, and because of the identification problem, they estimate Rz s from regressing each of the estimated factors on the individual macroeconomic variables that are used to estimate the factors. They interpret the factors by their loadings on the macroeconomic
variables. This chapter applies Stock and Watson's (2002a) exercise to the second

estimated factor by regressing this estimated factor on each of the 78 macroeconomic variables. It is found that the highest R2s are from regressing the second estimated factor (53-58%) (71%), the term and unemployment ; ate rate spread measures on employment (51%). Furthermore, the loadings of the second estimated factor on the employment and the term spread measures are positive while on the unemployment it is negative. This

97

innovations innovations in in the second the term and when spread explains why, estimated factor exist together in the model, one of them or both lose power in some of the model estimations. However both of them are significantly priced when the Kalman filter -based innovations to the second estimated factor is used and the model is estimated by the GMM.

lt could be concluded that the innovations in the estimated factors from a large dataset of macroeconomic variables bear important relationship to stock returns in the UK market. Even more the second estimated factor, which loads largely on employment and term spread, carry information that is not captured by the innovations to the state variables in Petkova's (2006) model which includes excess market return and innovations to dividend yield, term spread, default spread, short-term interest rates, HML and SMB, or by the Fama and French's (1993) HML and SMB. This new information seems to be related to shocks to the employment and unemployment. These findings are consistent with a number of studies. Mnch (2004) reports when he augments Campbell's (1996) model with the two estimated factors, the resulting model does a little better than his diffusion index model and much better than Campbell's (1996) model. Mnch (2006) reports that the estimated factors are important conditioning factors for stock returns and these estimated factors carry information beyond those contained in the widespread conditioning variables. He reports that one of the estimated factors is related to a number of variables including the term spread where the latter is consistent with the finding in this chapter. Mnch (2004,2006) also usesthe Fama and French's (1993) 25 portfolios.

98

Furthermore, common

Ludvigson

and Ng (2007) point out that they find the estimated stock market's conditional volatility is related to the

factor that forecast

employment employment Furthermore, factor

and real output. Flannery and Protopapadakis

(2002) report that they find variance.

is a potential risk factor that influences stock returns' conditional

Mnch (2004) report that he finds, for the US, the first priced estimated to unemployment, interest rate spreads and capacity utilization.

is related

Furthermore, (negative)

Boyd,

Hu and Jagannathan (2005) point out that they find a positive by stock market prices to the increase in unemployment during

reaction

expansion (recession) and as expansion times dominate the recession times, the general reaction is positive.

2.5.7

Visual Representation

Petkova (2006) depicts the fitted expected return against the actual average return for each of the 25 portfolios for her model which includes excess market return and and for visual models. both the

innovations to dividend yield, term spread, default spread and one month T-Bill the Fama and French's representation (1993) three-factor model. She explains that this

is informative

about the relative

performance

of the different

Petkova (2006) indicates that if the model fits the data then for all portfolios,

fitted (from the model) and actual average returns should hit the 45 degree line. She explains that the model specification fitted value for each portfolio is calculated from this exercise

follows This the parameters. chapter model's estimating for the Fama and French's (1993) three-factor

her in performing

model as shown in Panel (A) of Figure

(2.2) and for the final model which includes all the priced risk factors that are found in

99

this chapter to be priced in the UK market as shown in Panel (B) of Figure (2.2). This final model is Petkova's (2006) model that includes excess market return, innovations to dividend yield, term spread, HML and SMB [but without innovations to default spread and one month T-Bill as they are found not to be significant in the UK ], augmented with the innovations to the second estimated factor (that relates to employment,

unemployment and term spread). Note that the. Panel (C) of Figure (2.2) is for the same model in Panel (B) but Kalman Filter innovations replace VAR innovations.

Figure (2.2) shows that the points that represent the 25 portfolios are slightly closer to the 45-degree line in Panels (B and C) than in Panel (A). The calculated correlations between the fitted and actual returns for the portfolios are 0.92,0.97 and 0.96 for Panels (A, B and C) respectively. This supports that the model in Panels (B and C) seem to provide a better description of the average returns on the Fama and French's (1993) 25 portfolio than the Fama and French's (1993) three-factor model. Petkova (2006) reports that her model with excess market return, innovations to dividend yield, term spread, default spread and one month T-Bill outperforms the Fama and French's (1993) threefactor model in producing closer points for the 25 portfolios to 45 degree-line especially for those portfolios that defy the Fama and French's (1993) three-factor model. In addition, Hodrick and Zhang (2001) report that the return on some of the 25 size and book-to-market portfolios is overstated by the Fama and French's (1993) three factor model and in particular, for the small growth one. Furthermore, Hodrick and Zhang (2001) point out that, what they call, Fama and French's (1993) five factor model that has extra two factors which are the term spread a nd default spread has slightly smaller pricing error compared with the three-factor model.

100

However as reported earlier the hypothesis of jointly zero pricing error is rejected for this model, although under the GMM estimation the model specification is accepted using Hansen's J-statistics. However, Lettau and Ludvigson (2001) emphasize that there is a small-size problem of the Wald statistic.

2.6

Conclusion

This chapter shows that Fama and French's (1993) three-factor model and Petkova's (2006) model of the ICAPM of Merton (1973) are successful in explaining the cross sectional returns of the UK 25 Fama and French's (1993) 25 size and book-tomarket portfolios. It is found that the innovations in dividend yield and term spread are significantly priced risk factors, but the innovations in default spread and one-month TBill are not significantly priced. However, these state variables are not able to capture the pricing power of the HML and SMB, which continue to be significantly priced in the UK. The results are robust to the innovation estimation technique and the model estimation method.

This

chapter

augments

Petkova's

(2006)

model

with

innovations

to factors

estimated from a large panel of 78 macroeconomic

variables using the dynamic factor

model and principal components analysis of Stock and Watson (2002a, b). To uncover the macroeconomic forces that explain stock returns as required by Campbell (2000) and

Cochrane (2006), it is important not to confine the analysis to a small set of potential risk factors, but to a search in the wider set of the macroeconomy. This becomes important in

101

z or

light of the problem of omitted information (Ludvigson and Ng (2007)) and the findings is found factors factors in It Mnch (2004). that two are priced although estimated risk of the cross section of returns on stocks, one estimated factor (the second factor which relates to employment and term spread) is found to add information beyond those carried by the innovations to dividend yield, term spread, HML and SMB. In addition it is found that this second macro estimated factor is related to employment and term spread measures.This is consistent with Mnch (2004) who reports that augmenting Campbell's (1996) model with estimated factors provides better model than Campbell (1996) model. However, this chapter finds no link between the SMB and HML and the macroeconomic variables which contradicts the findings of Liew and Vassalou (2000) and Kelly (2003).

This chapter concludes that the potential best representation of the asset pricing model in the UK stock market is Petkova's (2006) model with excess market return, innovations to dividend yield, term spread, and HML and SMB [but without innovations to default spread and one month T-Bill as they are found not to be significant in the UK],
augmented with the innovations This has important implications to the second estimated factor relates to employment. to those who are interested in estimating average returns It warns them to not ignore the larger set of

on stocks that are traded in LSE. macroeconomic

variables when they consider the potential determinants of stock prices.

This chapter contributes by adding another attempt to the asset pricing literature that tries to address the challenges that are posited by Campbell (2000) and Cochrane (2006).

102

Table 2.1 Number

Year 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992

of Stock in the Sample

Number of firms per year 205 217 224 238 315 375 408 710 947 1023 1059 961 Year 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Number of firms per year 972 1006 1002 1149 1242 1186 1101 1187 1211 1134 1077 ; 143

1993 959 This table reports the number of UK stocks available for analysis in each year in the sample period of July-1981 to December-2005 after applying Fama and French's (1993) criteria regarding the availability of stock prices and book values. The data are from Datastream for non-financial stocks traded i n London Stock Exchange excluding foreign stocks

Table 2.2 Average Returns

on UK Fama and French's

(1993) Factors

HML SMB 0.59 -0.37 Average (%) (5.64)*** (-1.92)* This table reports the monthly average excess returns on the market portfolio, the average returns on the SMB and HML portfolios. The t-statistics is reported in parentheses. The portfolios constructed following Fama and French (1993 and 1996). The study sample period is July-1981 to December-2005. ***significant at 1%, ** at 5% and * at 10% Market 0.07 (0.24)

103

Table 2.3 Fama and French's (1993) Three-Factor Model

Pane! A: Fama and MacBeth (1973) Pane/B: GMM

Parameters Intercept -0.87 (-1.68) [-0.70] 0.96 (1.63) [1.37] -0.47 (-2.02)** [-1.86*] 0.93 (5.14 )*** [4.43 ***] 82%
44.52 (0.00)

/=s
0.07

Parameters

it=s
0.07

Market

(1.51)
-0.37 (-0.46) 0.59 (1.88)*

(0.40)

0.11

(0.15)
-0.37 (-0.56) 0.59 (0.99)

SMB

-0.47 (-2.48)** 0.69 (6.88)***

HML

Adjusted R'
x2

Hansen J-

28.80

statistic This table reports the parameters estimates of Fama and French (1993) three-factor model. Panel (A) reports the estimates from Fama-MacBeth (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios and the risk factors are market excess return (Market) and returns on SMB and HML. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken's (1992) corrected t-statistics is reported in square brackets. u is the average return on the traded risk factor i, 45 is the estimated risk premium for traded factor i and reported in the previous column, and /4 =8 is the restriction that Jagannathan and Wang (2002) impose on the risk premium when the factor is traded under the beta method. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R2.***significant at 1%, at 5% and * at 10%

104

Table 2.4 Petkova's (2006) Model of ICAPM : Five Risk Factors - VAR
Panel A: Fama and MacBeth (1973) Intercept -1.65 (-2.71)*** [-0.76] 1.32 (2.15)** [ 1.08] -6.39 (-5.53)*** [-2.60]*** -3.25 (-2.62)*** [-1.23] -2.71 (-1,88)* [-0.87] -3.65 (-2.87)*** [-1.34] 69% 36.77 (0.01) Panel B: GMM

Market

06 -0 . (-0.20) 67 -5 . (-3.15)*** 21 -8 . (-3.27)*** 1 30 . (0.55) 91 -1 . (-1.08)

FTDIV

TERM

DEFAULT

TBILL

Adjusted R2

xz Hansen J-

11.47 (0.65) statistic This table reports the parameters estimates of Petkova's (2006) model which includes the excess market returns (Market), innovations in dividend yield (FTDIV), term spread (TERM), default spread (DEFAULT), and one month T-Bill (TBILL), where the innovations are estimated from VAR model. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken's (1992) corrected t-statistics is reported in square brackets. ,r2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R' is the adjusted cross sectional R2. ***significant at 1%, ** at 5% and * at 10%

105

Table 2.5 Petkova's (2006) Model of ICAPM: Five Risk Factors - Kalman Filter
Pane! A: Fama and MacBeth (1973) Pane/B: GMM

Intercept

Market

-0.59 (-1.03) [-0.25] 0.38 (0.64) [0.30] -0.84 (-4.12)*** [-1.79] -0.57 (-2.04)** [-0.87] -0.25 -0,95 [-0.40] -0.02 -1.18 [-0.50] 68% 5.13

07 -0 . (-0.23) -0.48 (-1.97) 35 -2 . (-3.22)*** 0 15 . (0.32) 0.07 (2.12)

FTDIV

TERM

DEFAULT

TBILL Adjusted R2

x2
Hansen Jstatistic

(1.00)
I 7.25 (0.92)

This table reports the parameters estimates of Petkova's (2006) model which includes the excess market returns (Market), innovations in dividend yield (FTDIV), term spread (TERM), default spread (DEFAULT), and one month T-Bill (TBILL), however the innovations are estimated based on KalmanFilter technique. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth (1973) method, the Fama and MacBeth (1973) t-statistics is reported in parentheses and Shanken (1992) corrected t-statistics is reported in square brackets. r2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests , for model specification. R2is the adjusted cross sectional R2. ***significant at 1%, ** at 5% and * at 10%

106

Table 2.6 Petkova's (2006) Model of ICAPM: Seven Risk Factors - VAR
Panel A: Fama and MacBeth (1973) Intercept -1.05 (-2.05)** [-0.67] 1.08 (1.85) * [1.261 -3.57 (-3.21)*** [-2.08] ** -2.27 (-2.20) ** [-1.43] -0.71 (-0.54) [-0,35] -1.92 (-1.58) [-1.02] 2.26 (3.19) *** [2.13]** -0.63 (-1.74) * [-1.33] Adjusted R z x
Hansen Jstatistic This table reports the parameters estimates of Petkova's (2006) model

Panel B: GMM

Market

0.07 (0.24)

FTDIV

-3.73 (-3.03)*** 69 -4 . (-2.94) *** 0 02 . (0.01) -2.53 69 ()*1.69 . 2 07 . (2.33)** 95 -0 . (-2.1 1)* 1) .1

TERM

DEFAULT

HML

93% 40.74 (0.00)

8.25 (0.97) which includes the excess market

returns (Market), innovations in dividend yield (FTDIV), term spread (TERM), default spread (DEFAULT), one month T-Bill (TBILL), HML and SMB where the innovations are estimated from VAR model. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) tstatistics is reported in parentheses and Shanken (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R. ***significant at 1%, ** at 5% and * at 10%

107

Table 2.7 Petkova's (2006) Model of ICAPM: Seven Risk Factors - Kalman Filter
Panel A: Fama and MacBeth (1973)
Intercept (-1.88)

Panel B: GMM

-1.06

Market

[-0.72] 1.10 (1.74) * [1.25] -0.40 (-2.39) [-1.67]* -0.63 (-2.27) ** [-1.55] -0.28 (-1.03) [-0.71 ] 0.02 (1.17) [0.80] 0.85 (4.64) *** [3.45] *** -0.62 (-2.57) ** [-2.13] **
94%

-0.05 (-0.16) 45 -0 . (-2.14) ** 43 -1 . (-2.52)** 08 -0 . (-0.20) 0 04 . (1.68)* 0 42 . (1.89)* 30 -0 . (-2.99) ***

FTDIV

TERM

DEFAULT

TBILL

HML

SMB Adjusted R2

i2 ,
Hansen Jstatistic

38.69 (0.00)
I 6.68 (0.99)

This table reports the parameters estimates of Petkova's (2006) model which includes the excess market returns (Market), innovations in dividend yield (FTDIV), term spread (TERM), default spread (DEFAULT), one month T-Bill (TBILL), HML and SMB but the innovations are estimated based on Kalman filter technique. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assetsare UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth (1973) t-statistics is reported in parenthesesand Shanken's (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R2.***significant at 1%, ** at 5% and * at 10%

108

Table 2.8 Estimated Factors as Risk Factors - VAR

Panel A: Fama and MacBeth (1973) Panel B: GMM

Intercept

Market

DFI

DF2

-0.53 (-0.89) [-0.23] 0.34 (0.56) [0.27] 7.50 (3.59) *** [1.58] 1.75 (I. 72) * [0.77] -0.20 (-0.14) [-0.06] -1.00 (-0.73) [-0.33] 3.81 (2.04) ** [0.90] 2.72 (2.35) ** [1.05]
29%

20 -0 . (-0.69) 10 86 . (2.52)** 0 14 . (0.07) 3 35 . (1.07) 2 72 . (1.01) 3 56 . (1.15) 4 17 . (1 71)*

DF3

DF4

DF5

DF6 Adjusted RZ z X Hansen Jstatistic

5.50 (1.00) 12.67

(0.81)

This table reports the parameters estimates when innovations to six estimated factors (F1-F6) replace the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term spread, default spread and one month interest rate. The six factors are estimated from a large set of 78 macroeconomic variables using the dynamic factor model and principal components of Stock and Watson (2002a,b). The innovations are estimated from VAR model. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R2.***significant at 1%, ** at 5% and * at 10%

109

Table 2.9 Estimated Factors as Risk Factors - Kalman filter

Panel A: Fama and MacBeth (1973) Intercept 0.45 (0.83) [0.21] -0.44 (-0.74) [-0.38] 0.10 (3.64)*** [1.691* 0.02 (0.60) [0.28] -0.05 (-1.21) [-0.57] 0.01 (0.16) [0.07] -0.01 (-0.12) [-0.06] 0.17 (3.54)*** [1.66]* 75%
4.45 (1.00)

Panel B: GMM

Market

0 11 . (0.36) 0.21 (2.09)** 0 00 . (0.02) 02 -0 . (-0.16) 0.29 (1.45) 0.07 (0.49) 0.26 (1.64)

Fl

F2

F3

F4

F5

F6

Adjusted R2

xZ

12.35 (0.83) statistic This table reports the parameters estimates when innovations to six estimated factors (Fl-F6) replace the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term spread, default spread and one month interest rate. The six factors are estimated from a large set of 78 macroeconomic variables using the dynamic factor model and principal components of Stock and Watson (2002a,b). The innovations are estimated based on Kalman filter technique. Panel (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. RZis the adjusted cross sectional R2 '**significant at 1%, ** at 5% and * at 10%

Hansen J-

110

Table 2.10 Estimated Factors as Risk Factors and Fama and French's (1993) HML and SMB
Pane! A: Fama and MacBeth (1973) Pane! B: GMM Panel C: Fama and MacBeth (1973) Panel D: GMM

Intercept

F1

-0.93 (-1.60) [-0.68] 0.95 (1.48) [1.13] 1.42 (0.87) [0.64] -1.85 [-1.42] 0.42 (0.29) [0.21] 0.67 (0.61) [0.45] 1.00 (5.15)*** [4.06] *** (-1..77)* [-1.56] 92% o 33.45

-0.79 (-1.50) [-0.62] 0.18 18 . (0.65) 1.16 (0.93) 57 -2 . (-2.89)*. * 02 -0 . (-0.01) 1.58 (1.29) 1 02 . (4.33)*** 39 0 - . (-1.65)* 1.01 (5.33)*** [4.42]*** (-198)** [-1.79]* 92% e 29.72 0 83 . (4.07)*** -0 .40 (-1.79)* (-219)** [-1.73]* 51 -3 . (-3.64)*** 40) . (10.84 [1.13] 0.20 (0.73)

F2

F5

F6

HML

SMB Adjusted R2

xZ
Hansen
J-statistic

(0.01)
18.23
(0.44)

(0.10)
17.45
(0.68)

Panels (A and B) of this table report the parameters est imates when the innovations in the four estin estimates estimated factors (Fl, F2, F5 and F6) replace the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term spread, default spread, one month interest rate, HML and SMB. The factors are estimated from a large set of 78 macroeconomic variables using the dynamic factor model and principal components of Stock and Watson (2002a, b). More specifically these are four out of six factors used earlier. The innovations are estimated from VAR model while the HML and SMB are real returns. Panels (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from the GMM on beta representation following Jagannathan and Wang (2002). Panels (C and D) estimate the same model in Panels (A and B) after excluding the estimated factors that are not significant. Panels (C) reports the estimates from Fama-MacBeth's (1973) crosssectional regressions with betas estimated using one multiple regression over the full sample period. (D) report the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken's (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen

(1982) J-statistic tests for model specification. R2 is the adjusted cross sectional R` ***signifiicant at 1%, ** at 5% and * at 10%

112

Table 2.11 Estimated Factors as Risk Factors and Fama and French's (1993) HML and SMB-Kalman filter
Panel A: Fama and MacBeth (19 73) Intercept -1.24 (-2.27)** [-0.83] 1.28 (2.07)** [ 1.47] 0.05 (1.70)* [1.15] -0.01 (-0.49) [-0.34] 0.07 (1.42) [0.96]
0.00

Panel B: GMM

Market

0 11 . (0.40) 0 04 . (1.61) 02 -0 . (-0.64) 0 04 . (0.86)

Fl

F2

F5

F6

(0.03)
[0.02] 0.98 (4.98)*** [3.64]*** -0.44 (-1.87)* [-1.57] 92%
55.53

(0.18) 0 99 . (4.42)*** -0 .29 (-1.23)

0 01 .

HML

SMB

Adjusted R2
x z

(0.00)

Hansen Jstatistic

40.78 (0.00)

Panels (A and B) of this table report the parameters estimates when the innovations in the four estimated factors (Fl, F2, F5 and F6) replace the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term spread, default spread, one month interest rate, HML and SMB. The factors are estimated from a large set of 78 macroeconomic variables using the dynamic factor model and principal componets analysis of Stock and Watson (2002a, b). More specifically these are four out of six factors used earlier. The innovations are estimated based on Kalman filter technique, while the HML and SMB are real returns. Panels (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth (1973) t-statistics is reported in parentheses and Shanken's (1992) corrected t-statistics is reported in between brackets. %2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2 is the adjusted cross sectional R2 ***significant at 1%, ** at 5% and * at 10%

113

Table 2.12 Estimated Factors as Risk Factors and HML and SMB-Kalman
Pane! A: Fama and Mac Beth (1973) Intercept -1.14 (-2.06)** [-0.84] 1.20 (1.93)* [1.49] -0.03 (-1.26) [-0.95] 1.09 (5.64)*** [4.48]*** -0.53 (-2.24)** [-1.971* 88% 54.95 (0.00 21.49
(0.429)

filter

Panel B: GMM 0.29 (1.07) -0.15 (-2.88)***

Market

F2

HML

0.92 (3.54)***

SMB

-0.44 (-1.84)*

Adjusted R2 2 , Hansen Jstatistic

This table reports the parameters estimates when the innovations in the second estimated factor (F2) replaces the innovations in the four state variables in Petkova's (2006) model that includes excess market returns and innovations to dividend yield, term spread, default spread, one month interest rate, HML and SMB. The second factor estimated from a large set of 78 macroeconomic variables using the dynamic factor model and principal components of Stock and Watson (2002a, b). More specifically this is the second factor out of the six factors used earlier. The innovations are estimated based on Kalman filter technique while the HML and SMB are real returns. This Table is the equivalent to Panels (C and D) in Table (2.10) with the Kalman filter innovations replace VAR innovations. Panels (A) reports the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (B) reports the estimates from GMM on beta representation following Jagannathan and Wang (2002). The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth (1973) t-statistics is reported in parentheses and Shanken (1992) corrected t-statistics is reported in square. XZ tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R2 ***significant at 1%, ** at 5% and * at 10%

114

Table 2.13 Estimated Factors as Risk Factors as Augmenting Factors

Panel A: Fama and MacBeth (1973) Intercept -0.82 (-1.54) [-0.61] 0.86 (1.42) [1.101 (-2.27)** [-1.70] * -1.98 [-1.37] DF2 -1.48 (-1.44) [-1.08] 0.82 (4.0)*** [3.20]*** (-2.46)** [-2.14]** 94% 30.72 (0.04) 7.86 (0.99) Panel B: GMM Panel C: Fama and MacBeth (1973) -1.09 (-1.97)** [-0.78] 0 00 . (0.01) (1.79)* [1.35] (-2.80)*** [-2.09]** (- 42) [-1.03] -0.03 (-1.14) [-0.83] 0.92 (4.82)*** [3.74]*** (-2.45)** [-2.11]** 92% 47.67 (0.00) 8.66 (0.98) 0.08 (0.27) Panel D: GMM

FTADY

(-2.94)*** -5.35 (-2.95)*** 10 -0 . (-0.08) 0 70 . (2.86)*** 72 -0 . (-2.95)***

(-2.56)** -1.76 (-2.34)** -0.08 (-1.68)* 0 72 . (2.47)** 77 -0 . (-2.89)**:

TERM

HML

SMB Adjusted R2 z X Hansen J-statistic

This table reports the parameters estimates of Petkova's (2006) model that includes excess market return, innovations to dividend yield and term spread, HIM and SMB ( without the innovations in the default spread and one month T-bill ) augmented with the innovations to the second estimated factor (F2) that is found to be related to employment and term spread. The second factor is estimated from a large set of 78 macroeconomic variables using the dynamic factor model and the principal components of Stock and Watson (2002a, b). More specifically this is the second factor out of the six factors used earlier. Panels (A) and (C) report the estimates from Fama and MacBeth's (1973) cross-sectional regressions with betas estimated using one multiple regression over the full sample period. Panel (A) shows the results when the factors' innovations are estimated from VAR model and Panel (C) shows the results when the factors' innovations are estimated based on Kalman filter technique. Panels (B) and (D) report the estimates from the GMM on beta representation following Jagannathan and Wang (2002). Panel (B) shows the results where the factors' innovations are estimated from VAR model and Panel (D) shows the results when factors' innovations are estimated based on Kalman filter technique. The study sample period is July-1981 to December-2005. The test assets are UK Fama and French's (1993) 25 portfolios. For the parameters estimated using Fama and MacBeth's (1973) method, the Fama and MacBeth's (1973) t-statistics is reported in parentheses and Shanken's (1992) corrected t-statistics is reported in square brackets. X2 tests the hypothesis of jointly zero pricing error. J- statistic is Hansen (1982) J-statistic tests for model specification. R2is the adjusted cross sectional R2 ***significant at 1%, ** at 5% and * at 10%

115

Figure 2.1 Average Returns on IIML (A): Monthly

15 10
c Li

and SM13

IIML's

Return
FFHML

Q 10 -15

23

45

67 89 111 133 155 177 199 221 243 265 287 Time

(I3): Monthly
15 10

SMB's

Return
FF-SMB

U 5 rn O -5 10 15 1 23 45 67 89 111 133 155 177 199 221 243 265 287 Time
This Figure depicts the monthly average return on the Fama and French's (Panel A) and SMB (Panel B) over time. (1993) IIML

116

Figure 2.2 Fitted Versus Actual 25 Portfolios

Returns on the UK Fama and French (1993)

2 1.5

0.5

w1 0.5 1 1.5 2

-2

-1.5

-1

ff ""

40.5 0 -1 -1.5 -2 Actual Return

117

This figure depicts, for each of the UK Fama and French's (1993) 25 size and book-tomarket portfolios, the fitted return on the y-axis against the actual return on the x-axis. Panel (A) shows the results for Fama and French (1993) three-factor model. Panel (B) shows the results for Petkova's (2006) model that includes excess market return, innovations to dividend yield and term spread, HML and SMB ( without the innovations in the default spread and one month T-bill ) augmented with the innovations to the second estimated factor (F2) that is found to be related to employment and term spread. The innovations are estimated from a VAR model. Panel (C) shows the results for the same model specification in Panel (B) with the Kalamn filter innovations replace VAR innovations. The sample period is from July 1981 to December 2005.

118

Appendix

(A)

This Appendix shows the UK 78 monthly macroeconomic variables used in the chapter. The sample period is July- 1981 to December - 2005. All variables are collected from the Datastream. These variables are (1) those used by Lagan and Mountford20 (2005) covering their macroeconomic categories. However, as this chapter does not manage to retrieve the entire 105 variable used by them for its balanced dataset over the period July December 1981 2005, few by to those to additional variables are obtained used similar (2) Artis, Banerjee and Marcellino21 (2005) (UK) and (3) additional few variables are obtained similar to those of Kapetanios, Labhard, and Price22(200 6) (UK); and (4) variables of UK default spread similar to Ludvigson and Ng (2007) (US) are also included in the dataset. Note that every study of these three studies has constructed a dataset of macroeconomic variables to cover the UK economy. However, this chapter focused on those variables that are used by Lagan and Mountford (2005) and then also used variables from the other two studies to increase the number of variables that can be collected from Datastream. Also note that some variables will be shared by more than one study.
Employment UK LFS: ECONOMIC INACTIVITY RATE, ALL, AGED 16 & OVER SADJ UK LFS: ECONOMIC ACTIVITY RATE, ALL, AGED 16-59/64 SADJ UKLFS: EMPLOYMENT RATE, ALL, AGED 16-59/64 SADJ UK LFS: IN EMPLOYMENT, ALL, AGED 16-59/64 VOLA UK UNEMPLOYMENT VOLA UK LFS: UNEMPLOYMENT RATE, ALL, AGED 16 & OVER SADJ UKTOTAL CLAIMANT COUNT VOLN Government finance UKBOP: EXPORTS - MANUFACTURES CURN UK EXPORTS VOLUME INDEX VOLN UK IMPORTS VOLUME INDEX VOLN UK TERMS OF TRADE NADJ UK UK UK UK UK UK INDUSTRIAL INDUSTRIAL INDUSTRIAL INDUSTRIAL PRODUCTION PRODUCTION CONFIDENCE INDICATOR - UK SADJ PRODUCTION INDEX VOLA PRODUCTION INDEX - MANUFACTURING VOLA PRODUCTION - MANUFACTURING VOLA OF TOTAL MANUFACTURED INVESTMENT GOODS VOLA OF TOTAL MANUFACTURED INTERMEDIATE GOODS VOLA

20 Lagan and Mountford (2005) use the following categories "employment; government finance; output; housing starts and vehicles; consumer and retail confidence; prices; money and loans; interest rates; composite leading indicator; and stock prices and exchange rates". See their Appendix (Lagan and Mountford (2005, p.94-97)) `1 Artis, Banerjee and Marcellino (2005) use the following categories "Real output and income, Employment and hours, Retail, manufacturing and trade sales, Housing,, Stock prices, Exchange rates, interest rates, Money aggregates, Price indices, Wages and Miscellaneous". See their Appendix (Artis, Banerjee and Marcellino (2005, p. 295-297)) 22 Kapetanios, Labhard, and Price (2006) use the following categories "Real output and income, Employment and hours, Trades, Consumption, Real inventories and inventories sales, Stock prices, Exchange rates, interest rates, Monetary and quantity credit aggregates, Price indices and surveys", see their Appendix (Kapetanios, Labhard, and Price (2006, p.34-38))

119

UK INDUSTRIAL PRODUCTION: ELECTRICITY, GAS & WATER SUPPLY VOLA' UK INDUSTRY SURVEY: EXPORT ORDER BOOK POSITION - UKSADJ UK INDUSTRY SURVEY: ORDER BOOK POSITION - UK SADJ UK INDUSTRY SURVEY: PROD. EXPECTATION FOR MTH. AHEAD - UK SADJ UK INDUSTRY SURVEY: PRODUCTION TRENDS IN RECENT MTH. - UK SADJ UK INDUSTRY SURVEY: STOCKS OF FINISHED GOODS - UK SADJ UK INDUSTRY SURVEY: SELLING PRC. EXPECT. MTH. AHEAD- UK SADJ Housing starts and vehicles UK NEW ORDERS FOR TOTAL CONSTRUCTION VOLA Consumer and retail confidence UK RETAIL SALES: ALL RETAILERS - ALL BUSINESS VOLA UK CONSUMER CONFIDENCE INDICATOR - UK SADJ UK BOP: BALANCE - TRADE IN GOODS CURA' UK BOP: BALANCE - MANUFACTURES CURN' Prices UK RPI NADJ UK RPI: ALL ITEMS EXC. MTG. INT. PMTS. (%YOY) CURN UK RPI: PERCENTAGE CHANGE OVER 12 MONTHS - ALL ITEMS NADJ UK RPI: ALL ITEMS EXCLUDING HOUSING (%YOY) UK AEI: WHOLE ECONOMY INCL. BONUS NADJ UK CPI NADJ UK CPI - HOUSING NADJ2 UK CPI - FOOD NADJ` UK MARKET PRICE INDEX - UK BRENT NADJ2 Money and loans UK OFI : BLDG. SOCIETIES MORTGAGES COMMITMENT FOR ADVANCES CURA UK BUILDING SOCIETIES MTG. COMMITMENT FOR ADVANCES CURN UK MO WIDE MONETARY BASE (END PERIOD : LEVEL CURA UK NOTES AND COIN -I MONTH CHANGE SADJ UK NOTES AND COIN -6 MONTH ANNUALISED CHANGE NADJ UK BOE: BANKING DEPARTMENT: RESERVES & OTHER A/C. S OUTSTANDING Interest rates UK MONEY MARKET RATE( FEDERAL FUNDS) UK GOVT BOND YIELD - MEDIUM TERM UK GOVT BOND YIELD - LONGTERM UK YIELD 10-YEAR CENTRAL GOVERNMENT SECURITIES NADJ UK LENDING RATE (PRIME RATE) UKLONDON INTERBANK RATE -3 MONTH (EP) UK3 - MONTH MONEY MARKET (MEAN) NADJ UK 3 MONTHS TREASURY BILLS YIELD (EP) UK 3 MONTHS TREASURY BILLS YIELD (EP) UKSTERLING ONE YEAR INTERBANK RATE NADJ UKSTERLING ONE WEEK INTERBANK RATE NADJ UKSTERLING ONE MONTH INTERBANK RATE NADJ UK ABBEY NATIONAL - MORTGAGE RATE SPREAD20 SPREAD10 SPREAD5' DEFAULT20 DEFAULTIO Composite leading indicator UK COMPOSITE LEADING INDICATOR (AMPLITUDE ADJUSTED)NADJ UK COMPOSITE LEADING INDICATOR (RATIO TO TREND) NADJ UK COMPOSITE LEADING INDICATOR (TREND RESTORED)

120

UK COMPOSITE LEADING INDICATOR: 3MTH PRIME BANK BILLS NADJ UK COMPOSITE LEADING INDICATOR: 6-MONTHS RATE CHANGE AT ANNUAL UK COMPOSITE LEADING INDICATOR: 3MTH PRIME BANK BILLS NADJ UK COMPOSITE LEADING INDICATOR: FTSE-A NON FIN SHARE PRICE INDEX UK COMPOSITE LEADING INDICATOR: NEW CAR REGISTRATIONS VOLA UK COMPOSITE LEADING INDICATOR: PRODN. - FUTURE TENDENCY SADJ UK COMPOSITE LEADING INDICATOR: PRODN. - FUTURE TENDENCY SADJ Exchange rates UK US DOLLARS TO (EP) VOLN UK YEN TO 1 (PURCHASING POWER PARITY LEVEL: 1975 BASED) UK EURO TO NATIONAL CURRENCY UNIT (AVG) UK REAL EFFECTIVE EXCHANGE RATES VOLN` UK WEEKLY EARNINGS - WHOLE ECONOMY NADJ' UK UNIT LABOUR COSTS, RELATIVE NORMALIZED SADJ`

2those used by Artis Banerejee and Marcellino (2005) 'those used by Ludvigson and Ng (2007) for USA. The rest of the variables are used by Lagan and Mountford (2005), although few of them are also downloaded from Datastream guided by Lagan and Mountford (2005) dataset, however, they have slight differences in their names compared to the names of these variables in Lagan and Mountford's (2005) dataset.

muse useu oy tapetanIos, 20 but this chapter uses it.

raonara ana rrice kwvo), annougn iney wan t use spreau ror

121

Chapter 3 Idiosyncratic Risk and Time-Varying

Betas

122

3.1

Introduction

"More generally, one gets no compensation or risk adjustment for holding idiosyncratic risk. Only systematic risk generates a risk correction" (Cochrane, 2001, p. 18).

The findings from the previous chapter that systematic risk factors are priced factors in the cross section of UK stock returns are consistent with the statement of Cochrane (2001, Ch(1)) that systematic risk should be priced and compensated for. However, no examination of idiosyncratic risk was undertaken in the previous chapter which leaves the first part of Cochrane's (2001, Ch(1)) statement unexamined in the current thesis. The need to investigate idiosyncratic risk is important. Theoretically, Merton (1987) validates the potential volatility). He questions the

importance of stock's specific variance (idiosyncratic

practicality of the assumption of perfect market and develops a model based on incomplete information. He derives the cross-sectional stock returns as a positive function of the stock's specific variance, along with other variables.

However, Ang, Hodrick, Xing and Zhang (2006) report that they find that idiosyncratic risk is related negatively to stock returns which they describe as a puzzling finding. They point out that this opposes Merton (1987) and previous findings of a positive or an insignificant relationship. Furthermore, Ang, Hodrick, Xing and Zhang (200823)report that in 23 countries (including the UK), they find similar idiosyncratic volatility effect where lagged idiosyncratic volatility is related negatively to future stock returns. Diavatopoulos et al., (2007) point out that the short sales constraint and limits to arbitrage could be behind the
23A ng, Hodrick, Xing and Zhang's (2008) is forthcoming in Journal of Financial Economics, based on correspondence with professor Hodrick

123

perplexing

findings of Ang, Hodrick,

Xing and Zhang (2006). Diavatopoulos

et al., (2007)

indicate that the latter study may have combined constrained and unconstrained stocks which
may result in idiosyncratic risk being negatively priced. They report that idiosyncratic risk

has a positive relationship with future returns on stocks when the former is measured by implied idiosyncratic volatility. Furthermore, Huang, Liu, Rhee and Zhang (2006) report that the findings of Ang, Hodrick, Xing and Zhang's (2006) and Ang, Hodrick, Xing and Zhang's (later version as 2008) of negative idiosyncratic risk effect on stock returns is explained by the short term reversals in returns of high idiosyncratic risk winner stocks and disappear once this phenomenon and size are accounted for.

On the other hand, Spiegel and Wang (2005) report that idiosyncratic risk is positively related to stock returns based on monthly returns and they point out that they have no answer to the negative relationship uncovered by Ang, Hodrick, Xing and Zhang's (later published as Ang, Hodrick, Xing and Zhang (2006)) using daily returns. Furthermore, Malkiel and Xu (2006) report that idiosyncratic volatility is able to explain positively the cross section of return on stocks. Furthermore, Fu (2007) argues that idiosyncratic risk varies over time and Ang, Hodrick, Xing and Zhang's (2006) finding is not pertinent for the expected stock
returns relationship with idiosyncratic volatility. He points out that when he models the

expected idiosyncratic risk using EGARCH model he finds the relationship between this conditional measure of idiosyncratic risk and stock returns is positive.

A study that attempts to reconcile the evidence regarding idiosyncratic risk is Bali and

Cakici (2008). Bali and Cakici (2008) refer to abovecited studiesand statethat the empirical

124

by these and other studies as to how idiosyncratic risk is related to the provided evidence cross sectional expected returns on stocks is conflicting. They indicate that this is becauseof the differences in the methodologies employed by the studies. Furthermore, they report that Ang, Hodrick, Xing and Zhang's (2006) results are caused by illiquid and small stocks, and once these are excluded, the negative effect of idiosyncratic risk on the cross section of stock returns loses its significance. Bali and Cakici (2008) conclude by stating that idiosyncratic risk role in explaining cross sectional expected returns on stocks is not robust.

An important

issue related to idiosyncratic

risk is discussed by Chen and Keown

(1981). Chen and Keown (1981) point out that when market beta is time varying, then the residual risk is contaminated by an amount that is equal to the time-varying part of the beta

times the market excess returns. They point out that this will cause the estimated residual variance from the ordinary least square method not to be pure when the beta is time-varying. Furthermore, they indicate that when they capture the time variation the residual risk has no relationship that Miller with the market beta. Malkiel in beta, they find that

and Xu (2006) point out

and Scholes (1972) argue that the errors in measuring betas could be among the idiosyncratic risk. Therefore, combining this with Bali and

reasons behind the significant

Cakici's (2008) findings and Chen and Keown's (1981) argument and findings, the evidence of the ability of idiosyncratic risk to drive stock returns could arise from the failure to capture returns. In other

the time variation in the measures of risk (i. e. betas) when modeling stock words, idiosyncratic

risk is not robust in the cross-section of stock returns as concluded by risk actually

Bali and Cakici (2008) and in those studies that find otherwise, idiosyncratic captures the time-variation of betas rather than being significant

on its own as Chen and

125

Keown (1981) state the residual risk will be not be pure in this case. In fact, Malkiel and Xu (2006) point out that the residuals from any asset pricing model could be proxy for omitted factors and for this reason the residuals are interpreted as a measure of idiosyncratic risk relative to that particular model from which they are calculated.

Inspired by this, this chapter attempts to contribute to the literature of idiosyncratic in the cross section of stock returns by attempting to provide further

risk

evidence toward risk in the cross

reaching a more conclusive conclusion regarding the effect of idiosyncratic

sectional returns on stocks which is essential in light of the current findings. Toward this end this chapter examines if measuring accounts for the time-variation idiosyncratic risk from an asset pricing model that of Bali and

in the measures of risk, support the findings risk is not significant.

Cakici (2008) and others who find that idiosyncratic employs conditional

Fletcher (2007) risk in

asset pricing models among other models to study idiosyncratic an important

the UK market. However, time-varying

point in order here, Ghysels (1998) state that the otherwise it will be outperformed for explicitly by the

beta should be modeled correctly

constant beta. Avramov

and Chordia (2006) develop a methodology

modeling

time varying beta in asset pricing models. They point out that their methodology of modeling beta explicitly provides a significant improvement for the asset pricing models including the

Fama and French's (1993) three-factor model.

Therefore, this chapter starts first by examining idiosyncratic stock returns in the UK market by applying Ang, Hodrick,

risk in the cross sectional

Xing and Zhang's (2006,2008)

and Spiegel and Wang's (2005) studies on the UK stock market. Then it applies Avramov

126

and Chordia's (2006) conditional model and methodology to model the time-varying betas to examine if this captures idiosyncratic risk i. e. it reapplies Ang, Hodrick, Xing and Zhang's (2006) and Spiegel and Wang's (2005) studies but with explicit modeling of time varying beta following Avramov and Chordia (2006). It is important to make clear that Fletcher

(2007) studies idiosyncratic risk in the UK and uses conditional CAPM and conditional consumption CAPM among other models. He points out that he models the stochastic discount factor (SDF) model's coefficients of these models as function of both constant and lagged dividend yield. In addition, Ang, Hodrick, Xing and Zhang (2006) and Spiegel and Wang (2005), among other studies in idiosyncratic risk, allow for time variation in betas. In fact, Ang, Hodrick, Xing and Zhang (2006) point out that the potentially time-varying beta is the motivation behind their use of daily return over a month. In addition, Ang, Hodrick, Xing and Zhang (2006) indicate that all risk factors and the conditioning variables on which the time-varying beta is conditional need be known in order for the conditional models to be estimated accurately. Furthermore, Ghysels (1998) state that the crucial issue is the correct modeling of the time-varying beta and Avramov and Chordia (2006) point out that their explicit modeling of time-variation in beta using firm level characteristics is what

distinguishes their study from the previous studies.

The rest of the chapter is organized as follows: Section 3.2 reviews the relevant literature. Section 3.3 states the hypotheses to be studied, Section 3.4 presents the methodology employed and the data used. Section 3.5 presents the results and the discussion 3.6 findings findings finally Section the a and presents summary table and of empirical section 3.7 concludes.

127

3.2

Literature Review

3.2.1

Idiosyncratic

Volatility

and Stocks Returns

Merton (1987) develops a models based on incomplete information. He points out this is motivated by the fact that investors' portfolios are made of a small number of securities (2006) is Xu Furthermore, Malkiel to point out that really available. and what relative idiosyncratic risk is not priced in the CAPM, however when holding the market portfolio is not a realistic investment option for investors, then idiosyncratic risk can be rationally priced. They report that idiosyncratic volatility is positively related to the cross-section of stock returns and this relationship is robust to controlling for other explanatory variables such as size and book-to-marker value. In addition, Lehmann (1990) indicates that the findings that residual risk is insignificant contradict with the findings of the mean-variance inefficiency of market portfolio. He points out that as the latter is partially attributable to omitted risk factors, then coefficients on these factors should be included in the residual risk.

Goyal and Santa-Clara (2003) point out that they study the relationship

between return

is They that the stock risk which average report on stock market and average stock variance. mainly idiosyncratic is positively related to market return in the predictive time-series that this and is

regression while finding is in

the market variance is not si gnificant. line with investor risk

They point out further risk point premia out

heterogeneity-time-varying premia models. They

models as there

background

risk-time-varying

countercyclicality fluctuations

in average risk, it could be argued that average stock risk captures the

in business cycle. Then they report that average risk effect is robust to predictive

128

find Goyal In Bali (2005) they that that and report contrast, et al variables. macroeconomic Santa-Clara's (2003) findings lack robustness and are caused by small stocks and liquidity premium. Bali, et al (2005) also point out that once average stock variance is calculated on a value weighted basis rather than on an equal weighted basis as in Goyal and Santa-Clara (2003), the predictive ability of average risk to expected market return does not survive.

Guo and Savickas (2006) argue that aggregate idiosyncratic volatility is essential for factor's it because, ICAPM's risk premium pricing stock among other reasons, measures an conditional variance. They point out that aggregate idiosyncratic risk could be a candidate for a pervasive macro factor. They report that value-weighted idiosyncratic and stock market risks are jointly significantly related to future market returns with a negative and a positive relationship, respectively. They point out that the omitted factors could lie behind why the previous studies do not find such positive association. Furthermore, Guo and Savickas (2006) point out that Goyal and Santa-Clara's (2003) finding is due to the correlation between their measure of idiosyncratic risk and market volatility. Finally, they report that idiosyncratic risk is significant in other international markets including the UK (with negative sign).

Guo and Savickas (2008) report that they find generally market return is predicted jointly by idiosyncratic (negatively) and market (positively) risks in the G7 countries. In addition they point out that they uncover for the UK, among other countries, a positive In idiosyncratic between the addition they report that risk. value premium and association idiosyncratic volatility explains returns on stocks cross sectionally similar to the book-toit investment the factor shifts and value premium opportunities also proxies and market

129

volatility in Fama and French's (1993) model. Furthermore they point out that the negative between the aggregated book-to-market value and average idiosyncratic relationship
volatility could be behind the latter negative relationship with future market returns.

Furthermore, Angelidis and Tessaromatis (2008a) indicate that there is confusing findings related to the performance of idiosyncratic risk in predicting market returns. They report that they find a negative relationship between value-weighted idiosyncratic risk only in the UK and Germany (out of 10 European countries) and future market returns. Furthermore, they report that the SMB (including for UK) and HML premiums are forecasted (positively) by equally weighted idiosyncratic volatility and value premium is also related to value weighted idiosyncratic volatility. Angelidis and Tessaromatis (2008b) study the UK. They

point out that they use (1) all stocks based, (2) large capitalization stocks based and (3) small capitalization stocks based idiosyncratic risks. They report that they found the third measure (i. e. based on small stocks) of idiosyncratic risk forecasts robustly future SMB.

Ang, Hodrick, Xing and Zhang (2006) report that there is a cross sectionally negatively priced innovations to market volatility. They argue that this is in agreement with the ICAPM. On the other hand, Ang, Hodrick, Xing and Zhang (2006) also point out that they find that idiosyncratic volatility, estimated from Fama and French's (1993) three-factor model, is significantly and negatively related to average returns. They point out that this contradicts insignificant like Merton's (1987) theory the or relationship found positive others as well as by earlier studies. Ang, Hodrick, Xing and Zhang (2006) point out that their finding of this is by factor. is aggregate volatility a puzzle risk relationship and not captured negative

130

Furthermore, they point out that the reason that their findings are different from other authors is due to not using firm's level idiosyncratic risk as measure of risk or for forming portfolios by those studies.

Ang, Hodrick, Xing and Zhang (2008) point out that this negative influence of lagged idiosyncratic risk on stock returns is global. They report that they find this negative relationship is significant in the G7 and in the rest of the 23 developed counties that they study. They state that there may be risk factors responsible for this phenomenon. Furthermore, they point out that this, what they call, idiosyncratic volatility effect, is robust in the US to many economic explanations.

On the other hand, Spiegel and Wang (2005) point out that they studied the interaction of idiosyncratic risk with liquidity in capturing the cross sectional return on stocks. They report that stock return is related positively to idiosyncratic risk while negatively to liquidity. Furthermore, they point out that they find when the two variables present together in the
relationship with stock returns, idiosyncratic risk maintains its explanatory power of stock They point risk based

returns while only one measure of liquidity out that they find idiosyncratic on OLS method.

(dollar volume) remains significant. is superior to idiosyncratic

risk based on EGARCH

In addition, Fu (2007) points out that Ang, Hodrick, Xing and Zhang's (2006) findings Fu (2007) idiosyncratic high that to stocks. risk points out of return reversal occur are result that the latter study's findings do not apply to the expected relationship becauseidiosyncratic

131

volatility

varies over time. Fu (2007) points out that the EGARCH feature. He reports that the conditional idiosyncratic

model accounts for this model

time-varying

risk from EGARCH

has a positive relationship with expected stock returns. Brockman and Schutte (2007) support this positive

In addition Fu (2007) points out that contemporaneous volatility relationship using

international data and using his method of estimating idiosyncratic model.

using EGARCH

Conversely, Liang and Wei (2006) indicate that they calculate idiosyncratic volatility based on monthly returns and find, generally in 23 developed countries, idiosyncratic risk is negatively related to stock expected returns. They point out that these findings confirm that this puzzle is robust and state that idiosyncratic risk could be seen as capturing some sort of undesirable risk. In addition, Liang and Wei (2006) point out that they find that the relationship is positive using country market portfolios and point out that this is in agreement with Merton's (1987) global version model. Furthermore, they point out that innovation to local market volatility has a robust negative price of risk in the UK in addition to Spain, and the negative relationship applies to innovations to global market volatility as well.

Boehme, Danielsen, Kumar and Sorescu (2005) point out that this mixed evidence of idiosyncratic risk results from ignoring the short sales constraints when conducting the

analysis. They point out that Merton's

(1987) model of positive idiosyncratic

risk influence

on the cross sectional returns on stocks, assumes frictionless Danielsen, Kumar and Sorescu (2005) point out that Miller

market. Furthermore, Boehme, (1977) predicts dispersion of

opinion is negatively related to stock returns given that there is binding constraint on short

132

idiosyncratic has They that risk a cross sectional positive relationship with stock report sale. returns with no constraints. They point out that this agrees with Merton (1987). Furthermore, Boehme, Danielsen, Kumar and Sorescu (2005) point out that in agreement with Miller (1977) constrained stocks' idiosyncratic risk relates negatively to stock return.

Furthermore, idiosyncratic

Diavatopoulos,

Doran and Peterson (2007) indicate that they measure volatility and study its relationship with future

risk as implied

idiosyncratic

stock returns and find a cross sectional positive relationship. implied idiosyncratic volatility. risk outperforms They point the AR

Furthermore they point out that as the EGARCH idiosyncratic risk based return

(2) as well

idiosyncratic relationship

out that the positive

is more apparent in small and high book-to- market stocks and could be related effects. Furthermore they report that short sale constraint is negatively

to these two

associated with future returns on stocks

On the other hand, Huang, Liu, Rhee and Zhang (2006) point out that the finding of Ang, Hodrick, Xing and Zhang (2006) and Ang, Hodrick, 2008) of the negative idiosyncratic Xing and Zhang (later version as

future is a result of short-term return on returns risk effect

large They is because the the the that this on winner of returns of reversal state reversal. highest idiosyncratic risk stocks. They report that once the latter reversal effect and size are relationship disappears. In addition they point out that risk is not (1,1),

controlled for, the negative significant they find the relationship

idiosyncratic expected of expected stock returns with

robust whether they forecast the latter using, among other methods, GARCH, ARIMA, or past month idiosyncratic volatility.

EGRACH

133

French, Schwert and Stambaugh (1987) report that they find the unexpected market volatility has a negative relationship with the unexpected excess market returns that results

from the positive relationship between the expected components of two measures. They point out that they calculate monthly market volatility from one month of daily returns and

decompose it into two parts of expected and unexpected via ARIMA. out that they calculate volatility using GARCH

Furthermore they point

from daily and monthly returns. In addition, relationship

French, Schwert and Stambaugh (1987) indicate that studying expected volatility with expected excess return should also include time-varying risk's measure.

Furthermore, confusing

Chua, Goh and Zhang

(2007)

point

out that the inconclusive

and

evidence concerning

the importance

of idiosyncratic

risk could be due to the

practice of other authors of employing expected idiosyncratic

realized rather than expected returns on stock with

risk when the realized returns are not good measures of their expected volatility to its two components They point out that for.

values. They point out for this reason they split idiosyncratic

expected part and unexpected part and use AR (2) for the decomposition. the positive expected relationship They report that they find that

is uncovered once unexpected returns are controlled unexpected (expected) idiosyncratic volatility

has a

contemporaneously (expected)

robust positive relationship In addition

with its return counterpart;

i. e. unexpected relationship is

stock returns.

they point out that the unexpected

consistent with Merton's (1974) option effect.

Bali and Cakici (2008) report that they find idiosyncratic risk is not robustly related to expected returns on stocks. They point out that they find a number of key players namely, the

134

frequency of the data, the stock sorting breakpoints and the portfolios returns weighting method as well as stock's size, degree of liquidity and price decide whether idiosyncratic risk has any significant cross sectional relationship with stock expected returns. They report that they find idiosyncratic risk influence on returns is not significant using monthly returns while for daily returns it is significant just in the case of using value weighted returns on portfolios (with CRSP breakpoints). Furthermore, they point out that once the smallest, most illiquid and lowest price stocks are excluded, Ang, Hodrick, Xing and Zhang's (2006) finding of the negative relationship disappears as it is driven by these types of stocks. In addition, they point out that monthly-return-based idiosyncratic volatility is better measure of expected idiosyncratic volatility than daily-return-based estimates.

Fletcher (2007) studies the UK idiosyncratic volatility. He states that he studies to what extent idiosyncratic risk is correctly priced by several asset pricing models, including among others, Fama and French's (1993) three-factor model, what he calls Petkova's (2006) application of Campbell's (1996) model, conditional CAPM and conditional consumption CAPM. Fletcher (2007) reports that he finds that idiosyncratic risk is important and consistent with Ang, Hodrick, Xing and Zhang (2006) and Ang, Hodrick, Xing and Zhang (later version as Ang, Hodrick, Xing and Zhang (2008)) among others. Furthermore, Fletcher (2007) points out that for idiosyncratic risk to be correctly priced is not an easy task for some pricing models and whether to price this latter risk or systematic risk correctly is a matter of tradeoff for these models. Furthermore, Au, Doukas and Onayev (2007) report that they find in the UK market for stocks that have high idiosyncratic risk, short-interest is negatively

135

Xing Zhang Hodrick, line Ang, is in They that this and to with point out returns. related (2006)

3.2.2

Time-Varying

Betas

Chen and Keown (1981) point out that when beta is time-varying

the estimation of

for by heteroskedastic because OLS be biased the the unaccounted will and unsystematic risk beta's variability relationship significant will show up in residuals risk. They point out that they study residual risk for time variation in the betas and find no

with market beta after accounting relationship

between the two measures of risk. They state that this insignificant

relationship is unlike what was found previously.

In addition, Ang and Chen (2007) point out that the OLS delivers heteroskedastic residuals which are not independent of the risk factor when the risk measures are timebetas be beta's the the are will when variance underestimated varying and correct conditional be in by They OLS. that should returns sock no anomalies calculated point out rolling betas CAPM's in time the are varying context until considered significant conditional modeled.

In addition, Ghysels (1998) indicate that because market beta are found empirically to be time varying, there have been calls in the literature to replace the unconditional asset it is beta by He to that their crucial model out points conditional counterparts. pricing models

136

risk correctly in order for models with time-varying beta to deliver better results than unconditional models (with constant betas), otherwise the reverse will occur.

Avramov and Chordia (2006) argue that the correct beta dynamics is impossible to identify. They point out that, nevertheless, they model time varying factor beta using individual firm's characteristics and the business-cycle variables where the former includes size and book-to-market. They point out that as a result of this modeling of time varying beta,
the conditional unconditional characteristics version of the asset pricing They report that models they improve substantially over their

counterparts.

find

the book-to-market

and size

remain cross sectionally

significant

for stock returns under the unconditional

Fama and French's (1993) three-factor time-varying

model while Fama and French's (1993) model with power. Furthermore, they

beta cause these effects to lose their explanatory

report that they find the momentum effect disappears when the model's alpha is modeled to be time-varying.

In fact, also a number idiosyncratic

of idiosyncratic

risk

studies

point

to the potential

that

risk carries other effects, Ang, Hodrick, Xing and Zhang (2008) point out that

Ang, Hodrick, Xing and Zhang (2006) explained that Fama and French's (1993) three-factor model will not correctly price the portfolios their idiosyncratic that result from sorting the stocks according to

risk, given that there are missing factors in this pricing model and hence et

the its residuals reflect their these missing risk factors' effect. Furthermore, Diavatopoulos idiosyncratic between is (2007) that there state a potential association al.,

risk positive effect

137

book-to-market and stock's size and on stock returns stock's effects. They state that idiosyncratic risk could be interpreted as a risk factor that should part of the pricing models.

In addition, Fletcher (2007) studies UK idiosyncratic risk using among other models, conditional consumption CAPM and conditional CAPM. He points out that he models the SDF model's coefficients of these models as functions of two arguments which are constant and lagged dividend yield. Despite this he states that he leaves examining fully whether the conditional factor models (including the Fama and French's (1993) model) capability to price idiosyncratic risk is superior to their unconditional counterparts for the future.

Antoniou et al (2007) study the momentum and point out that they apply Avramov and Chrodia's (2006) conditional factor models to three countries including UK. They report that they find the time varying alphas are behind the momentum profits. They indicate that this is in agreement with Avramov and Chordia's (2006) findings. Antoniou et al (2007) indicate that their findings show that momentum profits might be captured by variables related business cycle, but not by stock return's idiosyncratic component.

In light of these findings of time-varying betas - although Fletcher (2007) uses conditional pricing models in studying UK idiosyncratic risk - this chapter examines whether capturing time-variation in betas by following Avramov and Chordia's (2006) study would die idiosyncratic to risk effect out. cause

138

3.3

Hypotheses Development

This chapter examines idiosyncratic risk in the cross sectional returns of stocks in United Kingdom and then it studies whether allowing for time variation in beta explicitly following Avramov and Chordia's (2006) conditional pricing model and methodology, has an effect on the relationship between idiosyncratic risk and stock returns. In other words it examines if time variation in risk measurescause the idiosyncratic risk effect to die out.

3.3.1. Idiosyncratic Risk in UK

For time-series studies; Guo and Savickas (2006) report a significant negative relationship between the UK future returns on stock market and idiosyncratic risk. Guo and Savickas (2008) confirm this negative relationship, in addition they report the value premium in the UK is positively predicted by idiosyncratic risk. Angelidis and Tessaromatis24(2008a) also report a negative and positive effect of idiosyncratic risk on future returns on the market and SMB portfolio, respectively. Angelidis and Tessaromatis (2008b) report that it is idiosyncratic risk calculated using UK small size stocks which predicts future SMB.

For cross sectional tests; Ang, Hodrick, Xing and Zhang (2008) report, using daily frequency, that a negative predictive cross sectional idiosyncratic risk effect on stock returns exists in the UK; i. e. higher idiosyncratic risk stocks are associated with lower future returns. They mention that the smallest size (5%) stocks are left out of the sample. Liang and Wei

Z' Angelidis and Tessaromatis (2008a) report a negative average return for SMB and positive average return for HML in UK over the period starting January 1988 and ending August 2005.

139

(2006) report using monthly frequency for idiosyncratic risk for stock level a negative insignificant idiosyncratic risk predictive effect for the UK. They also report that the market volatility shocks carry a significant price of risk in the UK with negative sign. Fletcher (2007) uses several asset pricing models including conditional pricing models and he points out that he finds assigning to idiosyncratic risk its correct price, is difficult for some of these models and Au, Doukas and Onayev (2007) relate idiosyncratic risk to short selling in the UK and both the last two papers indicate that their findings are in agreement with that of Ang, Hodrick, Xing and Zhang (2006).

lt appears from the above cited studies that the cross sectional relationship between idiosyncratic risk and stock returns in the UK market is not totally conclusive, although there is evidence of a negative relationship. However, it seems it is stronger using daily frequency for returns as in Ang, Hodrick, Xing and Zhang (2008) than using monthly frequency as in Liang and Wei (2006). Spiegel and Wang (2005) point out that different return frequencies result in different findings. Also Bali and Cakici (2008) report that monthly and daily frequencies result in different findings with the latter frequency shows significant

idiosyncratic risk effect while the former show insignificant effect. Given this, the first objective of this chapter is to study idiosyncratic risk in the UK cross sectional stock returns to examine the robustness of the relationship between idiosyncratic risk and returns to different frequencies and different measures of idiosyncratic risk. This objective is in line with Bali and Cakici (2008) who point out that they examine the effect of the variant methodologies on the relationship.

140

Given the above, the first two hypotheses of the chapter are stated as follows; Hypothesis (1) idiosyncratic risk is priced by the cross-sectional variation of expected stock returns in the UK market. Hypothesis (2) if the first hypothesis holds, then the relationship between idiosyncratic risk and stock returns is robust to data frequency and methods used to calculate idiosyncratic risk.

These hypothesises will be tested on the UK market in this chapter by applying the studies of Ang, Hodrick, Xing and Zhang (2006,2008) who focus on daily frequency returns and OLS and find negative relationship (including for UK), and Spiegel and Wang (2005) who use monthly frequency returns and EGARCH and OLS and find positive relationship. This is also motivated by Bali and Cakici (2008) who study variation in idiosyncratic risk studies' approaches and reject the idea that idiosyncratic risk has a robust effect on the cross sectional return in the US.

3.3.2

Time-Varying

Beta and Idiosyncratic Risk

Different potential explanations have been suggested for why idiosyncratic risk is a significant cross sectional explanatory variable for the returns on stocks. For the negative relationship documented by Ang, Hodrick, Xing and Zhang (2006,2008), for example,

Huang, Liu, Rhee and Zhang (2006) suggest that the return reversals is responsible for this it is Wei (2006) idiosyncratic Liang argue consistent with risk effect while and negative

141

for hand, Miller (1977)25. On the positive relationship, the other analysis of short-selling Diavatopoulos, Doran and Peterson (2007) suggest value and size effects are related to idiosyncratic risk effect in stock returns while Chua, Goh and Zhang (2007) point out that Merton's (1974) analysis of equity as a type of option is consistent with the positive sign of unexpected idiosyncratic risk effect.

French, Schwert and Stambaugh (1987) point out that alternative time varying risk measures, among other things, should be included in studying the relationship between market's expected excess return and volatility. They point out that they find in unreported results ambiguous findings when they examined these issues and point out that this could be attributed to problems with the used risk measures among other things. In addition, Fletcher (2007) points out that he is planning to examine whether the conditional factor models are able to assign the correct price to idiosyncratic risk superiorly to their unconditional versions in future work, although he examined conditional CAPM and conditional consumption CAPM in studying idiosyncratic risk in the UK.

Furthermore, Ang and Chen (2007) point out that they show the time-varying beta in the OLS should be estimated directly in order to know the inconsistency between the unconditional and the correct conditional alphas. They point out further that this

inconsistency could not be solved neither by employing short subperiods to estimate the risk Chen Keown frequency. In (1981) high by data (betas) and addition, nor using with measures OLS beta, for the that time-varying produce that residual risk not accounting when out point

Z5Liang and Wei (2006) don't have Miller( 1977) in their references, although they cite Miller(] paper as their above quoted argument show.

977) in their

142

issue Chen (2007) in beta. Ang to this time the this as point and variation part of contains
well. Furthermore, Malkiel and Xu (2006) point out that Miller and Scholes (1972) indicate

that the errors in measure of betas could be cne of the reasonsfor the significant idiosyncratic risk effect.

Motivated by the above cited studies, this chapter contributes to the literature of idiosyncratic risk by examining whether the time-varying beta conditional model and methodology of Avramov and Chordia (2006) can capture idiosyncratic risk effect. Therefore, the third hypothesis of this chapter is stated as following Hypothesis (3) if the first hypothesis holds, then modeling the time-variation in betas may capture the idiosyncratic risk effect on the stock returns, in other words, idiosyncratic risk is not priced per se.

Previous idiosyncratic risk studies also allow for time variation in betas in the return equation. Some use daily OLS with month- window for estimating idiosyncratic risk, such as Ang, Hodrick, Xing and Zhang (2006,2008) among the others. Others use EGARCH and
monthly OLS, such as Spiegel and Wang (2005) who calculate idiosyncratic risk for each

period using a window

of past 60 months (for OLS) and all past data (for EGARCH) in betas in the return equations. However,

and

hence allow for time variation points out that time-varying

Ghysels (1998)

beta should be correctly specified and failing to achieve this so beta model with substantially higher pricing errors

probably may result in a time-varying

than the pricing errors that are produced by a pricing model with constant beta

143

3.4

Data and Methodology

3.4.1. Data

Datastream is the source of data. This chapter employs daily and monthly returns on all stock that are still traded on London Stock Exchange as well dead stocks which were traded December, 2005. Therefore it starts at some point in time between 1st,July, 1981 to 3151, with the 3706 non-financial stocks (see Data section ((2.4.1.1. Stock Returns in Chapter (2)) but ends up with a sample of 2020 as the further refinements are made as follows.

For the daily returns on stocks to be included in the sample, it should have market capitalization and book-to-market value. Ang, Hodrick, Xing and Zhang (2006) point out that they need the number of daily returns for the stock in the month to exceed 17 observations. Fu (2007) demands the stock have at least 15 observations of returns associated with trading volume that is not zero in the month. He points out that this is to alleviate the infrequent trading problem. Following them this chapter requires for the stock to be included in the analysis it should have in excess of 17 daily returns as Ang, Hodrick, Xing and Zhang (2006) but it requires these observations to be all non-zero returns which is in line with Fu's (2007) be for infrequent 2020 the dataset these trading and will stocks met criteria criteria problem. for this chapter and all the analyses will use them including for monthly returns described below. Despite of the fact of applying these criteria, this does not totally eliminate the potential problem that the database could contain highly illiquid and very small shares that are not frequently traded which may severely affect the results and therefore the findings limitation. in light be this of should read

144

For monthly returns, Avramov and Chordia (2006) point out that they require the stock to have (1) return for current month and for the previous 36 months (2) market capitalization and (3) the previous December book-to-market value. This chapter follows Avramov and Chordia's (2006) requirement for stock to be included in the monthly analysis of idiosyncratic risk. In addition to be used for the EGARCH analysis, Spiegel and Wang (2005) point out that the stock should have 60 months of returns. This chapter therefore follows this Spiegel and Wang's (2005) criteria for EGARCH analysis.

The Monthly UK Fama and French (1993) HML and SMB and market portfolio are obtained as in chapter (2) [see Chapter (2) section 2.4. ]. 1 Stock returns, for more details]. For daily frequency similarly, Professor Krishna Paudyal provided the chapter with the UK Fama and French's (1993) HML and SMB from July 1981 to December 2003 which are constructed by following the methods explained in Fama and French (1993,1996). For the last two years of the sample period which are 2004 and 2005, the UK HML and SMB are by is following Fama French (1993), the used method same constructed which and closely Professor Paudyal to construct the factors for the period 1981-2003. Furthermore, similarly Fama following for is the and closely sample whole period constructed market portfolio French (1993). For all the details see Chapter (2) section 2.4.1.1. Stockreturns, which details the Fama and French's (1993) methods of constructing these factors.

145

3.4.2

Methodology

3.4.2.1 Idiosyncratic Risk Measures

a) OLS with Daily Returns

Ang, Hodrick, Xing and Zhang (2006,2008) explain that they measure stock's idiosyncratic risk for every month based on the residuals from Fama and French's (1993)

three factor model estimated over the preceding month as follows, Ang, Hodrick, Xing and Zhang (2006, p.283, Equation (8)) r; =a' MKT, +, .,. SMB, + 8Hmr, HML, + , M

Equation (1) above is in Ang, Hodrick, Xing and Zhang's (2006) notations. They define the notations in the above equation as follows; the dependent variable (r, ') is the stock's return in excess of risk-free rate. They explain that they estimate the equation above with month of daily returns. They define stock's idiosyncratic risk to be the square root (standard deviation) of the variance of E,. This chapter follows Ang, Hodrick, Xing and Zhang (2006, 2008) in computing idiosyncratic risk for daily frequency.

Ang, Hodrick, Xing and Zhang (2008) use the above equation at different levels (country, region and world levels) and annualize idiosyncratic risk. However this chapter
follows Huang et al (2006) in converting idiosyncratic risk into monthly basis. Huang et al

(2006) point out that they follow

Ang, Hodrick,

Xing and Zhang (2006) in calculating

idiosyncratic risk. Then Huang et al (2006) point out that they follow French et al., (1987)

146

by its daily idiosyncratic idiosyncratic by risk multiplying risk and calculate stock's monthly the square root of that month's total trading days.

b) OLS with Monthly Returns

Spiegel and Wang (2005) point out that they calculate stock's monthly idiosyncratic risk for each month, using monthly returns over the preceding 60 months, as the square root of the Fama and French (1993) three-factor model's residuals, where they estimate the model by the OLS method over this preceding 60 months. More technically, Spiegel and Wang
(2005, p. 9, ) define idiosyncratic risk as

2., (T - k)- _I E; The above equation is in Spiegel and Wang (2005) notations. They define the above notations as E2, is the Fama and French's (1993) model residuals, T and k are the number of non-missing time series observations (at least 24 observations) used in the regression and the number of regression coefficients, respectively. This chapter follows them in calculating monthly OLS measure simplicity. it for T however for denominator idiosyncratic the uses risk, of

RCH Conditional Heteroskedasticiy Generalized Autoregressive Exponential c) -EGA

Spiegel and Wang (2005) also use the EGARCH model to calculate conditional idiosyncratic risk. They point out that EGARCH accounts for time-variation in idiosyncratic Model EGARCH (2005, 10, Equation (3)) Wang Spiegel as p. write and risk.

147

' -RJ, =a, +Q;. R, SMB, )+Q;.., HML, +,. +E -Rj, tircr(R,ti, KT", Af HAI! Inh =lu, +lnh.,,
M=1

xv, (2)

+J: '7;. v, -,,, m=1 fv, -n-Elv, -n+K -

Equation (2) above is in Spiegel and Wang's (2005) notation. [Note that it is clear that m in P
m=1

in the second term is a mistake and it should be n. ] Spiegel and Wang (2005)

define the notations in the above equations as; h,,, tu,

yr, and v, as residuals'

conditional variance, Inh; , 's unconditional mean, the next three terms are parameters to be estimated and the final term is an i. i. d (0,1) term, respectively. They indicate that they estimate the above EGARCH model every month for every stock using all previous returns to that month and 60 monthly returns are required for the stock to be in the sample. Furthermore they point out that they employ as a measure for conditional idiosyncratic risk for month 1, the conditional idiosyncratic volatility's estimate for month t-1. Ang, Hodrick, Xing and

Zhang (2008) point out in their footnote number (11) that they, in unreported findings, use EGARCH (1,1) with daily data to calculate the subsequent day's conditional volatility. Also Fu (2007) employs EGARCH and Huang, et al (2006) use EGARCH (1,1). Furthermore, Bali and Cakici (2008) also use EGARCH (1,1).

Therefore, following calculate conditional conditional

Spiegel and Wang (2005) this chapter estimates EGARCH

(1,1) to

idiosyncratic

risk . To keep consistency, note that all the terms in the

mean equation in Equation (2) have the same meanings as their counterpart terms

in Equation (1) above.

148

3.4.2.2 The Cross-Sectional

Regression

The majority of idiosyncratic risk studies employ both the cross sectional regression of Fama and MacBeth (1973) and portfolio formation strategies (Ang Hodrick Xing and Zhang, (2006) to study the effect of idisoycnrtaic risk in the cross section of stock returns.

Ang, Hodrick, Xing and Zhang's (2008) use Fama and MacBeth's (1973) cross sectional regression to study the significance of idiosyncratic risk in explaining stock returns. They point out that every month (t) they run a cross sectional regression where the left hand side of the equation is the monthly excess returns on the stocks and the right hand side includes constant, the contemporaneous betas (of market, HML and SMB factors estimated over month (t) with daily returns (i. e, Fama and French's (1993) three factor model)), lagged idiosyncratic risk (1-1) and a number of firm's characteristics as controlling variables, including the logarithm of size (market capitalization) measured at the month's beginning (i. e, at t-1), six month lagged book-to-market and prior six month returns (momentum). They point out that they calculate t statistics based on four lags Newey-West (1987) to correct for serial correlation and calculate the adjusted R2 as average of the adjusted R2 resulting from the cross sectional regressions.

This chapter follows Ang, Hodrick, Xing and Zhang's (2008) application of Fama and MacBeth's (1973) cross sectional regression and calculation of the relevant statistics. However, it uses for past returns, measures similar to those in Spiegel and Wang (2005). Spiegel and Wang (2005) point out that they use natural logarithms of past returns cumulated fourth (included) (2) (1) to third the sixth months and (3) and months, second previous over

149

seven to twelve (included) addition, for simplicity

months, however this chapter does not take natural logarithm.

In

and consistency, this chapter replaces the six month lagged book-tovalue measured at time t- 1, where log [see section

market with the month's beginning log book-to-market is also used to keep consistency with Avramov 3.4.2.4: The Time-Varying

and Chordia (2006) definition

Beta Model of Avramov and Chordia (2006)]

As mentioned earlier this chapter also employs another two measures of idiosyncratic risk following Spiegel and Wang (2005) which are idiosyncratic idiosyncratic risk based on OLS with the

previous 60 months data and EGARCH and Wang (2005)

risk based on all previous data. Spiegel (1973) cross sectional

point out that they run Fama and MacBeth's

regression using the approach of Brennan, Chordia and Subrahmanyam (1998). Spiegel and Wang (2005) point out that under Brennan, Chordia and Subrahmanyam's the risk adjusted-returns (1998) approach

is calculated in the first stage as the difference between the realized

excess returns and fitted returns where the latter is calculated using the estimated parameters, and then the risk adjusted return is used as the dependent variable in the second stage cross sectional regression and is regressed every month on a number of characteristics including

amonth others idiosyncratic risk, natual log of size, three measuresof past returns. The latter measuresof past returns are mentioned and detailed earlier in the section. Spiegel and Wang (2005) point out that they estimate the Fama and French's (1993) factor betas (estimated parameters) (i. e. Fama and French's (1993) model) in the first step using the previous 60 Subrahmanyam Chordia Brennan, Spiegel Wang (2005) that and and out months. point (1998) indicate that this approach overcomes the problem of the errors in variable.

150

This chapter follows Spiegel and Wang (2005) in using Brennan, Chordia and Subrahmanyam's (1998) risk adjusted returns as the dependent variable in the second step of Fama and MacBeth's (1973) cross sectional regression for studying the cross sectional relationship between returns and the forecasted idiosyncratic risk (EGARCH and OLS with monthly data) as Spiegel and Wang (2005) do. To keep consistency with Fama and MacBeth's (1973) cross sectional regression used by Ang, Hodrick, Xing and Zhang (2008) and applied in this chapter for their measure of idiosyncratic risk as described earlier in this section, the size and book-to-market value used in the second step regression as independent variables have the same definitions as described above.

3.4.2.3 Portfolio Formation

Portfolio formation is used by most of studies including Ang, Hodrick, Xing and Zhang (2006), Bali and Cakici (2008), Fu (2007), and Spiegel and Wang (2005) among others.

Spiegel and Wang (2005) point out that for each month, they sort stocks on their OLS
based idiosyncratic risk (from Fama and French's (1993) model) estimated from the past 60 idiosyncratic risk (where the conditional mean equation is Fama

months data or EGARCH

and French's (1993) three-factor model as mentioned in section (3.4.2.1 (C) above) estimated from all previously available data, into ten deciles portfolios and calculate the value weighted average returns on each of these portfolios over one month. Furthermore, they perform

another sorting to control for other characteristics such as size. They explain this in detail that in each month, they first sort the stock on their size into five portfolios using their current

151

size (market capitalization) and in the second step, within each of these five portfolios, another ten deciles portfolios are formed by sorting stocks according to their idiosyncratic risk forecasted for the current month. They indicate that the value weighted returns are calculated for each of these 50 portfolios and the idiosyncratic risk portfolios are averaged across the size quintile portfolios.

Furthermore, Spiegel and Wang (2005) calculate, for all sorting whether single or double, the following statistics (1) Fama and French's (1993) alpha for each portfolio, (2) the difference in returns between the average monthly returns on the highest idiosyncratic risk portfolio and the average monthly returns on the lowest idiosyncratic risk portfolio as well as (3) the difference in Fama and French's (1993) alpha of the highest idiosyncratic risk

portfolio and Fama and French's (1993) alpha of the lowest idiosyncratic risk portfolio for the single sorting and within each first step- characteristic portfolio in the double sorting as well as (4) difference in alphas for highest and lowest idiosyncratic risk portfolios that control for the characteristic. Spiegel and Wang (2005) also calculate t-statistics based on Newey West (1987) for all of the above; i.e. differences in returns as well as for portfolios' alphas and differences in alphas.

This chapter follows on monthly following

Spiegel and Wang (2005) in forming portfolios risk and EGARCH idiosyncratic

for stocks based

OLS idiosyncratic

risk which are calculated

their approach. It also follows Spiegel and Wang (2005) in forming portfolios that

control for size effect and in addition applies their double sorting procedure for controlling for book-to-market effect. Furthermore, it follows Spiegel and Wang (2005) in calculating all

152

the above statistics. However, this chapter calculates the Newey-West (1987) t-statistics only for Fama and French's (1993) alphas and difference in alphas of the idiosyncratic risk portfolios from single sort and for Fama and French's (1993) alphas and difference in alphas of the characteristics (size of book-to-market value) controlled idiosyncratic risk portfolios from double sort.

Ang, Hodrick, Xing and Zhang (2006) point out that they construct idiosyncratic portfolios by concentrating on what they call strategy of 1/0/1 which stands for formation/ waiting/ holding periods, respectively. They indicate that, for each month, they create five idiosyncratic risk portfolios by sorting the stocks on their measures of idiosyncratic risk calculated over the previous month, using daily returns and then calculate the value weighted average return on each portfolio over the subsequent month. Furthermore, they point out that they also use double sorting procedure to control for other known characteristics such as size and value effects. Ang, Hodrick, Xing and Zhang (2006) explain this in detail that for each characteristic under consideration, they sort the stocks in each month on that characteristic into five quintile portfolios and then, within each of these portfolios, stocks are sorted using their lagged idiosyncratic risk into another five portfolios. They point out that portfolios with different levels of idiosyncratic risk but similar levels of that characteristic are then generated
by averaging the second-sort portfolios (idiosyncratic risk quintiles) across the first-sort

quintile portfolios (characteristic).

In addition, Ang, Hodrick, Xing and Zhang (2006,2008) calculate, for both the single and double sorting, the following statistics (1) the spread between the monthly average

153

idiosyncratic highest the on risk portfolio and the monthly average return on the lowest return idiosyncratic risk portfolio as well as (2) Fama and French's (1993) time series alpha for each quintile portfolio and (3) the difference in Fama and French's (1993) alphas for the highest and lowest idiosyncratic risk portfolios for the single sorting and within each firststep characteristic portfolio in the double sorting in addition to (4) the difference in Fama and French's (1993) alphas for highest and lowest idiosyncratic risk portfolios that control for the characteristic. They point out that alI t statistics are calculated based on Newey West (1987).

This idiosyncratic

chapter

follows

Ang,

Hodrick,

Xing

and Zhang

(2006,2008)

in forming

risk portfolios

with and without controlling

for size and book to market values, risk using daily

when, following

them, idiosyncratic

risk is calculated as lagged idiosyncratic them in calculating

returns with OLS. In addition, Newey-West (1987)based

it follows

all the above statistics and only for alphas and

t-statistic

where the latter is calculated risk portfolios

difference in alpha for the single sort- idiosyncratic in alphas for the characteristic controlled

and alphas and difference from the double sort. instead of and

idiosyncratic

risk portfolios

However, this chapter uses ten deciles portfolios five quintile EGARCH portfolios to be consistent risk portfolios with

for idiosyncratic its calculation

risk portfolios

for the OLS monthly

idiosyncratic

following

Spiegel and Wang (2005) as described

above. Also Ang, Hodrick, and report some results.

Xing and Zhang (2008) point out that they use deciles portfolios

154

3.4.2.4 The Time-Varying

Beta Model of Avramov and Chordia (2006)

To examine the main hypothesis of this chapter that is, if time-varying risk measures follows idiosyncratic the the conditional model and this significance of chapter capture risk,
methodology of Avramov and Chordia (2006). They point out that they extend Brennan,

Chordia and Subrahmanyam's (1998) approach that usesthe risk adjusted returns.

Avramov and Chordia (2006) apply their methodology to a number of models including Fama and French's (1993) model and CAPM. They describe the time series regression of the stock's excess returns under the most general specification, and assuming CAPM as an example, as follows (Avramov and Chordia, 2006, p. 1010, Equation (8))
r1, =a, +JolIrmr +/j2Zr-Irmr +N/3Size,, -Irmr +/, 4Z, _ISize,, +fj5BMjr-Irmr -Irm,

+, "lozi-IBM,, -Irml

+U),

(3)

Equation (3) above is in Avramov and Chordia's (2006) notation. They define the notations in the equation above as follows; the left hand side dependent variable is the return on stock j in excess of the risk free rate at time t, r,,,,, Sizej, , BM1, and z, are market _, _, _,

excessreturn at time t, log of market capitalization (size), log of book-to-marketvalue and

macroeconomic variable (default spread) respectively and all the last three variables are at time 1-1,. They point out that they also use alternatives to default spread but the results remain unchanged.

Then in the second step, they use the following cross sectional regression, (Avramov and Chordia, 2006, p. 1009, Equation (4))

155

Af

R' l

co,

+ m=1

Cmi Zm/, -I

+ e,,

(4)

Equation (4) above in Avramov and Chordia's (2006) notation, they define the notations in the above equation as follows; R',, = constant (a, ) plus residuals (u, ) from equation (3) ; i. e. the stock's j risk adjusted returns, and Zmj, , cm, and M are stock _, characteristics -which includes log market capitalization (size), log book-to-market value and three measures of lagged returns; returns cumulated over the previous (1) second to third months, (2) fourth to sixth months and (3) seventh to twelve months-, stock characteristics' coefficients and their number, respectively. They point out the Fama and MacBeth (1973) averages of the time- series coefficients (c,,, 's) should be significantly not different from zero if exact pricing holds.

Furthermore, Avramov and Chordia (2006) develop a conditional (time-varying beta) version of Fama and French's (1993) three-factor model by applying their time varying beta methodology to Fama and French's (1993) three-factor model in which they replace r,,,, in Equation (3) above by a vector which includes market, SMB and HML factors. Therefore this chapter applies Avramov and Chordia's (2006) conditional version of Fama and French's
(1993) three-factor model and methodology.

Avramov and Chordia (2006) point out that they run Equation (3) using the full sample, however, this chapter runs it every month over the prior monthly 60 observations for monthly OLS idiosyncratic risk and over all the previous observations for EGARCH idiosyncratic risk in order to calculate these monthly OLS and EGARCH measures of idiosyncratic risk to be

156

following Spiegel and Wang (2005) the these with ways measures consistent are calculated idiosyncratic in 3.4.2.1 Hence described these the subsection above. resulting measures of as but EGARCH) following Spiegel Wang (2005) (OLS and monthly are calculated and risk using Avramov and Chordia's (2006) conditional Fama and French's (1993) three-factor model as the asset pricing model instead of the Fama and French's (1993) three factor model used by Spiegel and Wang (2005). Spiegel and Wang (2005) use the Fama and French's (1993) three factor model for the monthly OLS and as the conditional mean equation for the EGARCH. Also this chapter adds these measures of idiosyncratic risk into the group of firm's characteristics in Eq. (4) each time which becomes similar to Siegel and Wang (2005) application of the cross sectional regression for studying idiosyncratic risk. Furthermore this chapter follows Spiegel and Wang (2005) in forming idiosyncratic risk portfolios (single and doube sort) based on these measures of idiosyncratic risk which are, as mentioned above, also calculated following Spiegel and Wang (2005) but using Avramov and Chordia (2006) conditional Fama and French (1993) three-factor model.

3.5

Results

3.5.1

Idiosyncratic

Risk and Stock Returns

This subsection tests the first two hypotheses of this chapter which examines whether idiosyncratic risk explains the UK cross sectional stock returns and whether this relationship is robust, in both cross sectional regression and portfolio formations analyses.

157

3.5.1.1 OLS Daily - Idiosyncratic risk / Cross Sectional Regression

Table (3.1) presents the coefficients of Fama and MacBeth's (1973) cross sectional regression of monthly excess return on the stocks on (a) Fama and French's (1993) factors betas, (b) idiosyncratic risks measured as in Ang, Hodrick, Xing and Zhang (2006,2008) using past month's daily excess returns, (c) and a number of firms characteristics which are size, book-to-market value and three measuresof lagged returns.

The regression that includes only Fama and French's (1993) factors betas and idiosyncratic risk, shows that idiosyncratic risk is negative and significant in the cross section of stock returns. Even after controlling for the other effects, the second regression shows that idiosyncratic risk's negative effect remains robust although the coefficient of idiosyncratic risk and the associated t-statistics become smaller. These findings are consistent with Ang, Hodrick, Xing and Zhang's (2008) findings for UK and US. Furthermore and also similar to Ang, Hodrick, Xing and Zhang's (2008), none of the Fama and French's (1993) three risk measures is significant in both specifications and the second regression's specification shows that size, book-to-market and past returns are all significant. Ang, Hodrick, Xing and Zhang (2008) point out that the finding that stock's non risk characteristics are significant while the risk measuresare not is in line with Daniel and Titman (1997).

3.5.1.2 OLS Daily - Idiosyncratic risk / Portfolios

Table (3.2) shows the results of sorting stocks based on their past month idiosyncratic frequency, daily into (2006,2008) Zhang in Hodrick, Xing Ang, using and risk, measured as

158

ten equal portfolios. The first row shows the Fama and French (1993) three-factor model's alphas for each of these ten portfolios. It shows that there is no clear association between idiosyncratic risk and Fama and French's (1993) alphas. Furthermore, the first row shows that the difference between the alpha of the highest idiosyncratic risk portfolio and the alpha is idiosyncratic but insignificant, lowest is the the and although sign of portfolio negative negative this is inconsistent with the findings of Fama and MacBeth's (1973) cross sectional regression in Table (3.1) above which shows that idiosyncratic risk is significant in explaining the cross sectional returns. The last cell in the first row shows that the difference in the average excess return on the highest idiosyncratic risk portfolio and the average excess return on the lowest idiosyncratic portfolio is negative. Au, Doukas and Onayev (2007) point out that their study's findings for idiosyncratic risk in UK and its relation with short selling are in line with Ang, Hodrick, Xing and Zhang (2006).

The second row in Table (3.2) shows the average size while the third row shows the average book-to-market value for each of the ten portfolios. Both measures; average size and average book-to-market value are calculated as in Ang, Hodrick, Xing and Zhang (2006). Ang, Hodrick, Xing and Zhang (2006) point out that they calculate the average size and the logarithms book-to-market for the the the of stocks' average of average portfolio as value book-to-market in the that that the values of the stocks' are portfolio and average of size log book-to-market however in Note this that that uses. chapter portfolio, respectively. stocks in decreasing is from trend the average size is It there the that a second row apparent value. from the second lowest to the highest idiosyncratic risk portfolio, this is also consistent with Ang, Hodrick, Xing and Zhang (2006) for the US, although note that they use five instead of

159

ten portfolios as mentioned above in the relevant section of the methodology of this chapter. For the average book-to-market value there is no clear pattern.

Table (3.3) presents the results of the double sorting procedure to control for size and book-to-market value effects. Panel (A) presents the Fama and French' (1993) alphas for the 50 portfolios that result from sorting the stocks on their market capitalization and then on their idiosyncratic risk. The last column shows the difference between the alpha of the highest idiosyncratic risk portfolio and the alpha of the lowest idiosyncratic risk portfolio within each size portfolio. In addition the last row presents the Fama and French's (1993) alphas for the portfolios that have different idiosyncratic risk but similar size; i. e. control for
size

It is clear after controlling for the size effect that, the difference in Fama and French's (1993) alphas between the highest and lowest idiosyncratic risk portfolios is negative and significant. By examining the alphas of each of the ten size-controlled idiosyncratic risk portfolios, it emerges that this significant difference results from the positive significant return on the lowest idiosyncratic risk portfolio rather than from the low negative return on the highest idiosyncratic risk portfolio where the Fama and French's (1993) alpha of the latter is negative but insignificant while the Fama and French's (1993) alpha of the former is positive and significant. This is different from Ang, Hodrick, Xing and Zhang (2006) pattern, Hodrick, in Ang, Xing and Huang (2006) the that the et al., point out as negative relationship Zhang (2006) is mainly due to the very low return on the portfolio with highest idiosyncratic

160

risk.

Similarly, Fu (2007) points out that the stocks with high idiosyncratic risk in Ang,

Hodrick, Xing and Zhang (2006) have exceptionally low subsequentreturns.

In addition, Panel (A) shows that within size portfolios, the high-low alphas are significant only for the smallest two portfolios but insignificant for the biggest three portfolios. In fact Bali and Cakici (2008) point out that they find that the negative effect of idiosyncratic risk of Ang, Hodrick, Xing and Zhang (2006) presents amongst stocks with small size, low liquidity, and cheap prices. Accordingly the finding in Panel (A) of Table (3.3) for the UK could be seen as consistent with Bali and Cakici (2008). In addition, Diavatopoulos et al., (2007) point out that they find idiosyncratic risk effect is positive and significant within the smallest of the five size portfolio and also within the highest two of the five boot-to-market portfolios. Diavatopoulos et al., (2007) point out that they find that idiosyncratic risk positive effect may be associated with size and book-to-market value. Ang Hodrick Xing and Zhang (2008) point out that their results show that large stocks show greater idiosyncratic risk effect than small stocks, using value weighted Fama and

MacBeth's (1973) regression in 23 countries (including UK).

Furthermore, Angelidis and Tessaromatis (2008b) point out that they find that the UK idiosyncratic risk estimated from small size stocks is a robust significant forecasting variable of future SMB but not other elements of market return. They state that whether this idiosyncratic risk is a risk factor or not remain to be investigated.

161

Panel (B) shows the results of the double sorting where book-to-market value replaces size (market capitalization) in the first-step sort in Panel (A). The last row shows that the difference in Fama and French's (1993) alpha between the highest idiosyncratic risk portfolio that controls for value effect, and Fama and French's (1993) alpha for its lowest counterpart, is negative but statistically insignificant. In addition the H-L Fama and French's (1993) alphas, as shown in the last column, are insignificant within each of the book-to-market value portfolios.

In summary,

Fama and MacBeth's (1973) cross sectional regression shows that Ang, Hodrick, Xing and Zhang (2006,2008) using

idiosyncratic risk calculated following

daily returns, is related robustly negatively and significantly to the cross sectional returns on UK stocks, while the portfolio formation's findings are not robust and therefore not consistent with the cross sectional regression findings.

3.5.1.3 OLS Monthly Idiosyncratic Risk / Cross Sectional Regression

This subsection presents the results of an examination of the effect of idiosyncratic risk
in the cross sectional returns on the UK stock, where idiosyncratic risk is calculated as in

Spiegel and Wang (2005) using OLS with the past 60 months of returns. Bali and Cakici (2008) point out that they find the frequency idiosyncratic risk is an important determining of returns that is employed to calculate

factor. Furthermore, Spiegel and Wang (2005) Xing and findings

point out that Ang, Hodrick, Zhang (2006)) idiosyncratic

Xing and Zhang's (later published as Ang, Hodrick, different daily is based that generates returns risk on

162

from idiosyncratic risk estimated from monthly frequency.

In fact the results below are

consistent with these studies that daily-based and monthly-based idiosyncratic risks provide different results.

Table (3.4) presents the Fama and MacBeth's (1973) cross sectional regression with risk adjusted returns used in the second step as the dependent variable following Spiegel and Wang (2005) who use Brennan, Chordia and Subrahmanyam's (1998) approach. Reg. ] shows that when idiosyncratic risk exists alone in the second step cross sectional regression, it is insignificant and has a negative sign. However when the other control variables are included in the regression as Reg.2 shows, idiosyncratic risk becomes positively and marginally significantly (at 10%) related to stock returns.

3.5.1.4 OLS Monthly Idiosyncratic Risk / Portfolios

Table (3.5) presents the Fama and French' (1993) alphas of the 10 portfolios formed by
sorting following the stocks on idiosyncratic Spiegel and Wang (2005). risk estimated using the prior 60 monthly returns

The H-L alpha for the difference risk portfolio

between Fama and

French's (1993) alpha of the highest idiosyncratic (1993) alpha of the lowest idiosyncratic finding

and the Fama and French's This

risk portfolio

is negative but insignificant.

is consistent with the Reg. l in Table (3.4) above. Also it is consistent with Spiegel risk,

find idiosyncratic Spiegel Wang (2005). Wang (2005) they that out and point and

has US based OLS 60 the no clear relationship with returns, on with prior monthly calculated stock returns. Furthermore, Liang and Wei (2006) study 23 countries and point out that they

163

calculate idiosyncratic risk volatility

via Fama and French's (1993) three-factor model

for UK 36 They the the a negative and previous of monthly report estimated using returns. insignificant Fama and French's (1993) alpha. Another observation in Table (3.5) is that the highest idiosyncratic risk portfolio has lower average book-to-market value than the lowest idiosyncratic risk portfolio. Guo and Savickas (2008) point out that they find the value premium is positively associated with idiosyncratic risk in some of the G7 including for the UK. Furthermore, they point out that they find aggregate book-to-market ratio is negatively associated with idiosyncratic risk for US.

Table (3.6) presents the double sorting procedure that controls for size and book-tomarket value each a time in Panel (A) and Panel (B) respectively. The last row in Panel (A) shows Fama and French's (1993) alphas for idiosyncratic It is clear that for these portfolios, the highest idiosyncratic idiosyncratic risk portfolio risk portfolios that control for size.

the difference between Fama and French (1993) alpha of lowest Fama French's (1993) the of and alpha and The sign here is in contrast to the risk and stock returns in Reg. 2 in

risk portfolio

is negative and significant. between idiosyncratic

positive sign of the relationship Table (3.4). significant

Panel (A) of Table (3.6) also shows the H-L Fama and French (1993) alpha is this is somewhat similar to Panel risk ).

only for the smallest and middle size portfolios

(A) of Table (3.3 ) (See section 3.5.1.2

OLS Daily - Idiosyncratic

On the other hand the insignificant difference in Fama and French's (1993) alphas in the single sorting in Table (3.5) as well as the negative and significant difference in Fama for in Panel (A) for idiosyncratic (1993) that French's size control alphas risk portfolios and

164

in Table (3.6) are consistent with their counterparts in Table (3.2) and Panel (A) of Table (3.3) for idiosyncratic risk that is measured based on past month daily returns as in Ang, Hodrick, Xing and Zhang (2006,2008). However there is a difference, for monthly

frequency in Panel (A) of Table (3.6) the negative difference in the alphas appears to be due to the low returns on the highest idiosyncratic risk portfolio that control for size which has a negative and significant Fama and French's (1993) alpha while the lowest idiosyncratic risk portfolio that control for size has positive but insignificant Fama and French (1993) alpha, the opposite occurs for the daily frequency as mentioned above.

Panel (B) of Table (3.6) presents the results of the double sorting when book-to-market value replaces size in Panel (A), in the first step sort. The last row shows the Fama and French (1993) alpha for the book-to-market controlled idiosyncratic risk portfolios. It shows that the difference between the Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio and Fama and French's (1993) alpha of the lowest idiosyncratic risk portfolio is negative but insignificant. Again this is inconsistent with Reg.2 in Table (3.4) but consistent with Panel (B) of Table (3.3) which shows the same double sorting procedure for idiosyncratic risk estimated as in Ang, Hodrick, Xing and Zhang (2006,2008).

Panel (B) of Table (3.6) also shows that idiosyncratic significant within the highest book-to-market portfolios.

risk effect

is positive observation

and is

This particular

consistent with Diavatopoulos find the positive idiosyncratic the other hand idiosyncratic

et al., (2007). Diavatopoulos

et al., (2007) point out that they

risk could be associated with high value and small stocks. On risk effect is negative and significant within the lowest book-to-

165

is Panel Although (B) there that no significant relationship shows overall market portfolio. between idiosyncratic risk and the cross sectional returns on UK stocks.

In summary, the cross sectional regression analysis shows a positive relationship between idiosyncratic risk, calculated using OLS with monthly frequency following Spiegel and Wang (2005), and stocks returns. However, this relationship is not robust to portfolio formation analysis. In addition there is a strong association between idiosyncratic risk and small and value stocks. Generally speaking, the findings so far suggest the portfolio formation does not provide robust results for idiosyncratic risk effect in the UK whether idiosyncratic risk is calculated using daily or monthly data.

3.5.1.5 EGARCH / Cross Sectional Regression

Fu (2007) uses EGARCH and points out that in the light of the time-variation in idiosyncratic volatility, the EGARCH model produces a better estimate of idiosyncratic risk than a lagged idiosyncratic measure of risk. Therefore, Table (3.7) presents the results, following Spiegel and Wang (2005) who use Brennan et al (1998) approach of Fama and MacBeth's (1973) cross sectional regression that uses risk adjusted returns as dependent is idiosyncratic forecasted following Also Spiegel Wang (2005), the risk variables. and calculated from EGARCH model using all previous data.

Reg. I shows the that forecasted EGARCH (l, 1) idiosyncratic risk is negatively related to the cross sectional regression of the UK returns. However, Reg.2 shows the EGARCH

166

idiosyncratic risk effect is not robust for controlling for the size, value and momentum variables, as it loses its explanatory power and become positive. This finding is consistent with Huang et al., (2006) who report that they find insignificant relationship between the cross section of returns and forecasted EGARCH (1,1) idiosyncratic risk in the US. However, it is inconsistent with Spiegel and Wang (2005) and Fu (2007) who report significant positive relationship for the US, based on both cross sectional regression and portfolio formation analyses. Fu (2007) makes it clear that he chooses the EGARCH idiosyncratic risk series from a number of EGARCH with different lags of variance and shocks.

3.5.1.6 EG ARCH / Portfolios

Table (3.8) presents the results of sorting stocks into 10 portfolios on their EGARCH idiosyncratic risk as in Spiegel and Wang (2005). It shows the difference between Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio and Fama and French's (1993) alpha of the lowest idiosyncratic risk portfolio is negative and significant. This is consistent with Reg. 1 findings in Table (3.7). Furthermore, it is apparent that the negative effect of idiosyncratic risk comes from the low returns on the highest idiosyncratic risk portfolio which has a negative and significant alpha while the lowest idiosyncratic risk portfolio has a positive but insignificant alpha. However, it is important to note that moving from the lowest to the highest idiosyncratic risk portfolios, shows no relationship between Fama and French's (1993) alpha and idiosyncratic risk. Taking into consideration that, as filtered for been has database in the very small the the thesis not employed motioned earlier,

167

and illiquid shares, this may suggest that the highest idiosyncratic risk stocks may be unrepresentative as that they are, possibly very small and illiquid shares.

Table (3.9) shows the results of the double sorting where stocks are first sorted on their size (Panel A) or book-to-market (Panel B) and then on EGARCH based idiosyncratic risk. The last row in Panel (A) shows Fama and French's (1993) alpha for the 10 idiosyncratic portfolios after controlling for size. The difference between Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio that controls for size and the Fama and French's (1993) alpha of its lowest counterpart is negative and significant; i. e. controlling for size does not cause idiosyncratic risk to lose its effect on the cross section of returns. The last row shows that once the size is controlled for, there is a better relationship between idiosyncratic risk and Fama and French's (1993) alpha (apart from the lowest and third lowest idiosyncratic risk portfolios) compared with Table (3.8). The last column shows the difference between the Fama and French (1993) alpha of the highest idiosyncratic portfolio and the Fama and French's (1993) alpha of the lowest idiosyncratic risk portfolio within each size portfolio. This column shows that in the UK, idiosyncratic risk effect exists within the for is the earlier consistent with patterns reported smallest and middle size portfolios which the other measures of idiosyncratic risk and reported in other studies (See section 3.5.1.2 above).

Panel (B) of Table (3.9) shows the results of the double sorting procedure where bookto-market value replaces the size as a criterion in the first step of sorting. The last row shows that, similar to Table (3.8), there is no clear relationship between Fama and French's (1993)

168

alpha and idiosyncratic risk portfolios by moving from the lowest to the highest idiosyncratic risk portfolios. Furthermore, it shows the difference between Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio and Fama and French's (1993) alpha of the lowest idiosyncratic risk portfolio is negative but insignificant. This is consistent with the findings of Reg.2 in Table (3.7) and it seems that book-to-market value captures the effect of EGARCH idiosyncratic risk. Guo and Savickas (2008) report that they find for a number of countries including UK that average idiosyncratic volatility positively predicts value

premium. Furthermore they point out that average idiosyncratic volatility has a negative relationship with aggregate book-to-market ratio in the US. Panel (B) also shows that idiosyncratic risk effect is negative and significant within the lowest two book-to-market portfolios.

In summary EGARCH idiosyncratic risk has a significant and negative relationship with the UK cross section of returns, when only comparing the alphas of the highest and lowest idiosyncratic risk portfolios. However, this negative relationship is not robust to controlling for size and book-to-market effect. Furthermore, it seems that idiosyncratic risk effect exists mainly in the smallest to middle size stocks and within lowest book-to-market stocks.

3.5.2

Time Varying

Beta and Idiosyncratic

Risk

if This subsection teststhat third hypothesis timethis examines chapter which of varying betacan captureidiosyncraticrisk effect.

169

The findings of the chapter so far regarding the effect of idiosyncratic risk in the cross sectional of returns UK stocks are confusing. Ang, Hodrick, Xing and Zhang (2006) point out that the Fama and French (1993) model's residuals may carry the effect of aggregate volatility factor conditional on the latter factor is a priced risk factor and hence Fama and French's (1993) model misses this factor out. However, they report that aggregate volatility factor can not totally explain the effect of idiosyncratic risk that they document. On the other hand, as mentioned earlier, Chen and Keown (1981) point out that the time-variation effect of factors' betas will contaminate the residuals of the constant beta model when the latter is estimated by the OLS. This chapter examines whether the time-varying beta that is modeled explicitly following Avramov and Chordia (2006) will resolve these conflicting findings. Therefore this chapter applies the conditional Fama and French's (1993) three-factor model of Avramov and Chordia (2006) and their methodology.

The chapter's findings above show that idiosyncratic returns becomes positively significant

risk based on OLS with monthly and past returns are

when size, book-to-market

included in the regression while when idiosyncratic insignificant. Contrary to this, EGARCH

risk exists alone in the regression it is is negatively significant but

based idiosyncratic

when size, book-to-market explanatory power. Avramov

and past returns are included

in the regression,

it loses its

and Chordia (2006) point out that Fama and French (1992) that is contained in market prices and affect stock's returns is and Chordia (2006) point

point out that the information

captured by the value and size variables. Furthermore, Avramov

out that also Ball (1978) states that the variation jr. the expected returns is captured by market ratios.

170

This section applies Avramov and Chordia's (2006) conditional Fama and French's (1993) three-factor models and methodology. In this conditional model, Avramov and Chordia (2006) model the factor beta as function of the following variables; size, book-tomarket value and default spread in the time-series regression step of the Fama and MacBeth's (1973) regression. Then Avramov and Chordia (2006) regress, in the second step cross sectional regression, the risk adjusted return on firm's characteristics which includes size, book-to-market and past returns. In this chapter, for each month, the time -series regression is regressed over the past 60 months and the standard deviation of the residuals is used as measure of idiosyncratic risk which is then added to the other firm's characteristics in the second step cross sectional regression. This measure of idiosyncratic risk is that of Spiegel and Wang (2005) except that here betas in the time series regression are explicitly timevarying modeled following Avramov and Chordia's (2006) as described above and in the methodology section. Table (3.10) shows the results

Table (3.10) shows idiosyncratic risk has a positive sign and is not significant in explaining the risk-adjusted returns, whether it exists alone in the regression or with other explanatory variables. In comparison Table (3.4) shows that when the betas are not modeled explicitly but estimated over a window of 60 months as in Spiegel and Wang (2005), this it in idiosyncratic is the cross exists when similar measure of positively significant risk sectional regression along with the other explanatory variables.

Furthermore, Panel (A) in Table (3.12) shows the results of the ten portfolios formed by idiosyncratic risk. It shows that when time-varying betas are accounted for, idiosyncratic

171

risk effect is not significant

whether in the single sorting or double sorting. Whereas Table

(3.6), where betas are constant over the 60 month estimation period, shows the difference between Fama and French's (1993) alpha of the highest idiosyncratic French's (1993) alpha of the lowest idiosyncratic risk, risk and Fama and significant for

is negatively

idiosyncratic risk portfolios that control for size.

Table (3.11) shows the results of applying Avramov and Chordia's (2006) conditional Fama and French's (1993) three-factor model methodology as in Table (3.10) but where the OLS monthly idiosyncratic risk in Table (3.10) is replaced by EGARCH (1,1) idiosyncratic

risk calculated also following EGARCH

Spiegel and Wang (2005) (i. e. the time-series step becomes idiosyncratic risk, the conditional mean

model). To calculate this EGARCH

equation of Fama and French's (1993) three factor model is the conditional

Fama and

French's (1993) three-factor model of Avramov and Chordia (2006), where they model the factor beta as function of following This book-to-market defaults size, spread. and variables; Spiegel and Wang

resulting EGARCH measure of idiosyncratic risk is calculated following (2005) except that betas are explicitly time-varying here.

It is clear from Reg. l and Reg. 2 that EGARCH (1,1) idiosyncratic risk, although it has a positive coefficient it is insignificant whether it is the only variable in the second step cross sectional regression or when the size, book-to-market for. In comparison significant Table (3.7) shows EGARCH and momentum effects are controlled (1,1) idiosyncratic risk is negatively firm's

when it exists alone, although it loses power when other explanatory

characteristics are controlled for.

172

Panel (B) of Table (3.12) shows the results of sorting stocks on EGARCH

(1,1)

idiosyncratic risk into deciles portfolio where betas in the conditional mean equation are time varying. It confirms the findings in Table (3.11). Its shows the difference between Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio and the Fama and French's (1993) alpha of the lowest idiosyncratic sort or after controlling risk portfolio is insignificant whether in the single

for size and book-to-market

each a time. In contrast Table (3.8) and risk in the single sort and when

Table (3.9) show a negative and significant controlling for size.

idiosyncratic

In summary, the results show that idiosyncratic risk is not priced in the cross sectional returns of the UK stocks. Even though they show that it has some power in explaining stock returns, these findings are not robust and more importantly when the time-variation betas are taken into account, idiosyncratic risk has conclusively in factors

lost its explanatory power. risk

This chapter does not examine the time variation calculated as in Ang, Hodrick,

in beta effect on idiosyncratic

Xing and Zhang (2006,2008),

using daily returns, as this

includes obstacles such as requiring daily frequency for the book value and this will be constant within the month. However, having found that the portfolio formation's OLS daily idiosyncratic for time-variation results for

idea is the that accounting this risk potentially supports not robust, risk to lose its

in beta could have the effect of causing idiosyncratic

explanatory power completely. Bali and Cakici (2008) report that they find that idiosyncratic risk of Ang, Hodrick, Xing and Zhang (2006) based on daily returns disappears when small stocks along with stocks with low liquidity Furthermore, Bali and Cakici and low price are omitted for US stocks. correlated with

(2008) point out that size is negatively

173

idiosyncratic risk when one type of breakpoint is used. Tables (3.2) of this chapter, also shows that there is a negative relation between averageportfolios size and idiosyncratic risk.

This chapter's conclusion of idiosyncratic risk is not significant in the cross section of returns, supports Bali and Cakici's (2008) and Huang, Liu, Rhee and Zhang's (2006)
conclusions that idiosyncratic risk is not priced.

174

3.6

Findings Summary

The results are mixed summary table below

and these non-consistent

findings

are summarized

in the

Summary Table
Idiosyncratic risk Measure Sign lagged OLS-daily / regression No control Controlling for Size Significance Controlling for Book-to-Market Sign Significance

Significance Sign Daily Frequency Significant

Negative

Negative and Significant

lagged OLS -daily / Portfolio

lagged OLS monthly / regression monthly / Portfolio EGARCH (1,1) / regression EGARCH (1,1) / portfolios lagged OLS monthly / regression lagged OLS monthly / Portfolio EGARCH (1,1) / regression EGARCH (I, 1) / portfolios

Negative

Insignificant
Mon

Negative

Significant

Negative

Insignificant

Negative Negative Negative Negative

Insignificant Insignificant Significant Negative

Positive and Significant Significant Negative Insignificant

lagged OLS -

Positive and Insignificant Negative Insignificant

Significant Negative Significant Time-Varying Betas Insignificant Insignificant Insignificant Insignificant Positive Positive

Positive Negative Positive Positive

Positive and Insignificant Insignificant Positive Insignificant

Positive and Insignificant Insignificant Positive Insignificant

3.7

Conclusion

This chapter attempts to study the relationship between idiosyncratic risk and the cross sectional returns on the UK stocks. It employs idiosyncratic risk measures that are based on daily frequency with OLS following Ang Hodrick Xing and Zhang (2006,2008) and monthly

175

frequency with OLS or EGARCH

models following

Spiegel and Wang (2005). Then it

beta following if time modeled varying examines the effect of idiosyncratic risk.

Avramov and Chordia (2006) does capture

This chapter sheds more light on the behavior of idiosyncratic section of returns and more importantly, methodology of Avramov

risk in the UK cross beta model and

it applies the time-varying

and Chordia (2006) to examine whether accounting for time-

importance factor decisive in beta helps in the regarding variation reaching a more evidence of idiosyncratic risk in stock returns.

3.7.1

OLS Daily Frequency

Ang, Hodrick, Xing and Zhang's (2006,2008)

measure of idiosyncratic

risk, which is

based on the daily returns with OLS, is found in this chapter to be negatively related to the cross sectional returns on stocks in the UK, when the relationship is estimated using the Fama and MacBeth's (1973) cross sectional regression. This is consistent with Ang, Hodrick, Xing and Zhang (2008) who find a similar negative relationship portfolio Zhang for the UK. However,

formation results provide mixed and confusing findings. Ang, Hodrick, Xing and (2008) point out that the alphas of these portfolios' are investable returns.

Nevertheless the results in this chapter suggest that the alphas of portfolios formed on this in UK. for investors idiosyncratic the risk are not source of returns measure of

176

The findings from the portfolio formation also suggest that there is a strong association between idiosyncratic risk effect, based on daily returns and stock's size. It has been shown

that the difference between the Fama and French's (1993) alpha of the highest idiosyncratic risk portfolio and the Fama and French's (1993) alpha of the lowest idiosyncratic risk

portfolio is negative within each of the five size portfolios but it is monotonically in absolute magnitude from the smallest to the largest size portfolio. earlier Bali and Cakici (2008) and Diavatopoulos association between idiosyncratic

decreasing

Indeed as mentioned

et al., (2007), point out that there is and

risk and small stocks in the US. Also Angelidis

Tessaromatis (2008b) point out that they find idiosyncratic has predictive power for SMB in the UK.

risk based on small size stocks

3.7.2

OLS Monthly

Frequency

When idiosyncratic

risk is calculated based on OLS and monthly returns following

Spiegel and Wang (2005), this chapter finds a positive and significant relationship between idiosyncratic risk and the cross section of returns in the presence of size, book-to-market and momentum variables, when the relationship is estimated using Fama and MacBeth's (1973) cross sectional regression. However, the findings from portfolios formation are not consistent with the cross sectional regression findings. negative but insignificant who report a difference It has been shown that the relationship is

based on the single sorting, this in line with Liang and Wei (2006) between the Fama and French's (1993) alpha of the highest

idiosyncratic risk portfolio and Fama and French's (1993) alpha of the lowest idiosyncratic

177

is insignificant for UK. On the other hand, this chapter shows that negative and risk portfolio
that when size is controlled for, the difference becomes significant although negative.

3.7.3

EGARCH

Monthly

Frequency

When idiosyncratic risk is calculated based on the EGARCH model, following and Wang (2005), this chapter finds a negative significant relationship (1,1) idiosyncratic

Spiegel

between EGARCH is

risk and the cross section of stocks returns, when the relationship

estimated based on Fama and MacBeth's (1973) cross sectional regression. However this relationship disappears after controlling for size and book-to-market and momentum

variables. The findings from the portfolio

formation are consistent with the cross sectional ratio that captures the effect of

regression findings and suggest it may be the book-to-market

idiosyncratic risk. The overall findings of EGARCH idiosyncratic risk effect in this chapter are in contrast with both Spiegel and Wang (2005) and Fu (2007) who report a positive relationship for the US.

3.7.4

Time-Varying

Betas and Idiosyncratic

Risk

As discussed above the findings of whether idiosyncratic risk is priced in the UK are mixed, question the potential usefulness of idiosyncratic risk in explaining the cross section of returns.

Then, factor betas are modeled explicitly to be time-varying following Avramov and Chordia (2006). This is done by applying the time-varying beta Fama and French's (1993)

178

three factor model of Avramov

and Chordia (2006) and then the residuals from this time risks

varying beta model are used to calculate the monthly OLS and EGARCH idiosyncratic following

Spiegel and Wang's (2005). The results show that, after accounting for timerisk is insignificant in the cross section of returns, based on

variation in beta, idiosyncratic

both Fama and MacBeth's (1973) cross sectional regression and portfolio formation analysis.

3.7.5

Concluding Remark

Although idiosyncratic

risk appears to have some potential explanatory power for the power completely disappears when time-variation in

UK stock returns, such explanatory

betas is accounted for. Therefore, it could be concluded that idiosyncratic risk is not priced in the UK market. The results suggest that the initial confusing findings may be explained by idiosyncratic risk having captured the effect of not correctly modeling time-varying risk is not significant risk

measures. The finding that idiosyncratic (2008) findings for the USA.

supports Bali and Cakici's

Although, this chapter does not examine the effect of time-varying

beta on idiosyncratic

risk effect, when the latter is calculated using daily return as in Ang Hodrick Xing and Zhang (2006,2008), the findings in this chapter show the negative relationship between this

measure of idiosyncratic risk and stock returns is significant using cross sectional regression but not portfolio formation. This questions the real usefulness of daily OLS idiosyncratic risk in stock returns.

179

O Q O

NN

U_

vi

>
ob b qm Uy OM4 7

it

'fl

NEm "y 'G y

CO OA

>

'4D

-m r-

eu 0 hNG

EC N

c
77 c O .,7

^3 avi
./

u" N
0 .o

ou
jM Lit O 'Z U G, N-7

wyN2 C/) Oy

>
.V y 'O N CC fd N

cu cd N
d e oM

cd c0 y>N vUi aLi Cy

CM c0 . "" o .r

.tN

,nNc O

Oy Y '
c

iF , + fv^j N OO^

CS Ny

OM
rN v

Zk

-5

1u..

N .NN cC MO

0 00

U
000

ce

"^

r C
n* I n N

m CMte, r= oh0 r- l) 3y
r wv
ONNyO ^vW 0-d il ld tL0 >

j,
O

NO

U.
>

'
.

0--

EOCb Ny OOy 00 O O N O I

O v'i

OO

Un. LUN

'D

u xCH
OMM pOO "0 yC0 e3 id -0 OU yyQ

72
vi

. Cy,

Ep

UWy4.

V cc

OnU n OO u

v
OO

00

Cl.

^-y

'p

.D .0

NNCOL
Uc QG

'n

00

(O

LL

V)

2 C

,^

N OG r2

c> n

CL27

OZZ 'fl Ov

>, -o t, v-w -e -Z C, 2w CO
C=VC

Qx

M "Z

r
O U

rn O "N EO Y O

Eo bA

0UU

c p

s wl o0 00 Q

E0 m RS VN7
NN 2l

r_ M
yy

3NsNO
tQ

td 'n

O', OU

NN

O E'cs

v. pv

OOO U ce w U 4,

o
OOO O0O

tcO

UCo.

Q) .90
b Cs
NO
M

>, 0M.
"
Os VJ

U
bq
NN cC 1-. N

Vl

P7

00

._> 0 O OU

0"

"' -7" 00

"C .. ` "r.+

NO

G.
UNM UU

Ow

y=
r O

tC O
+-+ N .1

.oE
N O O

GC , N co
. N

Q1 U 0I^ " t. l hN

N OO

ci

-o

Ov

tr

308

Cy QpO

,01
c 0 b 0
0 w 0 a

** **
N-

In v)
o\
NO NN

00 NO

t0c

d "

OC

to O

a
p v 0W
p0 C-

a
wCcv 'O .2?, s_ ei
C

v
> O+
rn

00

-W "

syyw,.,

-7 w -N119
0 d

pp

o C.

l Io

E co ACQ
-

ci. "v

cy

w
N

a3
-(V

v> i> r--

00 C` c a)

vi MM O Z,

Vl

Uy

c i
r

E__ o ". O ' a 3p , . 0pOv

OO w ps

iF

* iF

0N

o1 0ns
ci

CD. r, mgz
C cd

iLw a3

b)

>
y
y

MO
N

0 rM lu 3 0

.iOOC U
00 O NO M " Nv u>'i '

00

a 0 N
M
t M

vi E.

'9

v a Q u, u.

c 15 C- EE m oo 5 " Na sp w w N Y'N s
05

-m

CO

ca

ig

u e-2

E:

i-

tr

Vl

Vl V-

00

00

ir*-

iINN

V
'.

* -1 NOOOOO0

-.

LID

-S ci

00

RM --

O 0O0

hM

OOO

ri

00 C .:

7pN OD OOOO

00

Ov OOO

^ 'a +N 4. 00 00 t OOO 00 M (V O r R^ . -. W) OON l kn

'M' hNNM

\0

_
0 0^

et

N . -.

vMl. ppOOOOO

** 0

* vii

t+?

0000

_^

N 00

.
c

W-. i O^

O v' N O

O v1

ON 10 to mMNOM

II

C'

i 'IT -

OOOO

O^

OOO

0 a 0
y

**M

IT

v' V') O

N M --"

i rv1 *N l 00 O N OOO0O .... MvvMN

iF

00

O O

MN* V1 Cl) O O00

a
M

N^7* 0^ O4 `'

^-

O^ MvNvMM

:>

00 O[.

.*M O

NN Ca 00 OO

v1

* *l. O I-

. -* ?^ONC

00

[10

rr

*N. O0 ri 7,

w 1

w*w" *NO 0 -+ IN NOc; N.. . 00

* 00 as 00 O O 'D MMN Q O^ OOO --NM r4

.-. ** h^G

ei M
. 92

U o
ENM bb .aOON

cd

4+

CZ2N

' 10

crj

:vP.

ONMNN OOOpOv

Npp

W)

NM 000 0

"L QU3w

cv

"

ai

ag
VTcr

z
cvx cc

on

MNN oo p I OO

NNp000 .-..

OO i MwNYTUL

00 M \,O rM O MOe o^p M ' OO l-NN ke) -O, O, OOO O OO

rn v o

CU sE - vi ' c cri

:D fl-

"o- N

a 4) 4) o Q3v ED Ma oy o _o
ow
c oo c 0, w r_

aNiC.

00

le:

1-C NZ

Zp

CY,

;ZO *

vo coo

TDs v -o 7B :;

CL

so0,
O OOpp V1 O O CO) M p. t 00 ' p N
t L (Z -0

_Mvi=
t. `n

Z tu

ON

w O

r-

aye

.OO0

"y cULc (U A *Z o N -o

won 6 `s'
o Cl

a, v) Mp-ON

co pp o 00

00 NMM 0--

Ci

3
U

M.
" vA

acsN

0 -v
M

en

(y

UN

I) mYCEgOr

0 c
OOOO,

.
O

*
NM

*
M

o*ZN NN V1

0 EEdOO
ti L p C L O` t') I `n .LOp M2L -rUo ", Np .

21
CU) G'

0
Co M^ O

s COyyM N MDGNN O, MNMMOO OOvOOOM c0 O

0"

**
NN w. i U N7_ `n VO *TC ynU

. .+w 'C rea

t0 fA UUy rn

[ N O

N OO M OOOvOzz

/1

NM 10? Cr;

* * * _*
-v
+i ri

iC y0c -0 L, -ov, NA
7, A CyC

s' .

iF

wowOyC c
N

00 O

MN O

o0 -- O

--

. ,_,.,

vl ON

6F 0 0 o

p,

eC

ai Osww 3"O 52

", 'o

C U

O L>

>EQ.

vn y

ac* *
Z

3-= 0

eo-- ONM OO66OO

o* VI

v N

o* 00 OO pop 1O

4^

Ny

r"

0.

"L

C
c

' v

g tL"' E A ", tCC vOi cd Lyy tCV

c0

oy Im

.y 2.

2 'b .. U
NMd

>
N N'y

OO

o 20`2.

c"o

E2 ooC, - 0g-.

CC i C 4. cC

oJ,

Q
VN

4L

n CD -

m'p 'O U Op 31 C5UZ,

O\ * -0

3C

-0

U0 $=E CoE =L
U L=

iF

**yU My

co O 'gin _ Z `n CO . ^Z CO

'n ZO CyUc. OOO UOC NU

I> -r-

:p
Cl.
fA ice. i. %

t;._ -- c . + U bD
DO rad _N u

cd

>'n
IM U

*^

OL

YO

CL

O .L.
Uy

u C` O Gam)6U) 'a .

`n

s.

b x o o G
C O 'w c0 N "N999 Cy

as

U-"ca a NEy3N

" ti Cn

bA v

N ^' `I

td+ MGNCN 7-

as+

00

vEEQ O c0 C0>. Oy ODUYN .D tad "e >, vU D

w N

a U
U
0
U =

E COo=>

N_

I-

? y
'O

M-

NUN

e 09 qqO-

vii

u E-0

O_

LNtMT

4.' =LC3
N7--. y 0 N L

Q Q a c

OuNNMM u'C

n auy V CO C_

"a

>

y c NL-3 NYY
y) i 0)) R+

22mE
01
C

Oy

td UU j"N"V ca NN

.5 v

,O

COnbp

YMW.

uvu.
c'Vcv cd Q. "`

>, -o 15 ` -w
-= mi as

E Qx
MO :2NN OO

s"

Oo Ey

w-s

xQ v Z x

vN3

EcUr

eO Vl

MN OOO

00

`' c sN TNTby oE.. Z 9v . :. = c ei 3-

ra

i::

OV+.

.00 N. + 0
r_ 0

NN 00 OOO

N M

V 'b yTuM O 'fl

0_ -p + O, yN= "' y y OxU

eOd

,n O>m

'in

te
ONNN OO

r-,

pyC t' iyC L1N ZD

fx

't 7

c>C

(I

EMC

.-" >
of m^

N.

MM M -7

00 N

p, kn

,bU

n c oo c'

ri

V 3,

NO

OO-

C)

L)

:i

sw

V-

-E

caE
-'Y
CVl

I 00

**
M

C
L

of c 0
M

xo
o '
sy r-

vp0

U..

r, E
0u

",

to

c0

00

00

a x

2 0

Op OO

oEu E
W

'0
w

Co (A

_a

FstLCN4.

EA >! s C, 0 n' *--z

a
M

00

pm 0 -

Q4' o
ly

L V C O . w S C O
3_j

O UV

03
wy

0c
OOO fn M O O

ti sa
U ,
N

z eC

ti-

tqMM QQQ M M N

4. ' O O C O ZZ -U

5m
y

NE r

,U OUCi

cu

9y'O a ti ovc e E
oNa

aao

N Ca

M Q

"Y y

N '""'

wwNC

n
H
i0 M

'y rCo ,N .o
off. .2E.:.

CGs" [-

O". > '0

Z, -

Co

C. v

y",

* cC 'n a, 00

* 'c MNM 0000 00 00 -M OOOO c

O*.. 00 N
N

O\ MN --O

*'

'f 00 O

O\ 00 O

p ppOO

pip . -.

00 ^

(V

i v1 D Nh

F7

00

1,00 NN VOOOON

O0

''

ppOppOO

NN ,MNNN CA a, 00

00 x
,

O7OOOvOOOO

999

*
N 0OO0Q0

-OZ: Vl M,

OOO,

9O

00

O oo

"-;

O0OOOOO

0 4
0

110 00 In
oho ONNNM

7-

'7

n V1

MNtOM

--
/1 O M Ov V'?

--

OO0O

1)
a
M M rf M '.O NN*** '""' nNNM

*f
r

00

M N

OV

MM OOO

00 19

M^ M ,__,

N ON

0O -O^

i. n N^ OMN

a N
N

it M? O9 V CD O vNN 0^ O O

x ^ O OOO7 vii N

3 - 0

L'

(U

2NM

NU

Tt

Gn

12 v
12.
'

c = vO

.N

dMviNM 12,9 -99

Vii

ooo0

"QU NO

14o
rn iC Q"

be E
2U ya W_ iO 4"" '- U

2- -0

V.

'O NCv

GOA U cLd

O n

'M'

ONO M oo

o RO O

'

+'

D rn

OOOOI.

OOO
- "0 eO NN L1' a M

V LNO

1,0. CNj aN+ tl

r,

Cs

`nM O

^
ONOO

O-7

MNN

NM O p

00 r-ONO

000

X0 o ooac "__

G.

-0

7, ,,,

N M

00

i -" 00 '" ;M

\o

[-

OOOvO,

MMN

.-*

or ti

NO

; e c -' w'

__, O

wp o a-

ed i

t) 0e 4

r*

v:.

oo

o, * -

Ncs

mD2
wy >0v cd

c o10 0 -1

om

v1

oZ `O

CO O

NOOOOOOe -d

N v1 OOO

Q. 7

c-n> w-vcm oo> 'o py 0) u

a0+
ci

dUONU vw 4. 0

tu .OU
.-

'+
92-

NO-NONNOM O, vOOO66

N
O ^;
OOO O. y Oo

co a
rn1 ' -v

v -o

r00

'C

F"

O
O OO0000OO

,1
V')

f`4

\O

nn

e -rN

E a N_
to c ,

o2 O_ U

0 vi `
0 iF 'L7 O
N 'y 0.

um
C n

v=0Q
b e 'L

-. y -

a
M

NMb O OO NN -

M -+ cn O O --O

Vl N NO

00 o0 vi O

it
vl MN

w
N

"

3'
NNswo

E a aw
4. O

Co =1:

r om i! 1 lNNMN O Ov e Z e r1 OO

le:

N MN O

0ON C M Wy

w O Nb at N O ,..; M OCO v O^ N p pVd "v `

3 .a 0

,y vi

CN U> >C +'L-+

Op...

o N o
U'N

'

oV 0

fl
NM taO =Y

N>, c .So3

x
"" UG

8" -.

2 =50u

Lj

OE

all 30e-U OO
OE -dom CC C

MyD7
` -,

Q
v rv

=NZ. 7E cd UO

oh o CO

"

00

b0

Vl

7y UL CtO ONO0C Q

I'll .> $E
U

*U * OMv'N=M C-

(A

b)

-= "N e-y

U ed .

Ng CNUCON .Z

C0 -e

G2) eO O

NU
NyE

a E "yEZVl

' bD 7

N 25 vl rt

U^

, "w
pq "p

v3
OU g3

Ou,

7-

OO)

L', O c, mu 3 ') a) v0f

o ;, x o v o 00

* cj ` or Vo E a"i CC) 00 . .G0Np, ^ ^N

OC rOU CCO 00 .`N CO T "'' it W7

*iY-4

G)

0
N " , [

UOC CUN 22 NCU C0+

.E

E ; %EU". + LDnU + oE'. . 0 ^ E aCi + u N

O
U 9. to

Oyn

U CH O

NN-N ONO

N ONNa

M*_N

>, G ^ a ,
e0

D
u

* dO

^n

M `t,, C U

UU t. N rn M n *- . MO, (l Vl r

*tOv E--tO ^ OA C-

to O

r-

7D

cu--*

CL Mm
N3

r,

NyNy

+'

Nbn

U7"

ce

p4

s2yy
lE.

(U "g

vl :.

U,

E- ri

Ll it

(1)
cZ

AA-

l' UE A F2 n2 Q
iWMp NN NN UyA UnOyM

CO CO bA

pr

CO

it oho C t. UA n' ? .iC C O= N' `U d L1.

NN

"--O

c iA. OUQ

0 o

00 ny OO

ce 4' '^
c

VUi OO

'

s ai 'Cn=v
3ca Uru5Uv

ON

^
pp O

ri
^oUAcc

un-Yc
,i, ' '40 E O> L '
M , Nd

OOO

MNN

00 00 .-OC

1.0 IN

00 O

ri. W SL n0

"--"
c0

O
M 1 M

0o

ro

E w

OOO

UE9 cc ^v v-o_
",

!c
L

"G

on_o V1

0
w

Q)
O NN OOO

.x0ENA* U'C m OOO ._.

cu c_

uy

0 a

cv

Ev

svAUA

E-Z E y= o oEv? 3c

e1 '. .n4 0 '

sOC

u
L

"'

'IT

R Z ;7 ON N
OO

r cv

occ
m O ^ NsM o
N O N>.

V o ,

u 0 b

>E '+p od. MM O L .C _0. QA t .UN in aLi r_ vi

OOO

NONx_,

a d w M H

x U

M0

73
v 3 0 ONN 0O O+ NO MU 'Y a

0 uz,
CA E3Lrp3O. O N `0O A p

4) C
M cd

NaEMM-

v Qw w

_L
A c OA cV 3L

ug"zu
MG= LTyA

V>-

_ : wLU.

.=

Iti

<C

MN O '. O D

vl .OMO

Mn O

00 M

MM ON V)

IN

ON o wl

vi

00

O,

iO

o0 N

lP

7NSN

10

^O

10

O,

Ir

e"

MN
1 vu

ONOON

N o OpOO

"o

^^^^^^

pNON rtt 00 O

W) 00 N O
.

;.

00

O
OOO,,,,,

"0Om
. op

N
MONM 0 OOOO

vi

^_OO
O --O

.. ^ 00 110
OO

"

M OOO0O

00

OM-

r ^-

.DNs 0000O

aq O-_ 0OO O6 ,.

a_ 0
O _ "0

\D

MM

"o

ON m O

M-

r-

00 kn

NN

O\

00

m
.--.

(21 00

r-

10

000

OOOOOOO

0
a 0r iO N

O Ol
10 ^

z 0 w 0

N O

OOOO ,,

+F vl ^ MMN C7, OO N

r"I
OOq-;

r^ oO

'-It
Op

N
O

q 1) N N
.. i -

O
M

00 ^
M

V1 O'. W)
en

OOO.

Co

00
N^ 00

c
Oh MMN v) N '-+ 'NNN 00 Vl 0&1 N vS Vl --N N MN

OM

r)

"--O `-'

OO

--O

OO

x
^ 7N1 'O '0

a Q
w o M

tf)

NNO, 6i

OO

%.

v^i

M^ONh

W) M ONO-:

G) Nme U r-

Nt

_
O

(nw

0) 7/

.a

OO
-U

'x 00 ? i 00 ? *N N n v1 O N 00 --O O^ lip

NUUU CU > p

O ^OOO6

co n n M vl N
97 ii a. i v

N_ L R" "o o a0

=Q-0 OC Uy O_ 'O " ' C ar O "e

V
on
00 ^^., MN

N"-

^N

fn 'D

--*

Ov

000 N

NO I'D 10 OOO

/ 'U L,
0 O"L au

fn

g-Op uL
_N .0 y00

V1

YAmC

.9NM ct7

* NO .7eNe O MO

*i '^ D C 1

'9 "N

O N OOO OOO

CD. 2m`L -0 'm

U"OU

""

..

tNM

'ct

'O

00

.O0 00 O O OO

L N

E . M4c `-'
0. 0O

3
Op

p nom, a

C bA

p-

O06ON

N c ' L .OCOO OUy"20.

uU
O U
't N 00 CD -OMMNOM OOOO MNO 00 OOO V Ul V) *i . %4

"C .-U
u C9 mbU

_N
C, ,N

ONN NN C "--

.Z, co

CZ u,
oo Yc " =o 3.
4-.

"OEE>,, C 0 >n o0
01) Ln

10
+ C O

N OOON t O4 N kn i

C O TJ

OOOONO

O +F r, (I-

MO M

C-4 ,. 'r
n-N

= a " on 3 oEE a o on c0 e My4 a`

v Ly>
OO az V) M C-Uc* .O ar -U .-0U

>,

F 42

01

00

00

N Or N

^ a y k~ v 2-0 Je s 'v L , sE.. v v< N - 2 3 jo

`Cua.

Vn00M^N M^N ^-D

OOO =1 OMO vM

C*'. l
vi

o^
O-

ry
d -00 0C M

EEZ, 'E3 LL EC 0 C 0 y f0
n"

0'

6> a2i

O OOOOI.

. --+

0Nn-

..
O7O NvM

M 00 000 ^^, l-

'A ri

"NN2 y

>, 4o id

.2

'

oU9O>
c
v't ONOO 0 Oc 00 0, OHO ON O' N rn OOOM ( 00

ci

Z;;,

NV

ca_vE., 0
rn 2 C)

c
ci .1L

N' ` 'O

xa o
. co `+^ E 'y ,.

i^3 .a 00 0

N NM Os

vi

ty gUm 0 "p

++

yd NN3 U
iH +'a 0 tom/ z

O OC 4+ vi .N

6)

is
->Va

-cO

y N r 7

ON6> O

0 13. -1

.2.
U

M Ec

_a

s c 0 a O

N 'LSD

OO OO

N `/ sO

15

u
M

Z
-N

vi

LL
L C

.y

N ,L

(D Iy A

3
O

-a
r. a

.9t-
U U.

NN

.-

ci

C O

Y Vp

=s

C
GO ON

_ c

,
N

oM,

u >
000 VQ CN ON
y) ii v O

pq "

O
u 'O

r. U G: r
'O t'-ON 7
Uv

O
'O G O U N N" y7 rn '

_w U

c`0

V) O O

7'_'.

N ""'

L -

1'

rI O O
v i:

N Z
cC

r-a

CC

.,

b0

Z-> NO

ca

>
C
Q'

c0
'C

ca
.y
2

N cl

0'C o0
M

u+

C O M Ilt
N U

O NN

L O

O y r

O O 'O t c.

E'
y

'C 'i. 'C

im

U ar

p >m*

cC

E.
T pOp l0 v1 N Q 2

CQ

Vv C. ii O O

m M bC

`n OM O uO

0
c C

't

= N , S

V.. .

C1 Y

Co Co

H l0 ya.:

w 2- vo n

a Q U 11
.n V 41 C

O O M OOOOOO

I_ j

E 00

`UH'

>
A

IU c2.
^ AN

MC cn
2 F1 ce.

cd

(U

Co
0

c F" >
O '5

E i

'

NN

r;

u '3
N
0 = y

5-c

x U
a d w
r

ca

0u
n' T

a N i 7 NM

E' R" ^
C

c v
c '

3 cZ
'in LA

3
a 0
- o2

"l

""

3
O

ci
yU

VW C

11..

all,

u >o E3
ra Uv

oo
O "

cl C "
UU 5. Vi + y 2 Up ,l N C

Z4 .S

y U+

A u c as L

-" . 10

JM 'G ^

r)

c
ca

p CD=
. tom.
:m

JD

.2

G4

wN

ca
ed W)
v eV

AO eA
cl a

6xx oc
C O

M pO c

E
O "
.'

p 'O U-

EE 0
-r C G -0

> al -'4
-p
U0 al 0 y U `d

Ds*

F
0 N
ny

Y^ G. . N i

[-
OO

4p

0O E i--

Cg. U

re G

O U
Y

' N 000 . owN E

V7
V1 N

8. " cw

C14 vv

r-

00

00

NOO 'U
O b0

yi

re

L O

N 0

U b C
O

O ._

v O .a 3 C c0
YU In -

0 U G

O^

'n c^i v
Cy

v
NN

>a
c0 y

W
bA

'A U
cd

OO0OO

rn -Se yu

"U

s
cH
y LC

"a

F- C i

c O

Q= w
MC _ O CC 0

O E

C O

vi CC v3

g o"

do

EN

w N O99 Ov0 NNF 0OOOOO eO O rD v, M f

.Dyt.

O>
,NOy to

-W

U> '. """ N_ 1-

-p .. C Im

OppO

L u

cu v

cd T

,0 bU "

Co Z UU ECCy` w

''v

C w

Nt(OO

NMO -

NNE p" =.

NNNNh 00 00 N O

.. N
--'

r1 0

v N

0^

o, pp3 0-0

'

F,

UO 'a+ C m cC

'_' Z
v O 9OO OO

c=av

'

Nw

_. _

N* ,ei: q^ O N NN

*-, ". ^ ZN o llt

Ir) V

RC oo
N

N,

37
ON I

apN
'C vi >U "' p

w'i `g

E
00 M' O 00 MO t ' N 00 :NNNC o v l Y. C

a
co

ONCMN N OONN

M^ M

rc

`c C E s_
S

OO0O

Wp

NW

'L7

vUyO *** NN T ! 1 ^^ N O ONsMM OOOQ

M ' *N '". ppp

'E
o0 00 i O 00 r e -0 OMM OOO O

N"_N mC
y

0
CC
U

.a O c 0 a
p

,0 r-q

g2
O

MO I 00 NN

QE

'i, bA S _, o uca Lu NmNQ

C' 2

LL O
, OO\ \ NOMM O-^ I O [N ^ MV MO-M V1 M 2 MMNN

.a

rr
OA

R
o, bi 0 0 t 0 in.
0_

-.: =-

=, r

1.9
O9p
MN O

--O

o>>uv.
. Ze
M MO O N

MNm^ pOOOO...

n ONO

-7 O

. -..

C
N

6= po
U . ee
yy

As
w

CC

C O

r- a -v
Wd

o.
H O

*-1,0
9 i

c\

*
NN

\O

O N

00

t-

vl

! 1

[-,

MN

"V

Co

OOO
N

OOO

e
>

O O"Qwt O 0.
M O ^^ NNOOO [ M 00 'IT 00 Cl OO Ou L"' iF Z 37

y O

0 0.
a. i U

V1

tM OON

-NN

- w" o=Y .-0=UQ c -o U-0 Co c eu

^ OOO

n OMO 4OOO
vgl

Q. y
'2U . Eyy tN

CC

. --. 00000

VNM 0=

ei,

. -C)

OO

.. . + y

pes.,

Cd

ti

vl

vl

Vl

00

3 .a O
Ny ^; M O .r

Ir

. - NMNMM

O, OM

-.

M 1q,

00 O

__ O Vi
O "O ON

OOOOOO

U vs

VJ c0 N 'G aFo

cc cu
rp 42 p 1py 4G

an
Ov

=>

,j

UUU

u112

00Co" 2.0 yo

"o

s
L fd

O c02

t O a
C

C C

0 NV

iF

t_ bp

l ON

fd fCd n cc G fO

3 ZO T

-z
CV

tu 0

U TL

Oy -0 03 '

6i y
r

a+ ., r. +

O a as

Chapter 4 Downside Risk and Business Cycle

196

4.1

Introduction

"The attraction of the CAPM is that it offers powerful and intuitively pleasing predictions about how to measure risk and the relation between expected return and risk. Unfortunately, the empirical record of the model is poor-poor enough to invalidate the way it is used in applications. The CAPM's empirical problems may reflect theoretical failings, the result of many simplifying assumptions. But they may also be caused by difficulties in implementing valid tests of the model." (Fama and French (2004, first page (p. 25))).

Sharpe (1964) states that, in developing his capital asset pricing model, investors are assumed to be risk-averse and the investor's utility function is based on the expected return and its standard deviation. Levy and Markowitz (1979) point out that several researchers stressed that either the normality assumption or the quadratic utility function assumption should hold in order for the mean-variance rule to deliver the right solution. Furthermore, Sarnat (1974) indicates that the problems of the quadratic utility function that are discussed by others are the increasing absolute risk aversion and ultimate satiation. He points out but the mean-variance criterion is still valid as long as returns are normally distributed.

On the other hand, Post and Vliet (2006) point out that stock returns are not normal but kurtoic and positively skewed and Pedersenand Hwang (2007) report that normality Post Vliet (2006) in UK. for is high frequency the and point returns rejected assumption favour in is in light the that of the variance, a criterion that that of evidence not out

197

lower higher the and considers moments of the distribution of returns should substitute the mean-variance rule. In fact, Markowitz (1959) points out that the semi-variance, which focuses on losses, is better than the variance, but when the returns distributions are normal both have similar results. Bawa and Lindenberg (1977) develop a mean-lower partial moment capital asset pricing model that uses the mean-lower partial moment beta as a measure of risk.

More recently, Ang, Chen and Xing (2006) propose a model which assumes investors with a disappointment utility function. They reported that they find around 6% annual cross-sectional downside risk premium in the US stock market. They point out that the reason that previous studies fail to find strong support for downside risk is that these studies did not examine all individual stocks. Furthermore Ang, Chen and Xing (2006) point out that their methodology has a statistical power advantage to capture time variation in beta. Despite of all of this, Post and Vliet (2005) criticize Ang, Chen and Xing's (later published as Ang Chen and Xing (200626)) methodology. More specifically, Post and Vliet (2005) point out that the latter study does not perform a conditional downside risk's tests. Indeed, although Ang, Chen and Xing (2006) allow for beta to be time-varying but do not allow for time-variation in downside risk premium.

Fama and French (1989) reported that they find a negative relation between expected risk premium and economic (business cycle) conditions. They argue that this is
`' Post and Vliet (2005) uses and cites Ang Chen Xing (2004), which is missed from the references' section of Post and Vliet (2005). On the other hand, Post and Vliet (2005) use Equation (5) of downside beta of Ang Chen and Xing (2004), the same equation is given in Ang Chen and Xing (2006). Therefore, I Chen Xing (2006), Ang, Vliet is later (2005) Post that as and the published that and paper using anticipate

198

in line with the consumption smoothing of the inter-temporal asset pricing models. Even more, Cochrane (2006) points out that when the marginal utility of investor's wealth is high in bad times of the economy, expected return is high. The time-varying risk premium is supported by Ferson, Kandel and Stambaugh (1987), and Lettau and Ludvigson (2001).

Post and Vliet (2005) overcome this problem. They point out that they perform conditional tests to take into account the time variation in risk and risk aversion and this conditional on the

allow for market risk premium of downside risk to be time-varying

economic conditions. They reported that they find downside beta in the US captures the cross-sectional variation in returns on stocks superiorly better than beta of the CAPM.

Furthermore, they pointed out that they find this occurs in particular in bad times of the economy. However, Post and Vliet (2005) state that their findings of the superiority is not strong over Ang, Chen and Xing's of

downside risk over beta of the CAPM

(later

published as Ang Chen and Xing (2006)) sample period. In addition, (2005) point out that the two studies differ in the methodology

Post and Vliet

and the data. This raises

the issue of whether the priced downside risk premium that has been found by Ang Chen and Xing (2006) is applicable particular during all the stages of the business cycle, or during a risk has almost

economic state as Post and Vleit (2005) find that downside

perfect relationship with returns in bad times.

Therefore, downside risk-return relationship of Ang, Chen and Xing (2006) be investigated. This the to chapter attempts to economic on conditions needs conditional

199

issue. by first between downside It this the starts risk and examining study relationship returns on stocks by applying Ang, Chen and Xing's (2006) study on the UK market. Then it follows Post and Vliet's (2005) downside risk conditional test approach which includes splitting the full sample period into two periods. In the next step it reapplies Ang Chen and Xing's (2006) study over the recession and expansion periods, separately. This allows downside risk premium to vary with the business cycle conditions. Post and Vliet (2005) indicate that they conduct the conditioning test by splitting sample period into two periods of good and bad times and then re-conduct the test for these two periods separately and argue this is a simple conditioning approach.

Therefore, this chapter contributes to downside risk literature by first examining Ang, Chen and Xing's conditions (2006) downside risk premium conditional on business cycle

and secondly, by providing

an out-of-sample

test for Ang, Chen and Xing's

(2006) study. This is an important robustness exercise. Ang, Chen and Xing (2006) point out that the investor in their downside risk model is more concerned about losses than is attracted to gains. However, there is no reason to believe that investors in the UK market have such preferences. Blake (1996) reports that he finds a 35.04 weighted-average

coefficient of relative risk aversion in UK. In addition Blake (1996) points out that Mankiw and Zeldes (1991) finds a 26.3 coefficient for USA. Black and Fraser (2000) point out that they find the UK investors are highly affected by future economic uncertainty, and different from investors in the other countries including the US27.

'' Black and Fraser's (2000) study sample includes UK, Germany, USA, Australia, and Japan.

200

The remainder of the chapter is organized as follows: Section (4.2) includes the relevant literature review on downside risk and time-varying risk premium. Section (4.3) develops the chapter's testable hypotheses. Section (4.4) discusses the data and the followed by Section (4.5) which discusses the findings. Finally Section methodology, (4.6) concludes.

4.2

Literature

Review

4.2.1

Downside

Beta

Sharpe (1964) states that, in developing his CAPM, investor's preference is summarized by the expected return and its square root of variance (standard deviation). Markowitz (1959) points out that the positive and negative extreme returns are treated similarly under the variance. Bawa and Lindenberg (1977) use the lower partial

movement as measure of risk. They indicate that a criterion that uses this measure of risk along with the mean to be the measure of return works under any distribution of returns
on stocks and the mean-variance framework partial moment rule when normality the lower-partial moment becomes a special case of this mean-lower

assumption holds. Nawrocki28 (1999) points out that different types of utility functions of Von

accommodates

Neumann- Morgenstern as well as different risk attitudes and hence does not restrict the analysis to the quadratic utility measures of risk. function as in the case of variance and semi-variance

28 Available at http: // www56. homepage. villanova. edu/david. nawrocki/ Brief%20History Downside%20Risk%20%2ONawrocki. pdf

%20of %20

201

For the definition of downside risk, Price, Price and Nantell (1982) define the lower partial moment as a risk measure that is function of the returns dropping below a particular target, where the reference is to portfolio's risk and return. Bawa and

Lindenberg (1977) define the lower-partial moment risk as the risk that occurs when the market return falls below the risk-free rate of return. They point out that this latter rate of return is an opportunity cost that results form investor's choice of making risky
investment. 29 Ang, Chen and Xing (2006) use average market excess returns as well as

zero return and risk free rate of return as cutoff points for measuring downside risk. In addition, Ang, Chen and Xing (2006) define upside risk as the covariance of stock's return with the market return when the latter is above the cutoff point.

As mentioned

earlier,

Bawa and Lindenberg

(1977)

developed

an equilibrium

capital asset pricing model which employs the mean-lower partial moment. Bawa and Lindenberg (1977, p. 196, Eq.4 and Eq.5) derive the model as
E(R, )_r, (E(RA, )-rr: =nacina ) j= 1,2,..., M. (1)

CLPM,, (r1 ; M, J) LPM(r,:; M)

(2)

Equations (1) and (2) above are in Bawa and Lindenberg's (1977) notations. They define the notations in the above equations as follow: CLPM(r,; ; M, j) as colower partial
moment (n), M, RM, Rj as market portfolio, its return, and stock's j return, respectively,

M )LPM (r,,.; rate, and as risk-free r,:

as market portfolio's lower partial moment (n) at

29 Harlow and Rao (1989) point out that they find the target rate based on empirical data is the average market returns

202

30. Bawa and Lindenberg (1977) point out that in this framework if market return is r,; lower than the riskless rate of returns, then market portfolio will have risk.

In a recent study, Post and Vliet (2006) pointed out that they find the proxy for market portfolio is inefficient based on mean-variance rule but it is third-order stochastic dominance efficient and this mean-variance inefficiency can be explained by downside risk. They mention that the strong case for the mean-variance rule to be substituted with general rule comes from the empirical findings of the non-normality of returns on stocks' distribution and the psychological evidence of risk perception. Furthermore, they point out that although the issue of mean-variance inefficiency can be accounted for by other possible explanations, still a downside risk-based generalized CAPM can capture a number of financial anomalies.

Price, Price and Nantell (1982) formulate a theorem, in which they point out that when returns are lognormal, lower-partial is based that or on variance measure a risk

for but be for stocks with moment will stocks with average systematic risk, equal only high (low) systematic risk the former will be higher (lower) than the latter. They reported that they find empirical groups of stocks. Lindenberg. differences between these two measures of risk for the last two Bawa the of model and supports

They point out that this finding

Kim and Zumwalt (1979) developed what they call a two-beta model. They model into two

point out that this model divides the systematic risk in the single-market components; down-market

They risk. systematic reported and up-market systematic risk

30See Bawa and Lindenberg (1977, p. 192) for technical details of LPM,,

definition.

203

that they find that down-market beta is compensated with positive risk premium but upmarket beta has a negative price of risk. They point out that this is interpreted as while investors are ready to pay for taking upside risk, they require for downside risk a positive premium. Chen (1982) points out that Kim and Zumwalt's model suffers from and

heteroskedasticity that results from time-varying betas and from multicolinearity

these problems can be overcome by Bayesian time-varying beta model. He points out that the two-beta model is still valid under time-varying betas. He reports that he finds this time-varying model's findings confirm those of Kim and Zumwalt that investors ask for a positive premium for down-market risk while accept a negative premium for up-market risk. Furthermore, Chen (1982) points out that these results verify that downside beta is a better risk measure than the single market beta.

On the other portfolio's findings

hand Jahankhani

(1976)

indicates

that

the appropriateness

of

variance as its risk measure needs to be studied in the light of the unsupportive for the mean-variance CAPM. He reports that he finds empirically the two and

models; mean - variance and mean semivariance, slope's coefficients

fail to produce an intercept's framework.

that are in agreement with the underlying

Post and Vliet

(2005) point out that the inability mean-semivariance

of Jahankhani (1976) to find supportive evidence to the

CAPM over its variance counterpart is a result of not examining the Chen and Xing (2006) point out that

bear markets years. In a recent study, Ang,

Jahankhani (1976) and other early studies have not actually provided a direct examination have downside bearing is that not employed and the risk associated with of risk premium all individual find fail therefore to supportive evidence. stocks and

204

Ang, Chen and Xing (2006) propose a downside risk model. They point out that they assume investor's preferences are descried by Gul's (1991) rational disappointment aversion utility asymmetric function. They point out that in their model and framework, worry about downside the risk is for

and investors

risk and require compensation

bearing it and the CAPM beta is not the appropriate measure of risk. Furthermore, they indicate that under this utility function and assuming all other things are equal investors for

are prepared to give up part of the return, in form of a negative risk premium, investing in stocks that have high potential of upside risk. relationship

They point out that they

examine the contemporaneous

between downside risk and the cross-section as well as using individual stocks with

of US stock returns using portfolio Fama and MacBeth's (1973)

formation

cross-sectional

regressions. They report that they find

around 6% annual downside risk premium in the cross-section of US returns on stocks, which is robust for controlling for other effects, while robust results for a negative upside Furthermore, they point out that they find that

risk premium is not supported empirically.

except for highly volatile stocks, future downside beta is predicted by past downside beta. They point out that their methodology has a high statistical power as they use daily

returns daily returns over short 12- month periods instead of monthly returns over longer periods which suits the situation when betas are time varying.

However,

there is shortcoming

in Ang, Chen and Xing's

(2006) study which is

pointed out by Post and Vliet (2005). Post and Vliet (2005) point out that Ang, Chen and Xing (later published as Ang, Chen and Xing's (2006)) do not employ conditional tests

205

for downside risk. As mentioned earlier Ang, Chen and Xing (2006) allow for risk measuresto be time-varying but not downside risk premium.

Post and Vliet (2005) point out that they use a different methodology from that of Ang, Chen and Xing's (later published as Ang, Chen and Xing's (2006)) and use unconditional tests as well as also conditional tests to allow for time-variation in risk and downside risk premium conditional on economic states. Post and Vliet (2005) use the stochastic discount factor representation of asset pricing models and GMM, instead of beta representation. They point out that difference between the mean-variance CAPM and mean-semivariance CAPM is that in the latter model, the pricing kernel is linear in market return over losses and flat over gains. They point out that in light of the evidence of time-variation in risk and risk-aversion, the conditional asset pricing models is the appropriate choice. Furthermore, they state that they conduct the conditional test by using the median of a conditioning variable to split the full sample period into good times and bad times and then re-conduct the GMM chi square test for the two good and bad periods individually and also measure the fit of entire sample. They point out that they find the unconditional and conditional (on state of the economy) mean- semivariance CAPM do better in explaining the cross-sectional returns on stocks than its mean-variance counterpart, and US stock returns are driven by conditional downside beta. In addition they state that these results are robust for controlling for other effects including value and find downside More importantly they that they risk - return out others. point size among is during high is time premium which risk market near-perfect relationship particularly the bad statesof the economy

206

As cited above Ang, Chen and Xing (2006) and Post and Vliet (2005) provide support to downside risk in the US stock market. However, Post and Vliet (2005) point out that they find over the sample period of Ang, Chen and Xing (later published as Ang, Chen and Xing's (2006)), the results that the mean-semivariance CAPM is better than the mean-variance CAPM is not strong and the strong evidence occur during bad economic conditions. Post and Vliet (2005) criticize Ang, Chen and Xing's (later published as Ang, Chen and Xing's (2006)) study as using a questionable downside risk's definition, insample estimates, Fama and MacBeth's (1973) methodology, and unconditional test which led to their findings.

Bali, Demirtas and Levy31 study the inter-temporal relationship. They point out that the importance of downside risk stems from different reasons; one of these is the empirical failure of supporting a normal distribution for stock returns. They report that they find downside risk is positively related to the expected market returns regardless of the how downside risk is measured whether it is measured using value-at-risk, tail risk or expected shortfall.

4.2.2

Downside Risk in UK

Pedersen and Hwang (2007) point out that the problem of variance as symmetrical

measure of risk can be overcome by the lower partial moment of Bawa and Lindenberg (1977) which is asymmetrical. They state that their study examines the percentage of UK
31Available at: http: //w4. stern.nyu. edu/finance/docs/pdfs/Seminars/06 If-balipdf. September 2008. Final Access on 18

207

individual stocks that downside risk, measured by the lower-partial moment CAPM, better describe than CAPM beta. They point out that this is essential in order to know how downside risk affects individual stocks and whether it is a potential risk factor. They indicate that they use different return frequencies on the largest stocks (FTSE100), as well as on the (FTSE250) constituents and on the small stocks (FTSE SmallCap) that are available over the entire sample period. They point out that they find normality assumption is not appropriate for high frequency (weekly and daily) returns. They report that they find 23% more of small stocks for daily frequency are explained by the lower partial moment CAPM compared with the CAPM. They point out that the results imply that measuring risk should be customized to asset classes such as small stocks with daily frequency. Furthermore they point out that downside beta is better than CAPM beta, but its additional value may not justify using it in asset pricing models and the CAPM is the recommended model for normal returns.

Olmo (2007) formulate an economy in which investors are mean-variancedownside risk averse. He points out that in his model the stock's risk measure is the
weighted sum of its CAPM extends the CAPM beta and its comovement with downturn markets and this

downside by risk. He point out that even upside model considering

he indicates He investment. have that uses weekly returns on effect on movement still can UK sectoral indices and FTSE 100. He points out that he finds that stocks that covary positively (negatively) with the down market such as Chemical (Telecommunications)

have higher (lower) returns than estimated under the CAPM, affected by down market such as Oil and Gas.

while other stocks are not

208

4.2.3

Business Cycle and Time-Varying

Risk

Fama and French (1989) report that they find there is a negative relationship between expected stock returns and business conditions. They point out that this is in agreement with the consumption smoothing of asset pricing models, where investors increase (decrease) their saving in times of high (low) income which results in lower (higher) expected returns. Nevertheless, they indicate that this time-variation in returns may reflect changes in the risks of stocks. Furthermore, Fama and French (1989) point out that their findings are supportive to Chen, Roll and Ross's (1986) findings that small stocks have higher expected return and risk than large stocks. Perez-Quiros and Timmermann (2000) reported that they find the conditional distribution of returns on
stocks is asymmetrical between expansion and recession periods and it is more

asymmetrical

for small stocks than large stocks. They point out that this is because during

recession the tighter credit conditions have more adverse effect on small stocks' risk than large stocks which results in an increase in small stocks' expected returns in this bad time of the economy. In addition they pointed out that the increase in expected stocks returns (for both small and large) during recession reflects an increase in both the level and expected price of risk. Furthermore Perez-Quiros and Timmermann (2000) state that their

results indicate that asymmetries in the stocks' risk and their expected returns over the business cycle should be modeled in the cross-sectional returns on stock studies.

Lettau and Ludvigson (2001) reported that they find their conditional (scaled) Fama French's CAPM the the of and performance matches consumption version of (1993) three factor model. They point out that risk (risk level or risk aversion) is higher

209

during bad times of the economy than during good times. Furthermore, they point out that some stocks covary more with the growth in consumption during weak economic conditions than during strong economic conditions which makes the conditional version of the model more appropriate to describe the cross-section of returns on portfolios of stocks as the conditional model captures this time-varying risk premia. Furthermore, they point out that the risk of a stock is economic state-dependentas it varies over the state of the economy.

In addition, Ferson and Harvey (1991) point out that according to asset pricing the variations in the measure of risk and price of risk cause the predictable changes in the returns on the stocks. They point out that however less work is done on the latter source of variation. They report that they find the stocks returns' predictable variation is mostly captured by the market risk premium. Furthermore, they point out that they find the return's predictability results mainly from time-varying expected price of risk rather than the measure of risk. In addition, they indicate that their findings imply that the market risk premium is time varying conditional on the business cycle conditions.

Also Campbell (1998) points out that in the pricing model the risk-factor should be conditional on the state of the economy. Furthermore, Boyd,

premium Hu and

Jagannathan (2005) report that they find the reaction of stock market to the news of unemployment differs based on the state of the economy. They point out that they find when unemployment increases in expansion (recession) news

stock prices increase (fall)

periods. In addition, they point out that the information

contained in unemployment

210

differs based on the state of the economy and hence the return's sensitivity

to such news

is dependent on the economic state. In addition, Yogo (2006) points out that he finds the equity premium is countercyclical when investor's procyclical marginal utility as returns on stocks are low during recession periods is high and the value and small stocks have more

returns than their counterparts.

4.3 Hypotheses Development

This chapter attempts to contribute to the literature of downside risk and asset pricing by examining Ang, Chen and Xing's (2006) downside risk study conditional on the state of the economy. This is also important in the light of Post and Vliet's (2005) argument who pointed out that the superiority of downside risk over CAPM beta occurs mainly during bad times of the economy.

First, Post and Vliet (2005) point out that Ang, Chen and Xing (later published as Ang Chen and Xing (2006)) use a questionable downside risk's definition. However, it

could be argued that, Ang, Chen and Xing (2006) use a measure of downside risk that is consistent with their assumption of investor's disappointment and Xing (2006) point out that under this utility downside (upside) risk premium. This is different utility function. Ang, Chen

function there is a positive (negative) from Post and Vliet (2005). Post and

Vliet (2005) point out that their pricing kernel is flat over gains. Second; Ang, Chen and Xing (2006) state that the basis of the return and risk cross sectional relationship is to be

drawback hence by Post in-sample is tests as suggested a and not using contemporaneous;

211

Vliet (2005). Even more Ang, Chen and Xing (2006) report that they find past downside beta is positively related to future returns except for highly volatile stocks.

Third, Ang, Chen and Xing (2006) report that their findings based on the Fama and MacBeth's (1973) regression support their findings based on portfolio formations.

Therefore their support for downside risk is not dependent on Fama and MacBeth's (1973) methodology as criticized by Post and Vliet (2005).

In addition, Ang, Chen and Xing (2006) employ daily returns to estimate downside beta whilst Post and Vliet (2005) use monthly returns. Pedersenand Hwang (2007) point out that lower partial moment CAPM is preferable for high frequency data in the UK. Therefore this chapter applies Ang, Chen and Xing's (2006) study to the UK market. Then, it allows their downside risk premium to vary over the business cycle by splitting the full sample period into two periods following Post and Vliet (2005) and then reapplies Ang Chen and Xing's (2006) study over both periods separately. Post and Vliet (2005) argue that this is a simple conditioning approach and point out that this, what they call a "split sample approach", overcomes a number of problems which are related to

conditional models

4.3.1

Is Downside Risk a Significant Beta Factor in the UK?

The first step in this chapter is to examine downside risk in the UK market. Pedersenand Hwang (2007) point out that they study the percentage of UK stocks whose

212

returns are better captured by downside beta compared with CAPM beta. They point out that downside risk additional value does not justify using it in pricing models. Olmo (2007) proposes a model with investors who are mean-variance-downside risk averse and applies it to UK sectoral indices over a short period of time. Therefore, this chapter attempts to examine whether downside risk is able to explain the cross section of UK stock returns by applying Ang, Chen and Xing (2006) study to London Stock Exchange.

This provides an out-of-sample

test for Ang, Chen and Xing's (2006) study. This is

important as Ang, Chen and Xing (2006) point out that the assumption of disappointment aversion utility function leads to a priced downside risk in the cross-section of returns on stocks. They also report that the US investor's price downside risk. On the one hand,

Blake (1996) points out that he finds a 35.04 average coefficient of relative risk aversion in the UK. Also Blake (1996) points out that Mankiw and Zeldes (1991) find in US a 26.3 coefficient. Furthermore, Black and Fraser (2000) report that they find UK investors behave differently from other investors including the US investors. On the other hand Pedersenand Hwang (2007) report that they find high frequency (including daily) returns on UK stocks are not normal. Furthermore, they point out downside beta is better than CAPM beta but it is doubtful that it can be a risk factor that can significantly improve the pricing models. In light of this, the first hypothesis is stated as: Hypothesis 1: Downside beta is a significant risk factor in the UK cross-sectional returns on stock.

213

4.3.2

Downside Risk and Firm Characteristics

If the relationship between downside risk and the cross sectional returns exists in UK; i. e. if the first hypothesis holds, then the chapter examines the robustness of downside risk in the UK market. Ang, Chen and Xing (2006) report that they find downside risk is robust in the US cross-section of returns for controlling for several cross sectional characteristics such as size and book-to-market value among others.

In addition, Post and Vliet (2006) point out that an empirical shortcoming of the mean-variance principle is that stock returns are not normal, and Pedersen and Hwang
(2007) report that they find the UK small stocks are more non-normal than large stocks

and downside beta has an important role for small size stocks which are the main benefited of this measure of risk. Furthermore, they *point out that estimating risk has to be customized to small stocks with daily returns. Even more, they point out there is a relationship between size and downside risk and the former is important in deciding what model to use for risk (downside beta or CAPM beta). In addition Perez-Quiros and Timmermann (2000) point out that returns on small stocks and their risks are more Ferson Moreover, for business large the than and asymmetrical cycle. stocks across Harvey (1991) pointed out that during recession times, the betas of small stocks are high In high in low large this time those addition, premium. expected are of of stocks while Yogo (2006) reports that he finds the small and value stocks have more procyclical hypothesis is Therefore the their stated as second counterparts. returns compared with follows;

214

Hypothesis 2: Firm's size and book-to-market value are important factors in deciding the existence and significance of downside beta and its superiority to CAPM beta.

Therefore, hypothesis (2) tests the robustness of downside risk in the UK. Furthermore, Post and Vliet (2005) report that they find downside risk is better than CAPM beta in each section of the market (divided based on size, book-to-market or momentum) in the US. Similarly, Hypothesis (2) tests whether the cross-sectional downside risk-return relationship in UK is different across market capitalization and book-to-market stock categories and whether downside risk-returns is better than CAPM beta-return relationship across these categories as Pedersenand Hwang (2007) report that small stocks are the main benefited from downside beta.

4.3.3

Downside

Risk and Business Cycle

Post and Vliet (2005) criticize Ang Chen and Xing's (later published as Ang Chen
and Xing (2006)) study as for not examining conditional downside risk. In addition they

report that they find bad times of the economy have higher return and risk than good times. Pereze-Quiros and Timmermann (2000) point out that Fama and MacBeth's (1973)

method tests an average price of risk whilst the price of risk could be state-dependent and takes a significant value in one or more states of the economy while has a zero average.

Furthermore, they point out that it is important to model the asymmetries in stocks' risk

business despite based Therefore, the this, the fact that Ang, on across and return cycle.

215

Chen and Xing (2006) report that they find downside beta is robustly and positively related to average returns, if downside risk premium is time-varying, then similar results may not be found in other markets without allowing for downside risk premium to vary over the business cycle as Pereze-Quiros and Timmermann (2000) stressed. Furthermore Perez-Quiros and Timmermann (2000) point out that risk premia in the pricing model should vary substantially across (conditional on) the expansion and recession times in order for the model to have explanatory ability for the time-varying expected returns.

In addition, Fama and French (1989) report that they find expected returns on stocks are higher during weak economic conditions and lower during strong conditions of the
economy. Post and Vliet (2005) point out that they perform conditional tests of downside

risk, conditional on the state of the economy, to allow for time-variation in the risk, risk aversion and risk premium. They report that they find downside risk is superior to CAPM beta in describing the US cross-sectional returns on stocks. However they point out that they find an almost perfect relationship between the two variables (downside risk and return) occurs during bad economic times which are characterized by high equity
premium. Furthermore, Post and Vliet (2005) point out that during the sample period of

Ang, Chen and Xing's (later published as Ang Chen and Xing (2006)) study the evidence is not strong for the relative superiority of downside risk. Consequently, hypothesis (3) of this chapter examines the downside risk premium of Ang, Chen and Xing's (2006) model conditional is hypothesis Therefore, third the the the state on of economy.

Hypothesis 3: The risk premium of downside risk is time-varying conditional on the state during during it is higher than the expansion. recession economy and of

216

4.3.4

Downside Risk and Industry

The case for the potential importance of the industry in deciding the significance of the downside beta in explaining stock's returns is strong. Yogo (2006) point out that the marginal utility is countercyclical and value and small stocks have high returns because they have more procyclical returns than growth and large stocks. Furthermore, PerezQuiros and Timmermann (2000) point out that stock returns are asymmetrical between expansion and recession and this asymmetry is larger for small stocks than for large stocks. Black and Fraser (2000) point out that there the UK stock market may have higher cyclical stocks' percentage than other markets. Olmo (2007) reports that he finds for the UK sectoral indices, industries that co-vary positively (negatively) with downturn market have higher (lower) returns than that under the CAPM while downside beta is not priced in industries that do not move with the down market. Therefore, the fourth hypothesis is stated as follows; Hypothesis 4: Downside beta is priced within some industries but not within every industry.

Olmo (2007) uses sectoral indices and very short weekly sample period from 112003 to 4-2006. This chapter is different from his study in that it examines downside riskreturn relation within each industry using individuals framework different to and a stocks of Ang, Chen and Xing (2006)

that of Olmo. It applies the model and methodology within each industry.

Ang, Chen and Xing (2006) predict as a positive downside risk while Olmo (2007) points out that his to the CAPM beta a measure of

premium

and a negative upside risk premium to include

model extends the CAPM

in addition

217

downside beta. He points out that his measure of stock's risk is a weighted sum of both stock's CAPM beta and stock's comovement with down markets.

4.4 Data and Methodology

4.4.1

Data

Daily and monthly data on 2020 UK common stocks traded on LSE for the period of July 1981 to December 2005 are obtained from Datastream. These are the same 2020 stocks used in Chapter (3) of the thesis. Daily and monthly returns on FTSE-all share as a proxy for the return on the market is also obtained from Datastream. Also book-to-market value (book value divided by price) and market capitalization for each stock at monthly frequency are obtained from Datastream. Industry code for each individual stock also obtained from Datastream. The industry code from the Datastream for each company is used to group the stocks into their corresponding industries. In addition, the dates of the UK recession and expansion periods are obtained from the ECRI- Economic Cycle Research Institute's website (http: //www. businesscycle.com) and these dates are used to split the sample period into expansion and recession periods. The business cycle dates obtained from ECRI that occur during the sample period of the chapter (July -1981 to December-2005) are Peak (May-90) and Trough (March -1992).

218

The UK coincident index (UKCI) is also provided by the Economic Cycle Research Institute (ECRI32) and Figure (4.1) depicts the changes in this UK coincident index (UKCI). The Figure shows the biggest negative change in this index appeared to have happened during the recession period where the latter is identified by the dates from the ECRI as mentioned above.

Table (4.1) shows descriptive statistics for the sample over the period July- 1981 to December-2005. The FTSE-all share is used as proxy for the value-weighted average return on the UK stock market. In addition an equally-weighted market portfolio is constructed using all the stocks in the sample. Fama and French (1989) point out that
value-weighted portfolio attaches more importance to large stocks' returns while equally-

weighted portfolio attaches more importance to small stocks' returns.

The descriptive statistics displayed in Panel (A) is for daily frequency while Panel (B) displays the statistics for monthly frequency. Jarque-Bera test is used for testing normality following Pedersen and Hwang (2007). Normality assumption for the

distribution of returns on the UK stock is rejected by Jarque-Bera test for both valuefrequencies. daily for returns, and weighted and equally-weighted market monthly Although both daily and monthly UK returns are found to be non-normal, the daily Daily from deviation the than market returns. normality monthly show more returns returns have higher kurtosis than monthly market returns for both value-weighted and daily have the market returns market portfolios equally-weighted and equally weighted

32We are grateful to the Economic Cycle Research Institute (http: //www. businesscycle.com) for providing index index. (UKCI) Coincident UK the with us

219

higher negative skewness than the equally-weighed monthly returns. On the other hand, have higher kurtosis daily the and negative equally-weighted market returns while frequency, For daily the than the returns. monthly market value-weighted skeweness kurtosis have higher and negative skewness than the market returns value-weighted equally-weighted market returns.

Pedersen and Hwang 33 (2007) report the UK daily and weekly returns are nonnormal. They point out that the FTSE-all share daily returns' non-normality their fat tails. By examining non-normal arises from

Panel (A) of Table (4.1) of this chapter, it appears that the skewed especially for the equally-

daily returns are kurtoic and negatively

find Hwang Pedersen (2007) Furthermore, they that and report weighted market returns. the normality is appropriate assumption for monthly returns on FTSE all share, while

Panel (B) of Table (4.1) of this chapter rejects the normality assumption for both monthly value-weighted and equally-weighted market returns.

Fama and French (l 989) point out those large stocks are relatively

more

represented in value-weighted portfolio while the small stocks are more represented in the equally weighted portfolio. Pedersen and Hwang (2007) point out that the UK small CAPM is downside benefited from daily the risk while are most returns stocks with appropriate for large stocks with monthly returns. Furthermore, they point out when is better CAPM lower does hold, the then a choice partial moment not returns' normality beta in downside light Table (4.1) this chapter In CAPM. this, supports using of over the for UK daily and monthly returns, as both of them are non-normal, and for the daily
" Pedersen and Hwang (2007) sample period is from the first of August, 1991 until the 31 of July, 2001.

220

returns on small stocks as the small stocks, consistent with Pedersenand Hwang (2007), show more non-normality than large stocks, this is implied by the equally-weighted market returns results compared those with the value-weighted market returns34.In deed, Pedersen and Hwang (2007) point out that the frequency of return and stock's size are important in determining whether returns on stocks are normal and when daily frequency is combined with small stocks the lower partial moment CAPM is the proper model.

4.4.2

Methodology

4.4.2.1 The Downside Risk Model of Ang, Chen and Xing (2006)

This chapter applies Ang, Chen and Xing's (2006) model. Ang, Chen and Xing (2006) assume an investor has Gul's (1991) rational disappointment aversion utility function. Ang, Chen and Xing (2006) point out that under this utility function, investor is more concerned about downside risk and this results in downside risk being priced in the

crosssectionalexpectedreturnson stocksand CAPM beta is not the adequate measure of

stock's risk. They point out that they use as downside risk's measure that of Bawa and
Lindenberg's (1977) downside beta.

Ang, Chen and Xing (2006, p. 1197, Eq.5) define their downside beta as:

34Similarily, Bali, Demirtas and Levy Final Access on 18 (Available at http://w4. stern.nyu. edu/finance/docs/pdfs/Seminars/061f-bali. pdf . September 2008.) point out that they find, VaR is better in predicting returns on the equally-weighted index as small stocks have more non-normal returns than larger stocks. They comment further that is because equally weighted index weighs smaller stocks more than value weighted index.

221

cov(r,, rm
var(r,, \

\ rm
rm <p)

<Pm

(3)

Equation (3) above is in Ang Chen and Xing's (2006) notations. They define the notations in the above equation as follows r, , r,,, and im are the excess return on security i, the excess return on the market portfolio and the cut-off point between down and up markets. They use for the latter i. e. the cut-off point, the average market excess
return, although they use alternatives to is as a robustness check.

Ang, Chen and Xing (2006) point out that the disappointment aversion investors are have (i. high invest high in times to that e. wealth return at of prepared stocks produce high positive covariation with the upside market), assuming all other thing are equal, at a
discount Ang, Chen and Xing (2006, p. 1 199, Eq. 9) estimate upside risk as:
cov(r, 18+ = r, \ rm > m 1()

var(rm

rm

>p

Equation (4) above is in Ang Chen and Xing's (2006) notations.

Then this chapter aims at examining the downside risk premium of Ang, Chen and
Xing (2006) conditional on business cycle conditions. The approach used is applying

Ang, Chen and Xing's (2006) model of downside risk and their study and methodology to the UK market over the full sample period following and expansion periods separately. This conditioning them and then over the recession following is applied approach Post

bad into times and Vliet (2005) two and of good periods period splits sample which and reapply the test over the periods individually. Ang, Chen and Xing (2006) use both

222

portfolio formation and Fama and MacBeth's (1973) cross sectional regressions on individual stocks.

4.4.2.2 Risk Measures

To examine the relationship between risk and average stock returns, Ang, Chen and Xing (2006) estimate CAPM beta ('3 ), downside beta (-) and upside beta (+) of

individual stocks, Ang, Chen and Xing (2006, p. 1236, Eq.B-7 and Eq.B-8):
ritrnn lrir j) _ 1 j)_I nn<M N+ rmr rmt irnir>iry _ L rmt (Yt (5)

7/\

rZ

Equation (5) above is in Ang Chen and Xing's (2006) notations, they define the notations in the above equation as follows; r,,,, and r, as the demeaned excess returns (r;,+) as the demeaned

on the market portfolio and stock i, respectively, rm, (r,,; ) and i-

excess returns on the market portfolio and stock i, respectively, when the excess return on the market is below (above) the cutoff point or target (A. ). Ang, Chen and Xing (2006)
use as cut-off market portfolio, point between down and up markets the average excess return on the the zero rate of return and risk-free rate of return and point out that the

results are robust to the cut-off point. Furthermore, Ang, Chen and Xing (2006) point out that that they use a 12-month (t - t+12) interval, i. e annual horizon, of stock's daily returns to estimate stock's betas as this compromises between an enough number of observations for estimating the risk measures conditional on the down or the up markets and not too

long period for not allowing for time-variation in the risk measures. They point out that

223

those stocks with a number of missing observations that exceed five are excluded. Furthermore, Ang, Chen and Xing (2006) estimate CAPM beta (), relative downside beta, which they define as (, - - ) the difference between downside beta and CAPM beta, and relative upside beta, which they define as (+ -), the difference between

upside beta and CAPM beta.They predict a positive relation between downside risk and stocks returns and a negative relation between upside risk and stocks returns. This chapter follows Ang, Chen and Xing (2006) in estimating all the above betas following their procedures using only zero rate of return as the cut-off point.

4.4.2.3 Portfolio

Formation

4.4.2.3.1 Risk Sorted Portfolios

Ang, Chen and Xing (2006) point out that to examine the contemporaneous relationship between average excess returns and downside risk they form equallyweighted five quintile portfolios by sorting all the stocks on the basis of their estimated risk characteristcis - or what they call it realized - CAPM betas (), downside betas (-) or upside betas (, 6'), relative downside beta p- -, among other measures, at the
beginning of the 12-month interval which is used for estimating these risk measures.

They point out that then they calculate the equally-weighted realized excess returns on is five 12 that the these months of portfolios over very same period of used to each estimate the betas by calculating the cumulative excess return on each individual stock. Furthermore, Ang, Chen and Xing (2006) point out that these step are repeated at the

224

beginning of each month, even though 12-month of daily return is used as (horizon) a base for estimating betas. They indicate that for this reason they adjust for the overlapping in the sample periods by calculating Newey-West (1987) with 12 lags based t-statistics for the difference between the average excess return on the highest beta portfolio and the average excess return on the lowest beta portfolio. Furthermore, Ang, Chen and Xing (2006) point out that they also used monthly returns with 60-month interval as estimation period for betas and returns and found similar results. This chapter follows Ang, Chen and Xing (2006) in forming the portfolios as described above and estimating the risk measures using both daily returns and monthly returns.

The importance of the effect of return frequency on downside risk - average return relationship in UK is stressed by Pedersen and Hwang (2007) who point out that frequency has a significant role in deciding the normality of stock returns and report that that they find for the monthly frequency of returns and especially for large stocks CAPM beta is appropriate.

In addition Ang, Chen and Xing (2006) point out that to make use of downside risk relationship with contemporaneous average stock returns, downside beta should be

predicted. Ang, Chen and Xing (2006) point out that in order to examine this they employ Fama and MacBeth (1973) cross-sectional regressions, in which they regress realized

relative downside beta measure to be predicted on a number of stock characteristics that are the investor's information set beforehand, which include35, among others, stock's (1)

's Ang, Chen and Xing (2006) use in addition to firm characteristics other risk variables to predict downside risk measure.

225

past downside beta measure, (2) past standard deviation, (3) logarithm of market capitalization (size), (4) book-to-market value, (5) past 12-month excess returns. They point out that the first two variables are calculated over the preceding 12 months using daily returns and the next three variables are calculated at the beginning of the period. They explain that they run different monthly cross sectional regression specifications which include regressing realized relative downside beta measure on one firm

characteristic at a time with industry dummies and then they run one regression of realized downside beta measure on all firm characteristics and industry dummies all together. They point out that becausethese regressions are carried out on a monthly basis, they use Newey-West (1987) with 12 lags to correct for overlapping. This chapter follows them and applies their procedure to predict UK downside beta and CAPM beta.

This chapter also follows investable portfolios.

Ang, Chen and Xing (2006) in forming out that they form five quintile

what they call at the

They point

portfolios

beginning of the month by sorting all the stocks on the basis of their past downside beta estimated over the previous 12-month interval of daily returns. Then they calculate the equally-weighted (average) realized excess return on each of these portfolios over the

next month, in addition they calculate the difference between average excess return on the highest past downside downside beta portfolio beta portfolio and average excess return on the lowest past

with Newey and West (1987) t-statistics. This chapter follows excess returns over the next month as well as

them and estimates the average portfolio

over the next 12 months and average returns are reported on annual basis.

226

4.4.2.3.2 Accounting for Size and Book-to-Market

Value Effects

Post and Vliet (2005) point out that they use the double sorting procedure to control for size and book-to-market among other effects. They explain that they first form two

portfolios by sorting the stocks based on one of the characteristics and then within each of the two portfolios, stocks are sorted into 10 portfolios based on either their CAPM beta or

downside beta. In addition, Ang, Chen and Xing (2006) use this procedure to control for, for example, co-skewness, They explain the details as follows; into quintile portfolios they first sort all stocks

based on their co-skewness at the beginning of every 12-month

period where the latter measure is estimated over this subsequent one year period. They point out that then in the second step, within each of these five portfolios, stocks based on their downside beta into equally weighted five quintile then the realized excess return on each downside beta portfolio they sort all portfolios and

is averaged over the first-

step five quintile co-skewness portfolios to form downside beta portfolios that control for co-skewness. They point out that the co-skewness measures, downside betas and realized

excess returns all are calculated over the same 12-month period using daily returns and they calculate Newey-West (1987) with 12 lags to compute the t-statistics for the difference between average excess returns on the highest and the lowest downside beta portfolios within each of the five co-skewness portfolios and for the co-skewness controlled five downside beta portfolios. This chapter applies this quintile double sorting procedure as in Ang, Chen and Xing (2006) to control for size and book-to-market value; i. e. the stock's market capitalization or book-to-market value replaces the co-skewness. The market capitalization or book-to-market value calculated at the beginning of the 12for for is because Chen Xing (2006) Ang, the this and reason period, control month when

227

these characteristics in the Fama and MacBeth's (1973) cross-sectional regression, they calculate the log size and book to market at the beginning of the 12-month period (see next sub-section).

4.4.2.4 Fama and MacBeth's (1973) Cross-Sectional Regression

Ang, Chen and Xing (2006) point out that they use Fama and MacBeth's (1973) cross-sectional regressions on individuals stocks to show that downside risk is in fact not the same as other firm's effects that are found empirically to explain the cross sectional returns on stocks. Ang, Chen and Xing (2006) point out that they run Fama and

MacBeth's (1973) cross sectional regressions of 12-month excess returns of individual stocks calculated over the subsequent 12 months on (1) stock's characteristics including, among others, logarithm of market capitalization and book-to-market ratio, both calculated at the beginning of every period and (2) contemporaneous risk measures including, among others, CAPM beta or downside beta and upside beta, calculated over the same 12 month period) and this regression is repeated at monthly frequency. They point out for this overlapping they use Newey-West (1987) with 12 lags is to calculate the t-statistics. This chapter follows them in their application of Fama and MacBeth's (1973) cross-sectional regressions to control for these two characteristics.

In detail, Ang, Chen and Xing (2006) run different specifications of Fama and MacBeth's (1973) regression, among these which are applied in this chapter are; individual stock annual realized excess return on (1) downside beta and upside beta (2)

228

downside beta, upside beta, logarithm of market capitalization and book-to-market, among other variables (3) CAPM beta and (4) CAPM beta and logarithm of market capitalization and book-to-market, among other variables.

4.4.2.5 Downside Risk Premium and Business Cycle

After applying

Ang, Chen and Xing's

(2006) model and methodology

to the UK

market over the full sample period following contribution which is examining

them, this chapter proceeds to its main

if the risk premium of downside risk of Ang, Chen and

Xing (2006) is conditional

on the business cycle and whether it provides different results risk premium. formation It does so by reapplying their model and

from that of the unconditional methodology (both portfolio

and Fama and MacBeth (1973) cross sectional)

regression over expansion and recession periods, separately.

To condition cycle conditions,

downside risk premium of Ang, Chen and Xing (2006) on business this chapter divides the full sample period into two periods as in Post

and Vliet (2005) who carry out the conditional downside risk tests by splitting the sample period into two periods. However, Post and Vliet (2005) use the median of a

conditioning

variable (dividend

yield, credit spread or earnings yield) to split the full recession

into into bad but the times, this sample splits and chapter period good sample and expansion periods following Antoniou,

Lam and Paudyal (2007). They point out that

business dates Research Cycle Institute's Economic the the cycle to split their they use of in UK the into the to the study momentum over recession and expansion sample period

229

business cycle. Therefore, this chapter, as mentioned in the Data section above, splits the
sample period into two economics states; i. e. recession and expansion guided by the as 12 months of

business cycle dates from the Research Cycle Institute36. Furthermore, daily returns is used to estimate the betas, it is difficult

to restrict the recession and

expansion periods exactly to the months obtained form the ECRI. Therefore, the rule applied is if any of the months used to calculate the beta happened to be during the recession period, then that beta is considered to be part of the recession periods.

4.5 Results

4.5.1

Downside Beta and Realized Returns

This sub-section tests the first hypothesis which examines the significance of downside risk in the UK market. Table (4.2) presents the results of portfolios formed on CAPM betas (Panel A) downside betas (Panel B) relative downside betas (Panel C), and upside betas (Panel D), for the whole sample period using daily returns over 12-month period. Ang, Chen and Xing (2006) report for each portfolio the equally weighted excess
return (realized), CAPM beta, downside beta, upside beta, where they calculate the three measures over the same one year period, the difference and quintile (1) portfolios and the difference between the beta of quintile (5)

between the average excess return on the Similar

highest risk portfolio

and average excess return on the lowest risk portfolio.

calculations are presented in this chapter or the UK market.

' We thank the Economic

Research Cycle Research Institute for providing

this information.

230

Panel (A) shows a positive relationship between CAPM beta ()

and average

excess returns from the lowest CAPM beta portfolio (quintile 1) to the fourth CAPM beta portfolio (quintile 4). However, the relationship breaks for the fifth highest CAPM beta portfolio which provides lower average excess returns than the previous lower risk (fourth) portfolio. In addition the difference between the yearly average excess return on the highest and yearly average excess return on the lowest CAPM beta portfolios is 4.02% which is positive but marginally insignificant at the 10% level. Higher CAPM beta is associated with both higher downside beta and higher upside beta and the difference between the beta of the highest CAPM beta portfolio and the beta of the lowest CAPM beta portfolio is positive and takes the values of 1.11,1.14,1.12 for CAPM beta,

downside beta and upside beta respectively, which are very similar in magnitudes.

Panel (B) shows similar findings for downside beta (/3-) and the average excess returns on portfolios sorted on downside beta. There is a positive relationship between downside beta and average excess return from the lowest to the second highest downside beta portfolios and then the relationship breaks for the riskiest (quintile 5) downside beta portfolio, which has lower returns than the previous lower risk downside beta portfolio. However the difference between the average excess return on the highest downside beta is 3.91% lowest beta downside the the portfolio portfolio and average excess return on which is positive and significant at 10%. This is unlike the spread for the CAPM beta indicate This is above. may a marginally significant as mentioned portfolios which potential slight improvement when using downside beta as measure of risk over CAPM beta, although the positive risk-return relation has not been restored for the riskiest

231

stocks. This significant difference in returns between downside beta portfolios is also accompanied by larger difference between downside beta of the highest downside beta portfolio and downside beta of the lowest downside beta portfolio, which is 1.62, compared with a corresponding differences for CAPM betas and upside betas of 0.80 and 0.62 respectively, which is consistent with Ang Chen and Xing's (2006) findings for USA. However, the findings so far UK is slightly different from US, in Ang, Chen and Xing (2006) report that they find for the US stock market a strictly positive and increasing relationship between downside beta and average excess return for the five quintile portfolios.

Figure (4.2) depicts visually the relation between the different risk measures and the corresponding portfolios' average excess return that are reported in Table (4.2). It shows relationship (Downside) appears to be smoother than CAPM between downside beta and

that downside beta -return beta-return relationship

(CAPM).

Indeed the relationship (quintile

average excess return for the first four portfolios while this is not the case of CAPM beta portfolios.

I to quintile 4) is almost linear

Although the results show downside risk is positively related to the cross section of stock returns except for the riskiest stocks, CAPM beta provides, to some extent, similar results so far. Ang, Chen and Xing (2006) point out that one potential explanation for the positive association between downside risk and average excess return is that investors are indifferent to downside risk and this positive relationship is a result of the high correlation between downside beta and CAPM beta which occurs by construction, as

232

downside beta sorted portfolios

produce spread in both downside beta and CAPM beta. by sorting stocks on relative beta provides incremental

They point out to examine that they construct portfolios downside beta (--, Q) to gauge how beta. much downside

influence over CAPM

This chapter applies their exercise on the UK stocks are

reports the results in Panel (C) of Table (4.2).

The results in Panel (C) shows that sorting on relative downside beta does not change the previous findings, reported in Panel (B), for the existence of a positive relationship between downside beta and average excess return for the first four downside beta portfolios; i.e. except for the riskiest portfolio. This relationship is also depicted visually in Figure (4.2) (RelDwn). Furthermore the difference between the average excess returns on the highest relative downside beta portfolio and the average excess returns on the lowest relative downside beta portfolios is more significant now at 5%. In addition and although, the spread in average excess returns between the third and second highest relative downside beta portfolios is small, the difference between downside beta of the highest relative downside beta portfolio and downside beta of the lowest relative downside beta portfolio is 1.11 while the corresponding difference in CAPM beta and upside beta are -0.02 and -0.36 respectively. This confirms that downside beta is behind the positive relationship between downside risk and returns. These findings are consistent with Ang, Chen and Xing (2006) for the US. In fact Ang, Chen and Xing (2006) report that downside beta and not CAPM beta is driving the positive relationship as sorting on betas downside produces no spread in CAPM betas over relative downside beta relative portfolios.

233

Panel (D) of Table (4.2) presents the results of sorting stocks into upside beta portfolios. It shows no obvious relationship between upside beta and average excess

highest difference between the the upside returns on excess average returns and although beta portfolio However, and the lowest upside beta portfolio (2006) is negative, it is insignificant. relationship

Ang, Chen and Xing

report that they find a positive

between upside beta and average excess return on the US stocks and indicate that this is not in agreement with disappointment aversion investors who are prepared to invest in

high upside variation stocks at discount. They point out that this is because upside beta is contaminated with CAPM beta or downside betas effects and therefore they study relative upside beta portfolios. and MacBeth's This chapter does not perform this exercise on UK market as Fama

(1973) cross sectional regressions show that upside beta is not priced in

UK at all as shown below.

Table (4.3) presents the results of Fama and MacBeth's regressions.

(1973) cross-sectional stocks at

The first regression shows that the results of regressing individual

excess returns on CAPM 10%.

beta, is a positive market risk premium which is significant

The third regression shows downside and upside risk premia that results from betas. downside and upside stocks average excess returns on stock's is 4.3%, which is positive and significant at 5%. This is

regressing individual Downside

risk premium

consistent with portfolio

formations findings in Table (4.2) and supports the potential that

downside risk is better than CAPM beta. However, upside beta premium is positive with is and magnitude statistically small a very insignificant. Hence, although downside risk

Chen Xing's in line Ang, is with and premium

(2006) predictions of positive downside

234

risk premium, upside risk premium in UK is not negative as Ang, Chen and Xing (2006) predict for disappointment aversion maximizers. They argue that under such utility function investors are prepared to sacrifice part of the returns on stocks that have high upside betas as this is the time of low marginal utility of wealth and this results in a negative sign to the risk premium associated with upside beta. And even more it is not priced. When size and book-to-market are included in the regression to control for size and value effects, downside risk premium remains significant at 5%. On the other hand, upside risk premium becomes also significant at 5% but positive and not negative. Nevertheless, the magnitude of upside risk premium is smaller than downside risk
premium which, in terms of magnitudes, supports Ang, Chen and Xing's and findings for the US that price of risk is asymmetrical (2006) model

between down and up markets

and the latter is smaller in magnitude than the former. Despite the fact that this chapter finds the sign of upside risk premium in UK is not negative as they expect. Ang, Chen and Xing (2006) also report that the sign of upside risk premiums in US is not robust and not stable as it changes after controlling for other variables and becomes insignificant. for by Ang, Chen as control

This chapter does not control for all the variables that are controlled

and Xing (2006) but for their regression that includes size and book-to-market variables the sign of upside risk premium is negative and significant.

Finally the second

regression of Table (4.3) shows that the market risk premium is also robust to size and book-to-market value effects and it becomes more significant.

In summary, the results in this chapter accept the first hypothesis that downside risk is a priced risk factor in the UK market with a 4.9% annual risk premium. This positive

235

relationship between downside risk and the cross-section of stock returns is not due to the CAPM beta but results from the additional power of downside beta over CAPM beta. Furthermore, the results support that downside beta is potentially a better measure of risk than CAPM beta in UK. This is evidenced by the significant spread between average returns on the highest downside beta portfolio and the average returns on the lowest downside beta portfolios, while the corresponding spread for CAPM beta portfolios is only marginally significant. In addition, the visual depiction of the risk-returns

relationships suggests an almost linear relationship between downside risk and average excess returns for the first four downside beta portfolios while the relationship is not such smooth for CAPM beta portfolios, although it is positive. However, the difference in the performance between the two models is not that significant and both models fail to price
the riskiest stocks (quintile 5) in line with a positive risk-return relationship. This is in

contrast with Ang, Chen and Xing (2006) findings relationship Nevertheless

for US stocks market of a positive

between downside risk and average returns over the five quintile portfolios. the results support Pedersen and Hwang (2007) who point out that asset by downside beta as risk factor.

pricing models will not improve significantly

4.5.2

Downside Beta and Size and Book-to-Market

Value

This section examines the second hypothesis of the chapter which test whether book-to-market and market capitalization are important determinants of downside risk it beta. Specifically if its CAPM examines and superiority over relationship return downside risk is a better driving risk factor of returns on stocks for a particular category

236

of stocks rather than for all stocks. This is motivated by the findings of Pedersen and Hwang (2007). They point out that downside beta is more important for small UK stocks as these are more skewed and indicate that this clarifies the association between size and
downside beta. Indeed Ang, Chen and Xing (2006) point out that the small stocks are

In from to that addition they point out that there exposed more risk results asymmetries. is a possibility that the results of downside risk are driven by small stocks and equally weighted returns and therefore they check that this is not the case using different robustness checks. In addition, this exercise provides, similar to Post and Vliet (2005), a further check that downside beta is robust to the size and book-to-market effects. They point out that they use the double sorting procedure to examine if downside beta captures these and other effects, by studying downside risk within different segments (small versus large, value versus growth, among others) and also compare downside beta performance with CAPM beta performance within each of these segments.

4.5.2.1 Downside Beta and Size

Table (4.4) presents the average yearly excess returns on the 25 portfolios resulting from first sorting on stock's market capitalization and then on stock's CAPM beta (Panel A) or downside beta (Panel B). In addition it presents the average excess returns on the five beta portfolios that control for size in the last row of each Panel.

Panel (B) shows the robustness of downside beta to size. It shows the same positive

from lowest downside downside beta between the excess returns and average relationship

237

beta portfolio

to the second highest downside beta portfolio,

which then breaks for the

riskiest downside stocks. The difference between the average excess return on the highest downside portfolio beta portfolio and the average excess return on the lowest downside Nevertheless, the difference beta

is positive but now is insignificant.

between the

average excess returns on quintile (4) portfolio (1) portfolio is 4.58, which is positive

and the average excess return on quintile at 1%. These findings of the

and significant

robustness of downside risk to size effect, is consistent with the findings in Table (4.3) of Fama and MacBeth's (1973) cross sectional regression

By examining the relationship between downside risk and average excess return within each size quintile portfolio, a number of points appear. A positive relationship between downside risk and average excess returns generally holds within the smallest three size portfolios (except for either the lowest or highest downside beta portfolios). The difference between the average excess return on the highest downside beta portfolio and the average excess returns on the lowest downside beta portfolio is positive and significant within the two smallest size portfolios but the difference for the third portfolio, although positive it is insignificant. Within the second largest portfolio, the

relationship between downside beta and average excess return is bell-shaped and the difference between the average excess return on the highest downside beta portfolio and average excess return on the lowest downside beta portfolio is negative but insignificant. Finally within the largest size portfolio, the corresponding difference is negative and significant and the relationship between downside beta and average excess return is

238

negative except for the lowest downside beta portfolio. Panel (A) shows qualitatively similar results for CAPM beta.

In summary, first, the relationship

between downside beta and stock returns holds

for small to middle size stocks but not for large stocks. This is consistent with Pedersen and Hwang (2007) who point out that downside beta, in the UK market, is significant for

high frequency returns on stocks of small size to middle size. In addition they point out that although downside beta is superior to CAPM beta its additional value possibly does not provide better ability for explaining the variation in the cross section of returns on

stocks. Again this seems to be supported by this chapter's finding when it examines the unconditional shown above. downside risk premium of Ang, Chen and Xing (2006) on UK market as Second, the negative relationship between downside beta or CAPM beta are more

and average excess return within

large stocks may suggest that investors

concerned with the movements of small stocks than large stocks in line with Perez-Quiros and Timmermann deteriorating (2000). They point out that the small stocks are more subject to

credit conditions in the market during recession and small stock returns and than for large stocks. So this could be the interpretation of

risk are more asymmetrical

why downside beta and indeed CAPM beta works for small but not large stocks. Finally, downside beta and CAPM beta show similar performance within each size portfolios.

This is different from Pedersen and Hwang's (2007) findings. They point out that size is a crucial factor in deciding the suitability of CAPM relative to downside beta and an

asymmetrical

model, for high frequency returns. However, the results, presented in this

sub-section, suggest that although size is important in deciding whether the risk factors

239

studied here are suitable for explaining the cross section of returns, this applies to both CAPM beta and downside beta against a potentially other asset pricing model.

4.5.2.2 Downside Beta and Book-to-Market

Table (4.5) presents the average yearly excess return on the 25 portfolios resulting from, first sorting stocks' on their book-to-market value and then on stock's CAPM beta (Panel A) or downside beta (Panel B). In addition it presents the average excess return on the five beta portfolios that control for book-to-market value in the last row of each Panel. Panel (B) shows the relationship between downside beta and average excess return is robust to book-to-market value. As before, there is a positive relationship between downside beta and average excess return from the lowest downside beta portfolio to the fourth downside beta portfolio and the relationship breaks for the riskiest downside beta stocks. The difference between the average excess return on the highest downside beta portfolio and the average excess return on the lowest downside beta portfolio is positive and significant at 5%. These findings are consistent with Table (4.3).

Examining downside risk - return relationship within each book-to-market portfolio reveals that downside risk is significant within middle to high book-to-market value

stocks; the highest book-to-market portfolios. The difference between the average excess return on the highest downside beta portfolio and the average excess return on the lowest downside beta portfolio is positive and significant for the highest two book-to-market portfolios, while positive but insignificant for the middle book-to-market portfolio. In

240

for breaks downside between the risk and return positive relationship although addition, the lowest downside beta portfolio within the highest book-to-market portfolio and also breaks for the highest downside beta portfolio within the middle book-to-market

portfolios, there is a strict positive relationship between downside beta and average excess returns on the five portfolios within the second highest book-to-market portfolio. Within growth stocks, there is no obvious relationship between downside beta and average returns, this suggest that returns on these stocks may be driven by other risk factors.

Panel (A) of Table (4.5) presents the findings for CAPM beta. By examining the downside beta portfolios that control for book-to-market (last row in Panel B) and the corresponding portfolios for CAPM beta (last row in Panel A), it is obvious that downside beta produces, on average, larger spread in portfolios' average excess returns. In addition, the difference between the average excess return on the highest CAPM beta portfolio and the average excess return on the lowest CAPM beta portfolio is positive but insignificant for the CAPM while as shown above, it is significant for downside beta portfolios. In addition downside beta does a better job in maintaining the positive riskfindings These book-to-market lowest the portfolios. middle and return within second beta. CAPM indicate downside beta is better than that a measure of risk may

In summary, although the relationship between downside risk and returns is robust to book-to-market value, within book-to-market categories, the results indicate that downside beta is not a priced factor for all the stocks. Downside beta is a driving risk

241

factor for return on value stocks but not growth stocks, and still there is a problem of 37 highest lowest pricing the or risk stocks even within value stocks.

4.5.3

Downside Risk and Economic Conditions

The third

hypothesis

of this chapter is tested in this section. This hypothesis risk premium of Ang Chen and Xing's (2006) is varying

examines whether downside

over the state of the economy. To examine this hypothesis, the full sample period is split into recession and expansion times as described under the methodology after splitting section. Then

the sample period into recession and expansion, Ang, Chen and Xing's reapplied on each state of the economy; i. e. on the

(2006) model and methodology

recession period and the expansion periods instead of just over the full sample period.

Panel (C) of Table (4.6) presents the results of forming downside beta portfolios during expansion periods. It shows a strict positive relationship between downside risk and portfolio's average excess returns from the lowest risk stocks (lowest downside beta portfolio) to the riskiest stocks (highest downside beta portfolio). The difference between the average excess return on the highest downside beta portfolio and the average excess 5%. This is is lowest downside beta the at also and significant positive portfolio on return in Panel (A) of (1973) Fama MacBeth's the regression cross-sectional and consistent with Table (4.7). It shows downside risk has a significant positive price of risk of 6.0% in the is during UK the robust to stock's size expansion, which stock returns of cross-section
37 Ang Chen and Xing (2006) report that they find that downside beta is able to explain the cross sectional 25 book-to-market (1993) Fama French's portfolios. size and the and returns on

242

and book-to-market characteristics. Upside beta is not significantly priced when it exists in the regression with downside beta, however, when size and book-to-market variables are included in the regression, it becomes significant but again with positive and not negative risk premium. Nevertheless, as for the unconditional downside and upside risk downside is in than risk premium during premia, upside risk premium smaller magnitude expansion.

Panel (D) of Table (4.6) presents the results for downside beta portfolios during recession period. There is a bell-shaped relationship between downside risk and the average excess return, and the difference between the average excess return on the highest downside beta portfolio and the average excess return on the lowest downside beta portfolio is positive but insignificant. In addition Fama and MacBeth's (1973) cross sectional regression results, which are reported in Panel (B) of Table (4.7), show that neither downside beta nor upside beta are priced in the UK cross sectional returns during recession period when no other variables exist in the regression. However when size and book-to-market exist in the regression, both downside risk and upside risk premia became formation but These the results, combined portfolio significant results with negative. indicate that there is no clear and robust relationship between downside risk and stock's excess returns during recession.

Table (4.6) also presents the results for CAPM beta during expansion (Panel A) and between fact despite B). During the (Panel that the relationship expansions, recession CAPM beta and average excess returns on portfolios is positive and increases over the

243

first four portfolios,

it breaks for the riskiest stocks (highest CAPM beta portfolio).

This

is similar to the relationship over the full sample period. However, during the recession,

similar to downside beta results, there is no clear relationship between CAPM beta and portfolios' average excess returns. The results of Fama and MacBeth's (1973) cross sectional regression for CAPM beta during expansion and recession are reported in Table (4.7). These results support the findings based on portfolio formation in Panel (A) of Table (4.6) for expansion. But for the recession, it shows a negative risk premium for the market portfolio which is robust to controlling for size and book-to-market value effects. Figure (4.3) depicts the risk - return relationships visually.

From the above analysis, a number of important insights arise; first downside risk
premium is positive and significant during expansions and larger in magnitude than

downside risk premium during recession period. However, downside risk premium during recession is negative and becomes significant only when other stock's

characteristics exist in the regression. Furthermore, downside risk - return relationship is almost linear during expansion, as shown in Figure (4.3) while this relationship does not hold during recession. These findings are opposed to Fama and French (1989) who report
that stock returns are higher during bad times than good times of the economy. Also contradict Post and Vliet (2005) who report that they find the outstanding performance of downside risk over CAPM beta in the US occurs, in particular, during bad economic

states which are the times of higher return and risk. They report that they find that bad economic times have higher return and risk than good times.

244

Furthermore, Perez-Quiros and Timmermann (2000) point out that returns on stocks are cyclical as a result of changes in both their risks and the price of risk over the business cycle, which are higher during recession and also small stocks returns' premium over large stocks increase in recession.

Second, it seems that using unconditional downside risk premium estimated over the full sample period causes the break of downside risk - return relationship for the riskiest stocks. However, when downside risk is estimated over expansion and recession separately; i. e. it is allowed to vary over the business cycle, downside beta outperforms CAPM beta in explaining the cross-section of the UK stock returns and the improvement comes from pricing the riskiest stocks.

In summary, allowing for downside risk premium of Ang, Chen and Xing's (2006) to vary with business cycle conditions, is important for uncovering the risk-return relationship in the UK market during expansion times. However, during recession, neither downside beta nor CAPM beta, is priced based on portfolio formation, although
they are negatively priced based on Fama and MacBeth's (1973) cross sectional

regression. One possible reason for the lack of a stable significant

role for downside risk

during recession in the UK is the short recession period in the chapter's sample. In fact Post and Vliet downside risk. (2005) point out that the bear market in their sample is important for

245

4.5.4

Downside Risk and Industry

This section tests the fourth hypothesis, which examines whether downside beta is a priced measure risk of for stocks within some industries but not within every industry. Ang, Chen and Xing (2006) point out that they find downside risk of utilities industry is lower compared with other industries whereas there is little pattern among the industries. They point out further that the finding for the utilities industry is in agreement with this industry being defensive in down markets.

To examine the above hypothesis, Ang, Chen and Xing's (2006) model and their methodology of portfolio formation and Fama and MacBeth's (1973) cross sectional regression are conducted within each industry instead of using all the stocks in the sample as before. Table (4.8) presents the results of sorting stocks into five downside beta portfolios, and separately into five CAPM beta portfolios within each industry. The findings indicate that downside beta seems to have explanatory power for the average excess returns on Travel and Leisure, Personal and Households Goods, Technology and Retail industries. Within each of these industries, there is a positive relationship between downside risk and average excess return from the lowest downside beta portfolio to the fourth (second highest) downside beta portfolios. However, for Retail industry, the relationship is strictly positive over the five downside beta portfolios. The difference between the average excess return on the highest downside beta portfolio and the average for lowest downside beta is the and significant all positive portfolio excess return on those industries except for the Technology industry where the difference is positive but insignificant. Although CAPM beta performs similar to downside beta for Travel and

246

Leisure and Personal and Household Goods, for the Retail industry, it has a negative relationship with the average excess returns, and the difference in average excess returns on the highest CAPM beta portfolio and the average excess return on the lowest CAPM beta portfolio is negative but insignificant.

On the other hand, CAPM cross section relationship of returns

beta seems to have better explanatory within Media industry, although

power for the the positive for the

on stocks

between CAPM

beta and average excess returns is not maintained CAPM beta seems to be slightly

riskiest stocks. Furthermore,

better for stocks within with the

Industrial Goods and Services, as downside beta has a bell-shaped relationship average excess returns while CAPM beta has a positive relationship

with the average

excess returns for all stocks except for the highest beta stocks. However, the difference between the excess returns on the highest beta portfolio the lowest beta portfolio Within Healthcare is positive and significant CAPM and the average excess return on

for both downside and CAPM betas. relationship with average power

industry,

beta shows a positive

excess return. Finally, neither downside beta nor CAPM beta has an explanatory for returns on stocks within Construction Chemicals, and Materials, Other Industries38 which include Automobiles Telecommunications, Basic Resources, Oil

and Parts, and Gas,

Food and Beverages and Utilities.

This is not surprising,

as these are not

homogenous industries.

38The reason for combining all these industries in one category "Other Industries" is that the number of stocks in each industry is not enough to form portfolios.

247

To shed more light on the role of downside beta in explaining returns on the stocks for industry, is Fama MacBeth's (1973) regressions run each and sectional within cross the one year excess returns on individual stocks on their CAPM beta and then on their downside beta and upside beta (Ang Chen and Xing's (2006) model) just as before for the full sample. However, the results should be interpreted with caution as the number of stocks for some industries is small. Table (4.9) presents the results. The results show that downside beta has a positive and significant price of risk in the following industries; Automobiles and Parts (8%), Travel and Leisure (12%), Personal Households and Goods (5%), Retail industry (9%), Food and Beverages (6%) and marginally insignificant at the 10% level in Construction and Material (4%). The portfolio formation results in Table (4.8) show that downside risk and average returns have no relationship within the Healthcare industry. However, surprisingly, downside beta, upside beta and CAPM beta, all are have significant risk premiums in the Fama and MacBeth's (1973) cross sectional regressions for the stocks within this industry, with the signs of these risk premia being consistent with Ang, Chen and Xing's (2006) model. Furthermore, within the utilities industry, downside risk has a negative risk premium, although insignificant. This is could be seen as consistent with Ang Chen and Xing (2006). They point out that they find, in unreported findings, utilities are less exposed to downside risk.

In addition, significant

the results of the chapter show that CAPM the following industries;

beta has a positive

and

price of risk within

Industrial

Goods and Services

(5.3%), Media (8%), Travel and Leisure (16%), Personal and Households Goods (12%),

248

Technology (13%), Healthcare (15%). For the Basic Resources, CAPM beta has a negative and significant risk premium (-9%).

Even though both CAPM beta and downside beta have signifincat risk premiums in the cross sectional regressions within Travel and Leisure industry, downside beta performs better in the portfolio formation. Downside beta has a strict positive relationship with the average excess returns over the five downside beta portfolios while CAPM beta does not maintain its positive relationship with average excess returns for the highest beta portfolio.

A general conclusion that can be drawn for the above results is that downside beta seems to a better measure of risk for stocks in some industries but not for all industries which is line with Olmo's (2007) findings. He reports that stocks with positive covariation (negative) with the down market have higher (lower) returns than estimated This by down CAPM, the chapter, not affected market. under while other stocks are though, studies the relationship at individual stock level and it does not focus on the focuses beta. downside It CAPM that on magnitude of returns under compared with using whether downside risk as an appropriate measure of risk is industry-dependent.

These findings have important implications for the participants in the London Stock Exchange. First, the choice between CAPM or downside beta as a risk factor - possibly firm. Pedersen Hwang factors depends industry the the and of on along with other risk (2007) also call for paying attention to stock category when choosing the appropriate

249

pricing model, although their focus is on the size category combined with the return
frequency. Second it seems there are missing factors that influence stock returns which

need to be considered in order to successfully describe the behavior of all stocks' prices. This is consistent with the first chapter of this thesis.

4.5.5

Is Past Downside Beta a Good Proxy for Future Downside Beta?

The results presented so far are for downside risk's contemporaneous relationship with the average excess returns on stocks. Ang, Chen and Xing (2006) point out that to practicalize this contemporaneous risk-return relationship, downside beta need to be
predicted using previous information. This section presents the results of applying Ang,

Chen and Xing's (2006) methods for examining the relationship between downside risk and future stocks returns on the UK market. Table (4.10) presents the results. It shows the average excess returns on portfolios sorted on their past downside beta and separately on their past CAPM beta. Panel (A) presents the average excess returns on these portfolio over the next month and Panel (B) presents the average excess returns calculated over the next 12-month, both are on yearly basis. Unlike the contemporaneous case, there is a negative relationship between downside risk and portfolio's future excess returns. And the difference between the average excess return on the highest downside beta and the average excess return on the lowest downside beta is negative and significant. Similar results are found to CAPM beta portfolios.

250

The negative downside risk - return relationship can be consistent with the positive contemporaneous risk - returns found earlier if past downside beta is negatively related (1973)

to future downside beta. Ang, Chen and Xing (2006) use Fama and MacBeth's cross sectional regressions to predict future relative downside predictor variables

beta using a number of

each once a time and then all in one regression. The results of

applying their procedure for a number of predictors to predict future UK downside beta are presented in Panel (A) (coefficient of Table (4.11). The slope of Regression (1) is 0.30

on past downside beta) which is positive and significant,

therefore it rejects

the potential explanation that the negative relationship

between past downside beta and

average excess returns is due to the negative relationship between past and future downside betas. This has an important warning for the investors, they cannot use past downside beta as a proxy for future downside beta. In fact Ang, Chen and Xing (2006) report that they find that future downside beta cannot be predicted only by past downside beta. However, they report that they find there is a positive relationship between past downside beta and future returns except for the highly volatile stocks. So the negative relationship between past downside beta and future returns found here in the UK market, remains a puzzle. Similar results are found for CAPM beta in Panel (B) of Table (4.11).

Regression (2) in Panel (A) of Table (4.11) shows that past book-to-market positively

is

related to future downside beta. Panel (B) of Table (4.5) shows that the highest within the highest book-tovalue and future for the US relative

returns is achieved by the highest downside beta portfolio market portfolio. The positive relationship

between book-to-market (2006) findings

downside beta is similar to Ang, Chen, and Xing's

251

downside beta. They point out that their results show that the value stocks have higher relative downside risk than the growth stocks (note that in this chapter downside beta is predicted and not relative downside beta) when book-to-market value is the only predictor in the regression but the opposite occurs in a regression that includes all other in Panel (A) of Table (4.11) (6) However for Regression UK, together. the predictors confirms that the UK value stocks have higher downside risk than the growths stocks.

Regression (3) shows the past logarithm of stock's market capitalization (size) is firms larger have higher indicates future beta, downside that to positively related which downside risk than smaller stocks. However, Panel (B) of Table (4.4) shows that the highest return is achieved by the highest downside beta portfolio within the smallest size portfolio and the smallest stocks have higher average returns than the largest stocks. These findings taken altogether contradict a positive risk-return relationship. On the other hand, Ang, Chen and Xing (2006) report that they find for the US, a negative relationship between past logarithm of market capitalization and future relative downside beta.

Regression (4) shows that past standard deviation

is positively

related to future

downside beta. Regression (5) shows that the past 12 month returns have no relationship with future downside beta. The findings of Regression (4) is consistent with Ang, Chen and Xing's (2006) results for US relative downside beta, for the past 12 month while

they report a positive relationship between past 12 month excess return and future relative downside beta. However, this positive relationship is consistent with the findings for the

UK in Regression (6), which shows that when all variables are included altogether, the

252

past 12 month return becomes significant and positively related to future downside beta. In addition all other variables remain significant. Similar results are found for predicting CAPM beta as Panel (B) of Table (4.11) shows.

4.5.6

Monthly

Return and Downside

Risk

Ang, Chen and Xing (2006) point out that their results for downside risk is robust

to using monthly returns over 60 month interval period. On the other hand, Pedersenand Hwang (2007) point out the normality could fit the UK monthly return and CAPM works for normal returns. Table (4.1) of this chapter rejects the normality assumption for monthly returns, however, it still requires examining whether using monthly returns over
the 60 month, as Pedersen and Hwang (2007) point out, would result in CAPM beta be a

good measure of risk compared with downside beta for the UK stock returns. Table (4.12) presents the results of portfolios sorted on their downside beta and separately on their CAPM beta, estimated using monthly returns. It shows that there is a bell-shaped relationship between downside beta and average excess returns, and the difference between the average excess return on the highest downside beta portfolio and the average excess return on the lowest downside beta portfolio is positive and significant. For CAPM beta, a similar bell-shaped relationship exists between CAPM beta and average excess returns, however, the difference between the average excess return on the highest CAPM beta portfolio and the average excess return on the lowest CAPM beta portfolio is negative but insignificant.

253

These results show the importance of return frequency in deciding the significance of downside risk as well as the CAPM beta in explaining the cross section of returns. Pedersen and Hwang (2007) point out that returns frequency is important in deciding whether to use downside risk and whether returns' normality holds. However, the results in this chapter show that both CAPM beta and downside beta fail to price the highest two CAPM beta portfolios and the highest two downside beta portfolios, respectively.

In summary, it is found that, using monthly returns to estimate downside beta, the relationship between downside beta and average excess return is much weaker than using daily return and similar results found for the CAPM beta. This implies that the UK investors have to use daily data and not monthly data to estimate the measure of risk when using downside beta (or CAPM beta) as a risk factor in their asset pricing models.

4.6 Conclusion

This chapter applies the model of downside risk and methodology of Ang, Chen and Xing (2006) to the UK market, and then it contributes to the literature by allowing their risk premium of downside beta to vary with the business cycle conditions.

A number of important findings have been presented. First it has been shown that when the risk premium of downside risk is estimated unconditionally over the full sample fails but it beta has downside to price a positive relationship stock's returns with period, the riskiest stocks and it does not improve significantly upon the CAPM beta which

254

shows similar results. Second, the relationship between downside risk and the returns on stocks is robust to a number of effects which are the size and the book-to-market value. Third, it has been found that both downside beta and CAPM betas explain the cross sectional returns on the UK small and value stocks but not on large and growth stocks and although both measures show similar performances, downside beta seems to be a slightly better measure of risk for some stocks. In addition, the findings suggest that downside beta is a better measure of risk for some industries. Fourth, when the risk premium is allowed to vary over the business cycle's expansion and recession periods, a strictly positive relationship between downside risk and stocks returns has been found during expansion. However during recession times, it has been found no robust relationship between downside risk and stock returns, with a potential negative downside risk premium. Finally, the results show that in order for downside risk and market risk (CAPM beta) to be priced in the cross sectional returns of the UK stocks with the correct sign, they should be estimated using daily and not monthly data.

This chapter concludes that although the performances of downside beta and CAPM beta are largely similar when their corresponding risk premiums are estimated unconditionally, once the risk premium becomes time varying conditional on the economic conditions, it has been shown that downside beta is better in pricing the return on high risk stocks. In addition downside and CAPM betas have different performances in explaining the returns within industries, especially the Retail industry and Construction downside beta has industries industry. For Material these two the show while results and for industry is Retail and marginally significant which significant a positive risk premium

255

for Construction and Material, CAPM beta has negative reward, although it is not significant. Indeed among the different variables that are studied in this chapter, industry is the only factor that distinguishes between CAPM beta and downside beta.

The findings in this chapter also suggest that there may be other factors that drive returns on large and growth stocks that need to be considered. The negative relation between downside beta and future excess return remains a puzzle for this study.

Important implications have arisen from the results of this chapter for the investors who trade stocks on the London Stock Exchange. First, investors should estimate a risk premium for downside risk that is state-dependent,although a search for a better measure of risk during recession might be needed. Second, investors should use downside risk and CAPM beta as risk factors only for small and value stocks but not for large and growth stocks and they should use daily returns and not monthly returns to estimate the risk
measures.

256

Es
y .. NN

Q)
o_n
VM N
cC N

ON co 0

N O

0 0 0 m Lo C. im u 3 a,
m C O 12. 'c fd

cd

w N

" w, Tw
.aT
'-

E
=

cn

3
>

I'D

C-

0 0"

C 1c OO 'U

a.:

E 0

20
S

cm

XC OH

MNa

rlN`
bA
00

a)

3 >. CD c w

'^ o

00 00 6

MO
N

N GyN NU

3E .O OT cO

cr,
O\ 'm -M
NU7 y .

on

3
' y

_
O

.r

M Oi

N pp p

N M

G Co
ON

a
V C N C VF

Ev

rl

a)

T vUi

c"

to

"

>x
F

`a z

c,

-0 vi yyO .0 E+ E NX0. OU
dU

Table 4.2 Annual

Excess Returns on Risk Characteristic

Based Portfolios

Panel A: Portfolios based on CA PM CAPM-beta Portfolios Average Excess Return CAPM /3 Downside Beta Q Upside Beta 8+ Low 5.82% -0.03 0.11 -0.10 Panel B: Downside beta Portfolios Average Excess Return CAPM 8 Downside Beta Upside Beta Q Low 6.13% 0.10 -0.12 0.11 2 9.33% 0.18 0.35 0.09 3 10.43% 0.35 0.55 0.26 4 11.09% 0.57 0.79 0.49 High 9.85% 1.08 1.25 1.02 High - Low 4.02% (1.62) 1.11 1.14 1.12

Portfolios based on Downside Beta 8 2 9.16% 0.23 0.29 0.18 3 10.30% 0.37 0.54 0.30 4 10.96% 0.56 0.84 0.45 High 10.04% 0.90 1.50 0.72 High - Low 3.91% (1.69)* 0.80 1.62 0.62

Panel C:: Portfolios based on Relative Downside Beta Relative downside beta 3 Low 2 4 Portfolios Average Excess Return CAPM 6 Downside Beta Q 7.74% 0.51 0.15 0.59 9.22% 0.39 0.39 0.37 10.03% 0.37 0.53 0.30 10.15% 0.40 0.73 0.27

High 10.03% 0.49 1.26 0.23

High - Low 2.28% (I. 98)** -0.02 1.11 -0.36

Upside Beta )6+

258

Table (4.2) continued

Panel D:: Portfolios based on Upside Beta Upside beta portfolios Average Excess Return CAPM 8 Downside Beta Upside Beta + Low 9.00% 0.14 0.43 234 8.88% 0.22 0.42 0.05 10.14% 0.34 0.52 0.27 10.49% 0.53 0.69 0.56 High 8.69% 0.93 1.00 1.24 High - Low -0.31% (-0.21) 0.79 0.57 1.61

-0.37 This Table presents the portfolios constructed from sorting all the stocks in the chapter's sample, at the beginning of every month, into five quintile portfolios based on their CAPM beta (Panel A), downside beta (Panel B), relative downside beta (Panel C) and upside beta (Panel D). The sample period is July 1981 to December 2005. The measures of risk (the betas) are computed using daily return over the subsequent oneyear period and the equally weighted annual excess returns on the portfolios (Average excess Return) are calculated over the same one-year period. CAPM , Downside Beta Q and Upside Beta + are the averages of the betas of the individual stocks in each portfolio, all computed over the same one-year period. Newey- West (1987) based t-statistics are reported in parenthesis. ***significant at 1%, ** at 5% and * at 10%

259

Table 4.3 Cross-Sectional Regressions

CAPM Intercept 18 Regression (1) Regression (2) Regression (3) Regression (4) 0.109 (3.97)*** 0.234 (4.47)*** 0.111 (4.09)*** 0.236 (4.68)*** 0.074 (1.87)* 0.131 (2.31) 0.043 (2.53)** 0.049 (2.25)** 0.004 (0.37) 0.030 (2.19)** -0.078 (-4.29)*** 0.001 (1.29) -0.086 (-3.85)*** 0.001 (1.09)

Downside
beta QQ

Upside
beta

Logarithm
market Capitalization (Size) Book-toMarket

This Table presents the coefficients from monthly Fama and MacBeth's (1973) cross sectional regressions. The sample period is July-1981 to December - 2005. The annual excess return on the individual stock is calculated over the same one year-period used to calculate its measures of risk, which are calculated using daily returns. The excess returns on individual stocks are regressed on their measures of risk and size and book-to-market variables. The logarithm of the market capitalization (size) and book-to-market are calculated at the beginning of every period. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%

260

Table 4.4 25 Size and CAPM or Downside Beta Portfolios

Panel A: CA PM Beta Beta Portfolios Size Portfolios Small 2 3 4 Large Average Beta Portfolios Size Portfolios Small 2 3 4 Large 8.69 1.50 4.55 6.42 8.34 5.90 7 83 . 5 63 . 8.69 8 33 . 9.02 7.90 11.05 10 31 . 9.61 9.56 8.63 9.83 14.64 9.70 10.74 8.75 5.23 9.81 24.97 11.85 7.31 2.97 0.39 9.50 16.29 (5.48)*** 10.35 (3.39)*** 2.75 (0.80) -3.45 (-0.95) -7.95 (-2.26)** 3.60 (1.32) High-Low 15 13 . (5.54)*** 10.69 (4.11)*** 2.07 (0.71) -3.89 (-1.19)*** Low 23 4 High High-Low

Panel B: Downside Beta Low 9.01 0.71 5.57 6.24 8.46 23 7.30 6 73 . 8.59 9 25 . 8.61 11.11 8.89 9.37 9.45 7.72 4 16.13 11.51 10.44 9.04 5.80 High 24.14 11.40 7.64 2.34 1.25

-7.22 ** (-2..11 )) 3.36 Average 6.00 9.31 9.35 8.09 10.58 (1.42) This Table presents results of the 25 portfolios that result from sorting all the stocks in the sample into five portfolios based on their size (market capitalization) and then within each size portfolio, stocks are sorted into five portfolios based on their CAPM betas (Panel A) or downside betas (Panel B). The sample period is July -- 1981 to December (equally 2005. The average weighted) annual excess returns, which are reported in the Table, and the betas are calculated using daily returns over the same subsequent one-year period. The market capitalizations are calculated at the beginning of the period to match those in Table (4.3). This formation of the 25 portfolios is carried out at the beginning of every month. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%. The last row of each Panel shows the average annual excess return on portfolios that control for size.

261

Table 4.5 25 Book-to-Market

Beta Portfolios Book-toMarket Portfolios Low B'o Market 2 3 4 High Book-toMarket Average Beta Portfolios Book-toMarket
Portfolios

and CAPM or Downside Beta Portfolios

Panel A: CA PM Beta

Low

23

High

High-Low

-0.76 4.98 6.54 7.85 12.97 6.32

5.40 9.06 9.77 9.48 12.27 9.20

2 81 . 8.30 10.58 10.70 15.50 9.58

3.64 5.43 10.47 12.03 17.68 9.85

-1.79 4.00 8.70 12.72 20,60 8.85

(-0.35) -0.97 (-0.42) 2.17 (0.94) 4.87 (2.29)** 7.63 (3.22)*** 2.53 (1.33)

Panel B: Downside Beta Low 23 4 High High-Low 2.50 82 () 0. 0.93 (0.40) 1.56 (0.66) 4.82 (2.58)** 8.21 (3.87)*** 3.60 (2.02)**

Low Book-toarkt 2 3 4 High Book-toMarket Average -2.97 3.56 7.71 7.60 12.76 5.73

4.31 7.27 8.68 9.84 12.74 8.57

3.81 8.74 9.99 10.94 15.03 9.70

4.36 7.66 10.55 12.08 17.74 10.48

-0.47 4.48 9.26 12 41 . 20.98 9.33

This Table presents the results of the 25 portfolios that result from sorting all the stocks in the sample into five portfolios based on their book-to-market value and then within each book-to-market portfolio stocks are sorted into five portfolios based on their CAPM betas (Panel A) or downside betas (Panel B). The sample period is July - 1981 to December - 2005. The average (equally weighted) annual excess returns, which are reported in the Table, and betas are calculated using daily returns over the same subsequent oneyear period. The book-to-market values are calculated at the beginning of the period to match those in Table (4.3). This formation of the 25 portfolios is carried out at the beginning of every month. NeweyWest (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%. The last row of each Panel shows the average annual excess return on portfolios that control for book-to-market.

262

Table 4.6 Annual Excess Return on Risk Characteristics Expansion and Recession

Based Portfolios

in

Panel A: Portfolios based on CA PM 8 Expansion CAPM-beta Portfolios Average excess Return CAPM Downside Beta 83 Upside Beta Low 8.87 -0.03 0.11 -0.11 Panel B: Downside beta Portfolios Average excess Return CAPM /i Downside Beta QUpside Beta )6+ Low 2 11.62 0.18 0.35 0.08 3 13.02 0.34 0.55 0.24 4 13.68 0.57 0.79 0.46 High 13.19 1.07 1.25 0.99 High - Low 4.33 (2.33)** 1.10 1.14 1.10

Portfolios based on CAPM Recession 2 3 4 High High - Low 1.87 (0.77) 1.15 1.15 1.26

-17.38 0.01 0.13 0.00

-8.17 0.21 0.38 0.20

-9.24 0.37 0.58 0.40

-8.66 0.61 0.82 0.66 Expansion j64 13.37 0.55 0.84 0.43

-15.51 1.15 1.28 1.26

Panel C:: Portfolios based on Downside beta Relative downside beta Low 2 3 Portfolios Average excess Return CAPM Downside Beta Upside Beta + 9.59 0.10 -0.12 0.09 11.31 0.22 0.29 0.16 12.35 0.37 0.54 0.28

High 13.92 0.90 1.50 0.70

High - Low 4.33 (2.55)** 0.80 1.62 0.61

263

Table (4.6) Continued

Panel D:: Portfolios based on Downside beta - Recession Relative downside beta Low 234 High Portfolios Average excess Return CAPM 8 Downside Beta
Upside Beta 8+

High - Low 0.70% (0.24) 0.82 1.61

0.65

-20.02 0.13 -0.07

0.25

-7.16 0.26 0.32

0.32

-5.25 0.40 0.56

0.44

-7.29 0.60 0.86

0.61

-19.33 0.96 1.54

0.90

This Table presents results of splitting the sample period into expansion and recession period based on the dates of the business cycle from the Economic Cycle Research Institute. Then for each economic state separately, portfolios are constructed from sorting all the stocks in the chapter's sample, at the beginning of every month, into five quintile portfolios based on their CAPM beta (Panel A for expansion and Panel B for recession) and downside beta (Panel C for expansion and Panel D for recession). The sample period is July 1981 December 2005. betas) The (the to are computed using daily returns over the measures of risk subsequent one-year period and the equally weighted annual excess returns on the portfolios (Average excess Return) are calculated over the same one-year period. CAPM Q, Downside Beta Q and Upside

Beta Q+ are the averages of the betas of the individual stocks in each portfolio, all computed over the same one-year period. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%.

264

Table 4.7 Cross -Sectional Regression in Expansion and Recession

Panel (A): Expansion Intercept Regression (1) Regression (2) Regression (3) Regression (4) 0.134 (4.73)*** 0.284 (5.21)*** 0.135 (4.72)*** 0.285 (5.42)*** CAPM

Downside
beta 8,

Upside
beta 18+

Logarithm
market Capitalization (Size) Book-toMarket

0.087 (2.00)** 0.164 (2.63)*** 0.051 (2.67)*** 0.060 (2.5)** 0.005 (0.42) 0.040 (2.63)**

-0.108 (-4.62)***

0.001 (0.92)

-0.098 (-5.15)***

0.001 (1.15)

Panel (B): Recession Intercept Regression (1) Regression (2) Regression

(3)

CAPM
!j

Downside beta

Upside beta
Q+

Logarithm
market
Capitalization (Size)

Book-toMarket

-0.083 (-1.75)* -0.145 (-2.26)**

(-1.80)*

-0.024 (-1.77)* -0.124 (-6.36)***

(-1.21)

0.082 (6.19)*** -0.017

(-0.70)

0.001 (4.15)***

-0.075

-0.004

Regression 0.066 0.001 -0.134 -0.033 -0.042 (4) (-2.26)** (-1.79)* (-9.84)*** (4.91)*** (3.94)*** This Table presents the coefficients from monthly Fama and MacBeth's (1973) cross sectional regressions for expansion period (Panel A) and recession period (Panel B). The sample period is July-1981 to December - 2005. The sample period is split into expansion and recession based on the dates of the business cycle from the Economic Cycle Research Institute. Then for each economic state separately, the annual excess return on the individual stock is calculated over the same one year-period used to calculate the measures of risk, which are calculated using daily returns. The excess returns on individual stocks are regressed on their measures of risk and size and book-to-market variables. The logarithm of the market capitalization (size) and book-to-market are calculated at the beginning of every period. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%.

265

Table 4.8 Annual Excess Return on Risk Characteristics Based Portfolios within each Industry
Panel A: Ind. Goods & Services Low CAPM-beta Downside -beta 4.80 4.94 Low CAPM-beta Downside -beta 4.89 5.76 234 8.93 9.02 9.12 9.08 9.46 8.81 4 11.79 11.47 High 8.55 8.59 High 8.61 4.36 High - Low 3.75 (2.24)** 3.65 (1.99)** High - Low 3.72 (1.79)* -1.40 (-0.49) High - Low 7.88 (2.6)*** 8.73 (2.98)*** High - Low 7.38 (3.34)*** 6.03 (2.73)*** High - Low 6.23 (1.45) 5.91 (1.46) High - Low 9.16 (2.01)** 2.41 41

Panel B: Media 23 8.91 10.86 11.22 12.56

Panel C. Travel & Leisure Low CAPM-beta Downside -beta 1.15 1.30 Low CAPM-beta Downside -beta 1.65 2 79 . Low CAPM-beta Downside -beta 1.09 1.97 Low CAPM-beta -beta 5 08 . 5.24 23 3 95 . 4.87 7.80 7.36 4 9 15 . 7.84 High 9.03 10.03 High 9.03 8 82 . High 7.32 7 88 . High 14.24 7.65

Panel D: Pers & Househld Goods 23 4 6.04 8.61 11 15 . 10 60 . 4 12.96 11.43 11.32 13 34 . 4 7 72 . 10.57

8 03 6 46 . . Panel E. Technology 23 9.14 8.84

Panel F. Healthcare 23 7 20 . 9.88 6 57 . 8.11

266

Table (4.8) continued Panel G: Retail

Low CAPM-beta Downside -beta 11.50 9.59 Low 234 13.65 11.32 13.42 13.04 12.11 12.98 High 10.71 14.48 High High - Low -0.79 (-0.25) 4.88 (1.72)*

High - Low 0.66 CAPM-beta 11.08 8.25 7.59 9.49 11.20 (0.33) 1.00 10.13 Downside -beta 10.28 9.02 8.02 10.08 (0.48) This Table presents the portfolios constructed from sorting the stocks within each industry, at the beginning of every month, into five quintile portfolios based on their CAPM beta (first row of each Panel) and separately their downside beta (second row in each Panel). The sample period is July - 1981 to December 2005. The measures of risk (the betas) are computed using daily returns over the subsequent one-year period and the equally weighted annual excess returns on the portfolios, which are reported in the Table, are calculated over the same one-year period. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%. * Automobiles & Parts, Construct. & Material, Telecommunications, Basic Resources, Oil & Gas, Chemicals, Food & Beverage, Utilities Number of stock in each industry is; Ind. Goods & Services (569), Media (163), Automobiles & Parts (19), Travel & Leisure (168), Pers & Househld Goods (201), Technology (206), Construct. & Material (79), Healthcare (105), Telecommunications (25), Basic Resources (66), Oil & Gas (64), Chemicals (51), Retail (187), Food & Beverage (82), Utilities (35)

Panel H. Others * 234

267

Table 4.9 Cross -Sectional Regression within each Industry

Panel (A): Ind. Goods & Services

CAPM (1) Regression

Regression (2)

Q Downside beta

Upside beta

8+

0.053
(1.95)* 0.024 (1.23) Panel (B): Media 0.022 (1.85)*

CAPM (1) Regression

Regression (2)

Downside beta

Upside beta

0.08
(2.30)** 0.015 (0.76) Panel (C): Automobiles & Paris 0.016 (0.97)

CAPM
Regression (1) Regression (2) 0.07 (1.52)

Downside beta

Q+ Up side beta

0.08 (2.70)*** Panel (D): Travel & Leisure

-0.002 (-0.04)

R CAPM
Regression (I) Regression (2) 0.16 (2.55)**

Q Downside beta

Up side beta )6+

0.12 (2.78)*** Panel (E): Pers & Househld Goods

-0.017 (-0.54)

CAPM )6
Regression (1) Regression (2) 0.1 (2.66)***

Downside beta

U pside beta J6+

0.017 (0.88)

0.05 (2.37)** Panel (F): Technology

Q CAPM
Regression (1) Regression (2) 0.13 (1.63)

Downside beta
0.05 (1.27)

Upside beta )6+

0.019 (0.40)

Panel (G): Construct.

& Material

CAPM
Regression (1) Regression (2) -0.01 (-0.29)

Downside beta
0.04 (1.60)

Q+ U pside beta
-0.02 (-1.12)

268

Table (4.9) Continued Panel (H): Healthcare

CAPM Regression (1) Regression (2) 0.15 (2.47)* 0.12 (1.94)* Panel (J): Telecommunications -0.10 (-1.65)* Q Downside beta Q Upside beta d6+

CAPM Regression (1)

Regression (2)

Downside beta

Upside beta J6+

0.001
(0.01) -0.04 (-0.47)
Panel (1): Basic Resources

0.25 (1.05)

CAPM 18
Regression (1) Regression (2) -0.09 (-3.09)***

Downside beta

U pside beta

-0.02 (-0.78) Panel (K): Oil & Gas

-0.028 (-1.12)

CAPM
Regression (1) Regression (2) 0.05 (0.95)

Downside beta

U pside beta
0.11 (2.68)***

-0.03 (-0.97) Panel (L): Chemicals

Q CAPM
Regression (1) Regression (2) -0.09 (-1.30)

Downside beta

8+ U pside beta

-0.08 (-1.55) Panel (M): Retail

-0.12 (-1.27)

CAPM
Regression (1) Regression (2)
_n(17

Q Downside beta

+ Up side beta

(-1.32) 0.09 (1.77)*

Panel (N): Food & Beverage

-0.04 (-1.32)

Q CAPM
Regression (1) Regression (2) -0.001 (-OA3)

Downside beta
0.06 (2.02)* *

8+ Upside beta

-0.027 (-0.93)

269

Table (4.9) Continued Panel (0): Utilities

Q CAPM
Regression (I) Regression (2) 0.02 (0.16)

Q Downside beta

8+ Upside beta

-0.20

-0.02

(-0.90) (-0.51) This Table presents the coefficients from monthly Fama and MacBeth (1973) cross sectional regressions within each industry. The sample period is July-1981 to December - 2005. Within each industry, the annual excess return on the individual stock is calculated over the same one year-period used to calculate the measures of risk, which are calculated using daily returns. The excess returns on individual stocks are regressed on their measures of risk. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%.

270

Table 4.10 Future Annual Excess Return on Past Risk Characteristics Based Portfolios
Panel A. Low CAPM-beta Downside -beta 12.23 10.87 Low CAPM-beta Downside -beta 13 49 . 12.38 2 10.64 10 21 . 2 11 16 . 3 10.60 9.43
Panel B:

4 8.87 9.44 4 8.99

High 3.22 5.81 High 4.01

High - Low -9.01 (-2.55)** -5.06 (-1.72)* High - Low -9.48 (-4.59)* **

3 10.68

-6.78 9.85 5.60 9.83 10 96 (-3.93)*** . This Table presents the portfolios constructed from sorting all the stocks in the chapter's sample, at the beginning of every month, into five quintile portfolios based on their past CAPM beta (first row of each Panel), and past downside beta (second row of each Panel). The sample period is July - 1981 to December 2005. The measures of risk (the betas) are computed using daily returns over the past one-year period and the equally weighted annual excess returns on the portfolios, which are reported in the Table, are calculated over the next one month (Panel A) and next one year (Panel B). Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%.

271

Table 4.11 Does Past Information

Predict Downside Beta? Downside Beta

Reg. (1) Downside Beta BMV SIZE STDEV RET R2 9.42% 0.30 (14.77)***

Reg. (2)

Reg. (3)

Reg. (4)

Reg. (5)

Reg. (6) 0.15 (9.95)***

0.01 (2.25)** 0.17 (11.95)*** 7.77 (12.83)*** -0.001 (-0.04) 0.38% 8.44% 4.37% 1.68%

0.003 (1.95)* 0.19 (15.09)*** 12.51 (10.69)*** 0.07 (3.2)*** 21.42%

CAPM Beta
CAPM
Beta BMV SIZE STDEV RET R2 35.88% 0.63% 25.03% 3.98%

0.60
(24.28)*** 0.003 (2.6)*** 0.23 (16.21)*** 4.72 (9.77)*** 0.013 (0.66) 2.01%

0.39
(15.58)*** 0.002 (2.19)** 0.17 (15.31)*** 6.25 (8.41)*** 0.043 (2.91)*** 47.98%

This Table presents the coefficients of monthly Fama and MacBeth's (1973) cross sectional regression for predicting downside beta (Panel A) and predicting CAPM beta (Panel B). Downside beta (CAPM beta) is regressed on past downside beta (past CAPM beta), past book-to-market value (MBV), past logarithm of market capitalization (SIZE), past standard deviation (STDEV) and past one year excess returns (RET). Dummies for industries are included in every regression. Downside beta and CAPM beta to be predicted are calculated over the subsequent one-year using daily returns. Past downside and CAPM betas and past standard deviation are calculated using daily returns over the past one-year while past book-to-market, past size and past one year excess returns are calculated at the beginning of the period. RZ is the adjusted RZ from the regression. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%. The sample period is July 1981 to December - 2005.

272

Table 4.12 Annual Excess Return on Risk Characteristics Monthly Frequency

Based Portfolios

Low CAPM-beta 4.56

234 6.09 7.06 5.53

High 3.14

High - Low -1.42 (-1.26) 1.81 (2.81)***

Downside -beta

2.81

6.12

6.82

6.29

4.62

This Table presents the portfolios constructed from sorting all the stocks in the chapter's sample, at the beginning of every month, into five quintile portfolios based on their CAPM beta (first row) and downside beta (second row). The sample period is July 2005. The (the betas) 1981 December to measures of risk are computed using monthly returns over the subsequent 60 month- period and the equally weighted annual excess returns on the portfolios, which are reported in the table, are calculated over the same period. Newey- West (1987) based t-statistics are reported in parenthesis, ***significant at 1%, ** at 5% and * at 10%

273

1 a H

-uni o-uni -uni

0 0 N
y N U

^J

r,

p-uni p- uni 66- unii uni -uni 96-uni 6-uni 6-uni 6-uni -uni

o 0 u oo j .-u i
>, ^' O
N

u ) C

v7
-c ;r

Gl.

UY Un

UU

_ yU "}

-uni
g-uni 99-unr
c c _C O C. J

lIE

r r

UG

73

C_ y of C :a U -r
:J L bL U to

------

a y G on 5
nU

un oN. IONn ui a6uey3

;a"

: c " v

Figure 4.2 Relationship .,. no.

10.00%

Between Risk and Returns

8.00% 6.0 U'%

- CAPM

4.00% 2.00% 0.00%

- s-

Downside RelDwn Upside

-0.5

0.5

Beta

1.5

This Figure depicts the relationship between CAPI I beta, Downside beta, Relative Downside (RelDwn) Beta, or Upside beta and the equally weighed annual average excess returns on five portfolios sorted by their ('APM beta, downside beta, relative downside beta and upside beta, respectively. The y-axis is the porttoolio's average excess return and the x-axis is the portfolio's beta.

Figure 4.3 Relationship

Between Risk and Returns Over the Business Cycle

20 15
5 0 " ;-ADow n-Good Dow n- Bad CAP%Good CAPKBad

-0.5 >

-5 -10
-15

0-

Dr5r-_

.___

1.5

-25 Beta
This Figure depicts the relationship between CAP\l beta (CAPM-Good) and Downside beta (DownGood) during expansions and the average annual returns on the five portfolios sorted by their CAPA1 beta and downside beta respectively. Also it depicts the relation between CAPM beta (CAPM4-Bad) and Downside beta (Down-Bad) and average excess return during recession. The y-axis is the portfolio's average excess return and x-axis is the portfolio's beta.

275

Chapter 5 Conclusion

276

5.1

Conclusion

This thesis has studied the relationship between stock return and risk motivated by the recent developments in the area. It has examined the relationship in the light of a number of adjustments that have recently been re-examined by researchers.These include advancing from a static to an inter-temporal model (Merton (1973)); from pricing systematic risk to testing whether both systematic risk and idiosyncratic risk have significant effects on stock returns (e.g. Merton (1987)); and from CAPM beta risk to downside risk (e.g. Bawa and Lindenberg (1977) and Ang, Chen and Xing (2006)). The evidence from this thesis provides useful insights to add to other researchers' efforts toward understanding the relationship between stock rerun and risk.

5.2

Macroeconomic Factors and Fama and French Asset Pricing Model

In light of Campbell's (2000) and Cochrane's (2006) emphasis on the importance of macroeconomic variables to stock returns, the second chapter started by applying Petkova's (2006) study to UK. It examined the performance of the Fama and French's (1993) three-factor model on the UK Fama and French (1993) 25 size and book-tomarket portfolios. It found both the SMB (negatively) and HML (positively) are robustly priced in the cross section of returns on stocks. Then it applied Petkova's (2006) model that includes excess market return and innovations to dividend yield, term spread, default in includes her T-Bill that addition to the above model as well as spread and one-month five variables the HML and SMB. The findings showed that innovations to dividend yield innovation to that the term are results robust significantly priced, and and spread are

277

estimation technique and model estimation method. These results when taken as a whole are in line with Petkova's (2006) findings for US. She reports that innovations to term spread and one-month T-Bill are priced. In addition she reports that HML is positively priced while the SMB is not significant. However this chapter's results showed, unlike hers, the priced risk factors are not able to suppress the HML and SMB factors. Griffin (2002) points out that countries SMB and HML correlations that he finds are unlike those that would be found if they were driven by similar state variables.

Therefore, motivated by this lack of association between Fama and French's (1993)
SMB and HML factors and the above risk variables selected by Petkova (2006), this

chapter made use of the dynamic factor model and principal components method of Stock and Watson (2002a, b) to augment Petkova's (2006) model with innovations in factors

estimated from a large set of macro variables. This is also motivated by the findings of Mnch (2004,2006) and Ludvigson and Ng (2007). These studies applied this method over and above the variables

and reported that such estimated factors add information

generally used in asset pricing models literature. This augmenting step is important in light of the challenge that are put forward by Campbell (2000) and Cochrane (2006) that macro variables which influence stock return need to be unearthed and the call by

Cochrane (2006) among others, to link the SMB and HML to macroeconomic variables. Therefore, confining the analysis to a few macro variables will not provide a definitive answer to these challenges.

278

The results showed that the innovations to two estimated factors (the first and the sixth factors) are priced in the cross sectional returns on stocks. Mnch (2004) also reported that two estimated factors are priced in the cross section of return in the US. However, the final results showed that Petkova's (2006) model, that includes excess market returns and innovations to dividend yield, term spread, and the HML and the SMB [retaining only priced factors] augmented with innovations to the second estimated factor, is potentially the best choice to describe stock returns in the UK.

The results suggested that this second estimated factor, which is found to be associated with employment and term spread, may add information beyond those captured by innovations to dividend yield and term spread. The ability of the estimated factors from a large set of data to add additional information is consistent with Mnch (2004) who augments Campbell's (1996) model with two estimated factors, and
Ludvigson and Ng (2007) and Mnch (2006) who augment the conditional information

set with estimated factors. Nevertheless, the results showed, this second estimated factor does not eliminate the effect of the SMB and HML and therefore it remains unclear what

exactly these two factors do represent in the UK market.

Having employed a large set of macro variables reduces the possibility interpretation of Fama and French's (1993) HML and SMB

of finding an of the

in the context

macroeconomic variables in the UK. Despite the fact that Liew and Vassalou (2000) report that there is a link between the SMB and HML in the UK and GDP growth and

279

Kelly (2003) also reports that these two factors are connected to the real GDP growth in the UK. Furthermore, he reports that the HML is also linked to unexpected inflation.

5.3

Idiosyncratic

Risk and Time-Varying

Betas

The previous chapter examined whether a number of systematic risk factors were priced in the cross section of stock returns. However, there is recent evidence pointing the finger at idiosyncratic risk as a potential factor in capturing changes in stock prices (e.g. Ang, Hodrick, Xing, and Zhang (2006,2008), Spiegel and Wang (2005) and Fu (2007)) while others reject such findings (see Bali and Cakici (2008) and Huang et al (2006)). Therefore, the third chapter of this thesis examined whether idiosyncratic risk is priced in the cross section of UK stock returns. It examined idiosyncratic risk effect by following Ang, Hodrick, Xing, and Zhang (2006,2008) and Spiegel and Wang (2005), in order to establish whether it is a sustained effect or not robust as found by Bali and Cakici

(2008). This is important as there has been evidence of significant idiosyncratic risk effect in the UK documented by, for example, Guo and Savickas (2006), Angelidis and Tessaromatis (2008b), Ang, Hodrick, Xing, and Zhang (2008) and Fletcher (2007).

The results showed, consistent with Bali and Cakici's idiosyncratic insignificant risk effect is not robust. The results showed

(2008) conclusion, negative, positive

that or

effects in the cross section of UK stock returns for idiosyncratic

risk. In

detail, when idiosyncratic Ang, Hodrick, Xing,

following OLS based daily is with on returns risk examined the relationship between stock returns

and Zhang (2006,2008),

280

and idiosyncratic risk is negative and significant as reported by Ang, Hodrick, Xing, and Zhang (2006,2008). However these findings, unlike the previous studies are not robust to the testing procedure. Fama and MacBeh's (1973) cross sectional regression show a negative idiosyncratic effect while portfolios formed on their idiosyncratic risk show insignificant effect. Furthermore, the results suggested an association between

idiosyncratic risk and small stocks which is consistent with other studies such as Bali and Cakici (2008), Angelidis and Tassaromatis (2008b) for the UK and Diavatopoulos et al., (2007).

When lagged idiosyncratic following

risk is measured based on monthly returns with OLS risk could be

Spiegel and Wang (2005), the results suggested that idiosyncratic

positively priced in the cross section of returns only in the presence of other effects, also there is connection Furthermore, EAGARCH between small and value stocks and idiosyncratic idiosyncratic risk from monthly risk effect. returns using

this chapter calculated (1,1) following

Spiegel and Wang (2005). The results showed, in contrast to risk, that when idiosyncratic risk

the above findings for the OLS monthly idiosyncratic exists alone in the relationship

it is found to be significant

and negative in the cross (1,1) idiosyncratic risk

section of returns, while when other effects exist, EAGARCH effect vanishes. (2007) who

These findings are in contrast with Spiegel and Wang (2005) and Fu significant positive idiosyncratic risk effect when the latter

reported

estimated from EGARCH

models. All in all, these findings are line with the Baly and risk effect in the cross sectional return of

Cakici (2008) who conclude that idiosyncratic US stocks is not robust.

281

In light of these confusing findings of the importance of idiosyncratic UK stock return which

risk in the

are consistent with the documented confusing evidence, this between stock returns and idiosyncratic model and methodology of Avramov and

chapter proceeded to examine the relationship risk using the time-varying beta (conditional)

Chordia (2006) who point out that the advantage of their modelling of the variation in the stock betas is that it is explicit. Fletcher (2007) also uses conditional asset pricing model in studying idiosyncratic (2006) time-varying risk in the UK. The application of Avramov and Chordia's by Chen and

beta model and methodology

was also motivated

Keown (1981) and Malkiel

and Xu (2006) who point out that the residuals may include and Xu (2006) point out that Miller and Scholes (1972)

other effects. In fact, Malkiel

argue that the significant idiosyncratic risk could be due to errors in betas.

The results showed that once the time variation in betas is accounted for, no significant effect for idiosyncratic risk remains in the cross section of returns, regardless of idiosyncratic risk measure(monthly OLS or EGARCH) or the testing procedure (Fama and MacBeth (1973) or portfolios formation). However, this chapter did not examine the effect of the time-varying betas on idiosyncratic risk when the latter is calculated based on the daily returns. The reason for this is the daily book value does not change within a month's time period. Nevertheless, the results from portfolios formed basedon daily OLS idiosyncratic risk are less supportive to idiosyncratic risk effect which may support that idiosyncratic risk is not really priced in the cross section of returns.

282

5.4

Downside

Risk and Business

Cycle

Fama and French (2004) point out that theoretical shortcoming could be behind the CAPM failure. Furthermore, there is recent evidence of the significance of downside risk in the cross sectional returns. Therefore, and in light of these recent promising studies of downside beta such as those by Ang, Chen and Xing (2006) and Post and Vliet (2005), the fourth chapter sets the following objective: examining downside risk in the cross sectional returns on UK stocks following Ang, Chen and Xing (2006). This was also motivated by the evidence of Pedersenand Hwang (2007) and Olmo (2007) for downside risk in the UK market.

The results showed that downside risk is significant in explaining UK stock returns with a positive price of risk, a finding that is generally consistent with Ang, Chen and Xing (2006) and Post and Vliet (2005) for the US. However, the findings showed that downside risk has a problem pricing the riskiest stocks and did not improve significantly over CAPM beta. Pedersenand Hwang (2007) study downside risk in the UK and point out that based on their findings downside beta will not result in great improvement for asset pricing models. These findings of this chapter are not consistent with Ang, Chen and Xing (2006) for US market. They reported great explanatory power for downside risk in the cross section of returns. In addition, the results in this chapter showed that the return - risk (whether downside beta or CAPM beta) relationship is significant with the for of risk premium small size to middle size and value stocks but not sign only correct for large and growth stocks. The finding of an association between small and middle

283

is in line downside risk with Pedersen and Hwang (2007). and stocks

Pedersen and

Hwang (2007) report downside beta is essential for the small size to middle stocks in UK.

In the next stage, then the risk - return relationship was studied over the business cycle by dividing the study period into recession and expansion periods following Post and Vliet (2005) and then reapplying Ang, Chen and Xing's (2006) downside risk model and methodology over recession and over expansion separately. The recession and expansion periods are identified following Antoniou et al (2007) using the Economic Research Cycle Institute's dates of the business cycle. The results showed, during expansion, a strict positive relationship between downside risk and stock returns whilst CAPM beta still fails to price the riskiest stocks. On the other hand, during recession this chapter found no conclusive relationship between risk and return which could be attributed to the small recession period in this study. These findings support Post and Vliet (2005) who studied and emphasized the importance of time-variation in downside risk. But at the same time it supports the findings of Ang Chen and Xing (2006) that their measure of downside risk is robust and not weakened by allowing for time variation in the risk- return relationship over the business cycle, on the contrary it uncovers a better relationship. This is despite the fact that this chapter found no relationship during recession while Post and Vliet (2005) point out that they find downside beta's superiority is more apparent in bad economic times.

Furthermore, the results of the chapter suggested that industry could be an important distinguisher between downside risk and CAPM beta performances where the

284

former appears to be more suitable for some industries. This is consistent with the findings of Olmo (2007) for UK.

5.5

Implications

The evidence from this thesis has important implications for understanding the behaviour of stock returns, which is a necessity for every investor and researcher in finance who need to make decisions about changes in stock prices. The former group of beneficiaries includes management, stock holders, bankers etc., who need to make the decision to invest or not based on the risk-return aspects of their investment options. On the other hand researchersneed to be aware of what factors are important to price stocks, whether they directly study asset pricing or merely employ these models as a tool to adjust the return on their investing strategies which may change the findings on the empirical grounds. Although this thesis provides evidence that is generally in line with the literature, it also gives some insights related to some unresolved issues.

First,

it

strongly

supports Campbell

(2000)

and Cochrane (2006)

that

macroeconomic factors are crucial in shaping stock returns and it is important for LSE investors not to overlook any potential macro variable that may have association with stock returns. It also supports the importance of HML in particular and SMB as pricing factors in the UK stock market, although it does not find any direct association between them and the macroeconomic variables. Second, it joins those who very much doubt that idiosyncratic risk has any real ability to explain stock returns and stressesthe importance

285

help in for to time resolve which seems measures correctly risk variation of accounting this issue. Finally, it supports the importance of downside risk for stock returns and that allowing for time variation in the relationship strengthens it.

5.6

Further

Area for Research

5.6.1

Idiosyncratic Risk and Time-Varying Betas

Chapter (3) of this thesis studied idiosyncratic risk in the cross section of UK stock returns. Fama and French's (1993) three factor model is used as the asset pricing model to calculate idiosyncratic risk measure.This was done following other studies in this area, as explained in detail in the chapter, which employ this model as the underlying asset pricing model. The results of the chapter showed that there is mixed and inconsistent evidence as to whether idiosyncratic risk is priced in the UK or not. Furthermore, the results show that the UK Fama and French's (1993) three risk factors are not significant in the Fama and MacBeth's (1973) cross sectional regression of daily returns.

However, when Avramov and Chordia's (2006) conditional Fama and French's (1993) three-factor model, which accounts for time-variation in beta risk, is used as the underlying asset pricing model, the results showed that idiosyncratic risk, based on in frequency, becomes insignificant the cross sectional returns. On the other monthly hand, the findings of chapter (2) of this thesis showed a number of factors are priced in the UK stock market. These include Petkova's (2006) model augmented with innovations

286

to an estimated factor that relates to unemployment.

At the same time Petkova (2006)

her that model succeedsas a conditional model while Fama and French's (1993) reported model does not. In light of this, it would be interesting to apply this model as the underlying model in studying idiosyncratic risk. The hypothesis here is that if this model succeeds in capturing idiosyncratic risk, then this confirms that it is also a conditional model. This is because this hypothesis examines if this model really captures time variation in risk without an explicit modelling of beta risk as has been done in chapter (3) using Avramov and Chordia's (2006) methodology and their conditional Fama and French's (1993) model.

5.6.2

Downside Risk and Business Cycle

Chapter (4) of this thesis studied downside risk in the UK cross section of returns
by applying Ang Chen and Xing's (2006) study. The results of the chapter showed that downside risk premium, when examined over the full sample period (unconditionally), does not show significant improvement over the CAPM. Furthermore, within the largest between downside

stocks, it was found that there is, generally, a negative relationship

risk and stock returns. In addition within the growth stock, it was found that there is no clear relationship between stock returns and downside risk. Therefore, it would be

interesting to examine if downside risk modelled using Avramov conditional methodology, which accounts for time varying

and Chordia's (2006) improve the

beta, would

performance of downside risk within large and growth stocks.

287

Furthermore, it has been shown that there are some inconsistent findings for the between downside risk and stock average returns in the Healthcare industry relationship from the obtained results portfolio formation are compared to those using Fama and when MacBeth's (1973) cross sectional regression. it has been found that, using portfolio between downside risk and returns. However, using positive

formation, there is no relationship Fama and Maceth's

(1973) cross sectional regression produced a significant

downside risk premium and a significant

negative upside risk premium. In addition, for Chemicals and Utilities, the results based

some industries such as; Telecommunications,

on Fama and McBeth's (1973) cross sectional regression showed that neither CAPM beta nor downside beta is priced. This may suggest that there may be other risk factors affect stocks in these industries. In addition, the results could be affected by the fact that the number of the stocks available within some of these industries is not sufficiently Therefore, it would be interesting large.

to split the full sample of stocks into groups of

industries based on their covaraition with the business cycle, instead of working within each individual industry. The split based on the covaraition of the stock's industry with as Ang, Chen and Xing (2006) point out that they find the

business cycle is justified

utilities industry is less exposed to downside risk which is in agreement with the fact that these stocks protect their value in the down market. Furthermore, Olmo (2007) reported that he finds that downside beta is not priced for stocks that are not affected by the downturn market in the UK stocks, while stocks that have positive (negative )

downturn higher have (lower) returns compared with the CAPM. with market covariation

288

In addition, the results showed inconclusive findings concerning the relationship between stock returns and downside risk for the recession period. This could be due to the short sample period; therefore, it would be interesting to split the full sample as in Post and Vliet (2005) who split their sample period into equal size sub-samplesusing the median of a conditioning variable.

289

Bibliography

A. References

AI-Horani, A., Pope, P., and Stark, A., (2003), "Research and Development Activity Expected Returns in the United Kingdom", European Finance Review, 7, pp. 27-46.
Anderson, C., and Garcia-Feijo (2006), "Empirical Evidence on Capital

and

Investment,

Growth Options,andSecurityReturns",Journalof Finance,61, pp.171-194.

Ang, A., and Chen, J. (2007), "CAPM Empirical Finance, 14, pp. 1-40. Over the Long Run: 1926-2001", Journal of

Ang, A., Chen, J., and Xing, Y., (2006), "Downside Risk", Review of Financial Studies, 19, pp. 1191-1239. Ang, A., Hodrick, R., Xing, Y., and Zhang, X., (2006), "The Cross-Section of Volatility and Expected Returns", Journal of Finance, 61, pp.259-299.
Ang, A., Hodrick, R., Xing, Y., and Zhang, X., (2008), "High Idiosyncratic Volatility and Low Returns: International and Further U. S Evidence", NBER Working Paper series no. 13739j. Angelidis, T., and Tessaromatis, N., (2008a), "Does Idiosyncratic Risk Matter? Evidence from European Stock Markets", Applied Financial Economics, 18, pp. 125-137. Angelidis, T., and Tessaromatis, N., (2008b), "Idiosyncratic Volatility and Equity Returns: UK Evidence", International Review of Financial Analysis, 17, pp. 539-556. Antoniou, A., Garrett, I., and Priestley, R., (1998), "Macroeconomic Variables as Common Pervasive Risk Factors and the Empirical Content of The Arbitrage Pricing Theory", Journal of Empirical Finance, 5, pp. 221-240. Antoniou, A., Lam, H., and Paudyal, K., (2007), "Profitability of Momentum Strategies in International Markets: The Role of Business Cycle Variables and Behavioural Biases", Journal of Banking and Finance, 31, pp. 955-972.

Artis, M., Banerjee, A., Marcellino, M., (2005), "Factor Forecasts for the UK", Journal of Forecasting, 24, pp.279-298.
Asprem, M., (1989), "Stock Prices, Asset Portfolios and Macroeconomic Ten European Countries", Journal of Banking and Finance, 13, pp. 589-612. Variables in

'9 Forthcoming in Journal of Financial Economics, based on correspondences with Professor Hodrick. Acknowledge Professor Hodrick for giving us the permission to use his paper.

We

290

Au, A., Doukas, J., and Onayev, Z., (2007), "Daily Short Interest, Idiosyncratic Stock Returns", Working paper series (State Street Global Advisors' Global Equities Group in Boston, Massachusetts; Graduate School of Business, Old University; and State Street Global Advisors' Advanced Research Center Massachusetts).

Risk, and Enhanced Dominion in Boston,

Avramov, D., and Chordia, T., (2006), "Asset Pricing Models and Financial Market Anomalies", Review of Financial Studies, 19, pp. 1001-1040.

Bali, T., and Cakici, N., (2008), "Idiosyncratic Volatility and the Cross-Section of Expected Returns", Journal of Financial and Quantitative Analysis, 43, pp.29-58.
Bali., T., Cakici, N., Yan, X., and Zhang Z., (2005), "Does Idiosyncratic Matter? ", Journal of Finance, 60, pp. 905-929. Risk Really

Bali, T., Demirtas, K., and Levy, H., "Is There a Relation Between Downside Risk and Expected Stock Returns? ", Manuscript ( Zicklin School of Business, Baruch College, City University of New York; Zicklin School of Business, Baruch College, City University of New York Department of Finance; and Jerusalem School of Business The Administration, Hebrew University Available Jerusalem. at of http: //w4. stern. nyu. edu/finance/docs/pdfs/Seminars/06I f-bali. Ddf. Final Access on 18 September 2008 Ball, R., (1978), "Anomalies in Relationships Between Securities' Surrogates", Journal of Financial Economics, 6, pp. 103-126 Yields and Yield-

Bawa, V., and Lindenberg, E., (1977), "Capital Market Equilibrium in a Mean-Lower Partial Moment Framework", Journal of Financial Economics, 5, pp. 189-200.

Bernanke, B., Boivin, J. and Eliasz, P., (2005), "Measuring the Effects of Monetary Policy: a Factor-Augmented Vector Autoregressive (FAVAR) Approach", Quarterly Journal of Economics, 120, pp. 387-422.
Black, A., and Fraser, P., (2000), "International Comparisons on Stock Market ShortTermism: How Different is the UK Experience? ", Manchester School Supplement, 68, pp. 38 - 50. Black, F, Jensen, M., and Scholes, M., (1972), "The Capital Asset Pricing Model: Some Empirical Tests", In Studies in the Theory of Capital Markets, Eds., Jensen, M., Praeger Publishers Available at: ,

http://papers.ssrn.com/sol3/papers. cfm? abstract id=908569. Final Access on 18 September2008.

291

Blake, D., (1996), "Efficiency, Risk Aversion and Portfolio Insurance: An Analysis of Financial Asset Portfolios Held by Investors in the United Kingdom", Economic Journal, 106, pp. 1175-1192.
Boehme, R., Danielsen, B., Kumar, P., and Sorescu, S., (2005), "Idiosyncratic Risk and Working the Cross-Section of Stock Returns: Merton (1987) Meets Miller (1977)", Paper series ( W. Frank Barton School of Business at Wichita State University; Kellstadt College of Commerce at DePaul University; C. T. Bauer College of Business, University of Houston; and Mays Business School at Texas A&M University)

Boivin, J., and Ng S., (2006), "Are More Data Always Better for Factor Analysis?", , Journal of Econometrics, 132, pp. 169-194. Boyd, J., Hu, J., and Jagannathan, R., (2005), "The Stock Market's Reaction to Unemployment News: Why Bad News Is Usually Good for Stocks", Journal of Finance, 60, pp.649-672. Breeden, D., (1979), "An Intertemporal Asset Pricing Model with Stochastic Consumption and Investment Opportunities", Journal of Financial Economics, 7, pp.265296.
Brennan, M., Chordia, T., and Subrahmanyam, A., (1998), Specifications, Security Characteristics, and the Cross-Section Returns", Journal of Financial Economics, 49, pp. 345-373. Factor "Alternative of Expected Stock

Brennan, M., Wang, A., and Xia, Y., (2004), "Estimation and Test of a Simple Model of Intertemporal Capital Asset Pricing", Journal of Finance, 59, pp. 1743-1775. Brockman, P., and Schutte, M., (2007), "Is idiosyncratic volatility priced? The international evidence", Working Paper, University of Missouri, Columbia. (Cited by Fu (2007)) Burmeister, E., and McElroy, M., (1988), "Joint Estimation of Factor Sensitivities and Risk Premia for the Arbitrage Pricing Theory", Journal of Finance, 43, pp. 721-733. Campbell, J., (1993), "Intertemporal Asset Pricing without Consumption Data" American Economic Review, 83, pp. 487-512. Campbell, J., (1996), "Understanding Risk and Return", Journal of Political Economy, 104, pp.298-345.
Campbell, J., (1998), "Asset Prices, Consumption, Working Paper series no. 6485. and the Business Cycle", NBER

Campbell, J., (2000), "Asset Pricing at the Millennium", Journal of Finance, 55, pp. 1515-1567.

292

Chen. J., (2003), "Intertemporal CAPM and the Cross-Section of Stock Returns", Working paper (University of Southern California). Chen, N., (1991), "Financial Investment Opportunities and the Macroeconomy", Journal of Finance, 46, pp. 529-554.
Chen, N, Roll, R., and Ross, S., (1986), " Economic Forces and the Stock Market", Journal of Business, 59, pp. 383-403

Chen, S., (1982), "An Examination of Risk-Return Relationship in Bull and Bear Markets Using Time-Varying Betas", Journal of Financial and Quantitative Analysis, 17, pp. 265286.
Chen, S., and Keown, A., (1981), "An Examination of the Relationship between Pure Residual and Market Risk: A Note", Journal of Finance, 36, pp. 1203-1209.

Chua, C., Goh, J., and Zhang, Z., (2007), "Expected Volatility, Unexpected Volatility, and the Cross-Section of Stock Returns", Working Paper (Lee Kong Chian School of Business, Singapore Management University) Clare, A., and Thomas, S., (1994), "Macroeconomic Factors, The APT and the UK Stockmarket", Journal of Business Finance and Accounting, 21, pp. 309-330. Cochrane, J., (1999), "New Facts in Finance", Federal Reserve Bank of Chicago; Economic Perspectives,23, pp.36-58. Available at: http: //www. chica ofed.org/publications/economicperspectives/I999/ep3099 3.pdf. Final Access on 18 September2008. Cochrane, J., (1996), "A Cross-Sectional Test of an Investment-Based Asset Pricing Model", Journal of Political Economy, 104, pp. 572-621. Cochrane, J., (2001), "Asset Pricing", New Jersey: Princeton University Press, Princeton,
Cochrane, J., (2006), "Financial Markets and the Real Economy", (Graduate School of Business, University of Chicago) Working paper

Constantinides, G., (1989), "Theory of Valuation: Overview and Recent Developments", in Theory of Valuation: Frontiers of Modern Financial Theory. Eds., Bhattacharya, S., and Constantinides, G. Rowman and Littlefield: Totowa, N. J, pp. 1-23. Available at: http: //faculty. chicayoysb. edu/george. constantinides/documents/Valuation. pdf Final Access on 18 September 2008. Connor, G., and Korajczyk, R, (1986), "Performance Measurement with the Arbitrage Pricing Theory: A New Framework for Analysis, Journal of financial Economics, 15, pp. 373-394.

293

Connor, G., & Korajczyk, R, (1988), "Risk and Return in Equilibrium APT, Application Methodology", 255-289. Test Journal Financial Economics, 2 1, New of pp. of a
Daniel, K., and Titman, S., (1997), "Evidence on the Characteristics of Cross Sectional Variation in Stock Returns", Journal of Finance, 52, pp. 1-33. Daniel, K., and Titman, S., (2005), "Testing Factor-Model Explanations of Market Anomalies", Manuscript (Kellogg School at Northwestern University and NBER; and College of Business Administration, University of Texas, Austin and NBER).

Davis, J., Fama, E., and French, K., (2000), "Characteristics, Covariances, and Average Returns: 1929 to 1979", Journal of Finance, 55, pp.389-406.
Diavatopoulos, D., Doran, J., and Peterson, D., (2007), "The Information Content in Implied Idiosyncratic Volatility and the Cross-Section of Stock Returns: Evidence from the Option Markets", Journal of Futures Markets, Forthcoming. Dimson, E., and Marsh, P., (1999), "Murphy's Law and Market Anomalies". Journal of Portfolio Management, 25, pp. 53-69, cited by Dimson, Nagel and Quigley (2003). Dimson, E., Nagel, S., and Quigley, G., (2003), "Capturing United Kingdom", Financial Analysts Journal, 59, pp. 35-45. Fama, E., (1991), "Efficient the Value Premium in the

Capital Markets: 11", Journal of Finance, 46, pp. 1575-1617.

Fama, E., and French, K., (1989), "Business Conditions and Expected Returns on Stocks and Bonds", Journal of Financial Economics, 25, pp. 23-49.

Fama, E., and French, K., (1992), "The Cross-Section of Expected Stock Returns", Journal of Finance, 47, pp.427-465. "Common E.,andFrench, Fama, K., (1993),
Bonds", Journal of Financial Economics.

33, pp. 3-56.

Risk Factors in the Returns on Stocks and

Fama, E., and French, K., (1995), "Size and Book-to-Market Returns", Journal of Finance, 50, pp. 131-155. Fama, E., and French, K., (1996), "Multifactor Anomalies", Journal of Finance, 5 1, pp. 55-84.

Factors in Earnings and

Explanations

of

Asset

Pricing

Fama, E., and French, K., (1998), "Value Versus Growth: the International Evidence", Journal of Finance, 53, pp. 1975-1999. Fama, E., and French, K., (2004), "The Capital Asset Pricing Model: Theory and Evidence", Journal of Economic Perspectives, 18, pp 25-46.

294

Fama, E., and MacBeth, J., (1973), "Risk, Return, and Equilibrium: Empirical tests", Journal of Political Economy, 81, pp. 607-636. Ferson, W., and Harvey, C., (1991), "The Variation of Economic Risk Premiums", Journal of Political Economy, 99, pp. 385-415. Ferson, W., and Harvey, C., (1999), "Conditioning Variables and the Cross Section of Stock Returns", Journal of Finance, 54, pp. 1325-1360. Ferson, W., Kandel, S., and Stambaugh, R., (1987), "Tests of Asset Pricing with TimeVarying Expected Risk Premiums and Market Betas", Journal of Finance, 42, pp.201220.
Ferson, W., Sarkissian, S., and Simin, T., (1999), "The Alpha Factor Asset Pricing Model: A Parable", Journal of Financial Markets, 2, pp. 49-68. Flannery, M., and Protopapadakis, A., (2002), "Macroeconomic Factors Do Influence Aggregate Stock Returns", Review of Financial Studies, 15, pp. 751-782.

Fletcher, J., (2001), "An Examination of Alternative Factor Models in UK Stock Returns", Review of Quantitative Finance and Accounting, 16, pp. 117-130.
Fletcher, J., (2007), "Can Asset Pricing Models Price Idiosyncratic Returns? ", The Financial Review, 42, pp. 507-535. Risk in U. K. Stock

Fletcher, J., and Kihanda, J., (2005), "An Examination of Alternative CAPM-Based Models in UK Stock Returns", Journal of Banking & Finance, 29, p. 2995-3014.

Forni, M., Hallin, M., Lippi, M., and Reichlin, L. (2000), "The Generalized DynamicFactor Model: Identification and Estimation", Review of Economics and Statistics, 82, pp.540-554. Forni, M., Hallin, M., Lippi, M., and Reichlin, L. (2004), "The Generalized Dynamic Factor Model Consistency and Rates", Journal of Econometrics, 119, pp.231-255.
Forni, M., Hallin, M., Lippi, M., and Reichlin, L., (2005), "The Generalized Dynamic Factor Model: One-Sided Estimation and Forecasting", Journal of the American Statistical Association, 100, pp. 830-840. French, K., Schwert, G., and Stambaugh., R., (1987), "Expected Volatility", Journal of Financial Economics, 19, pp. 3-29. Stock Returns and

Fu. F., (2007), "Idiosyncratic Risk and the Cross-Section of Expected Stock Returns", Working Paper series (Singapore Management University, Lee Kong Chian School of Business)

295

Ghysels, E., (1998), "On Stable Factor Structures in the Pricing of Risk: Do TimeVarying Betas Help or Hurt?," Journal of Finance, 53, pp 549-573.
Gillitzer, C., Kearns, J., and Richards, A., (2005), "The Australian Business Cycle: A Coincident Indicator Approach", Working paper (Economic Group, Reserve Bank of Australia).

Goyal, A., and Santa-Clara, P., (2003), "Idiosyncratic Risk Matters! ", Journal of Finance, 58, pp. 975-1007. Griffin, J., (2002), "Are the Fama and French Factors Global of Country Specific?", Review of Financial Studies, 15, pp. 783-803.
Gul, F., (1991), "A Theory of Disappointment Aversion", Econometrica, 59, pp. 667-686.

Guo, H., and Savickas, R., (2006), "Idiosyncratic Volatility, Stock Market Volatility, and Expected Stock Returns", Journal of Businessand Economic Statistics, 24, pp. 43-56. Guo, H., and Savickas, R., (2008), "Average Idiosyncratic Volatility in G7 Countries", Review of Financial Studies, 21, pp. 1259-1296. Hahn, J., and Lee, 1-1., (2006), "Yield Spreads as Alternative Risk Factors for Size and Book-to-Market", Journal of Financial and Quantitative Analysis, 41, pp.245-269. Hall, S. (1991), "The Effect of Varying Length VAR Models on the Maximum Likelihood Estimates of Cointegrating Vectors", Scottish Journal of Political Economy, 38, pp. 317-323 Hamilton, J, (1994), " Time-Series Analysis", New Jersey: Princeton University Press Princeton.
Hanhardt, A., and Ansotegui C., (2008), " Do the Fama and French Factors Proxy for , State Variables that Predict Macroeconomic Growth in the Eurozone? ", Working Paper Series (ESADE Business School, University Ramon I, lull).

Hansen, L., (1982), "Large Sample Properties of Generalized Method of Moments Estimators" Econometrica, 50, pp. 1029-1054. Harlow, W., and Rao, R., (1989), "Asset Pricing in a Generalized Mean-Lower Partial Moment Framework: Theory and Evidence", Journal of Financial and Quantitative Analysis, 24, pp.285-311
Harris, R., and Sollis, R., (2003), "Applied England: John Wiley and Sons. Time Series Modelling and Forecasting",

296

Hodrick, R., and Zhang, X., (2001), "Evaluating the Specification Errors of asset Pricing Models", Journal of Financial Economics, 62, pp. 327-376 Huang, W., Liu, Q., Rhee, S., and Zhang, L., (2006), "Return Reversals, Idiosyncratic Risk and Expected Returns", Manuscript (Department of Financial Economics and Institutions, Shidler College of Business, University of Hawaii at Manoa) Hussain,l., Toms, S. and Diacon, S., (2002), "Financial Distress, Market Anomalies and Single and Multifactor Asset Pricing Models: New Evidence", Working paper series (Nottingham University Business School, University of Nottingham). Inklaar, R., Jacobs, J., and Romp, W., (2003), "Business Cycle Indexes: Does a Heap of Data Help?", Working paper (University of Groningen). Jagannathan,R. and Wang, Z., (1996), "The Conditional CAPM and the Cross-Section of Expected Returns" Journal of Finance, 51, pp. 3-53. Jagannathan, R., and Wang, Z., (1998), "An Asymptotic Theory for Estimating BetaPricing Models Using Cross-Sectional Regression", Journal of Finance, 53, pp. 12851309.
Jagannathan, R., and Wang, Z., (2002), "Empirical Evaluation of Asset-Pricing Models: A Comparison of the SDF and Beta Methods", Journal of Finance, 57 pp. 2337-2367.

Jahankhani, A., (1976), "E-V and E-S Capital Asset Pricing Models: Some Empirical Tests", Journal of Financial and Quantitative Analysis, 11,513-528. Jegadeesh,N., and Titman, S., (1993), "Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency", Journal of Finance, 48, pp.65-91.
Jones, C., (2001), "Extracting Factors from Heteroskedastic Financial Economics, 62, pp. 293-325. Asset Returns", Journal of

Kapetanios, G., Labhard, V., and Price, S., (2006), " Forecasting using Bayesian and Information Theoretic Model Averaging: An Application to UK Inflation", Working Paper no.566 (Department of Economics, Queen Mary University of London) Kelly, P., (2003), "Real and Inflationary Macroeconomic Risk in the Fama and French Size and Book-to-Market Portfolios", Working Paper series (W. P. Carey School of Business, Arizona State University) Kim, M., and Zumwalt, J., (1979), "An Analysis of Risk in Bull and Bear Markets", Journal of Financial and Quantitative Analysis, 14, pp 1015-1025.

297

Lagan, G., Mountford, A., (2005), "Measuring Monetary Policy In The UK: A FactorAugmented Vector Autoregression Model Approach", The Manchester School, 73, Supplement, pp. 77-98 Lehmann, B., (1990), "Residual Risk Revisited", Journal of Econometrics, 45, pp. 71-97. Lettau, M., and Ludvigson, S., (2001), "Resurrecting the (C)CAPM: A Cross-Sectional Test When Risk Premia are Time-Varying", Journal of Political Economy, 109, pp. 12381287. Levy, H., and Markowitz, H., (1979), "Approximating Expected Utility by a Function of Mean and Variance", American Economic Review, 69, pp. 308-317. Lewellen, J., (1999), "The Time-Series Relations among Expected Return, Risk, and Book-to-Market", Journal of Financial Economics, 54 pp. 5-43. Lewellen, J., Nagel, S., and Shanken, J., (2006), "A Skeptical Appraisal of Asset-Pricing Tests", Working Paper series (Dartmouth and NBER; Stanford and NBER; Emory and NBER) Liang, S., and Wei, K., (2006), "Volatility and Expected Returns Around the World", Working Paper series (Hong Kong University of Science and Technology)
Liew, J., Vassalou, M., (2000), "Can Book-to-Market, Size and Momentum be Risk Factors that Predict Economic Growth? ", Journal of Financial Economics, 57, pp. 221245.

Lo, A., and MacKinlay, A., (1990), "Data-Snooping Biases in Tests of Financial Asset Pricing Models", Review of Financial Studies, 3, pp.431-467. Ludvigson, S. and Ng, S., (2007), "The Empirical Risk-Return Relation: A Factor Analysis Approach", Journal of Financial Economics, 83, pp. 171-222. Malkiel, B., and Xu, Y., (2006), "Idiosyncratic Risk and Security Returns", Working Paper (Department of Economics, Princeton University; and School of Management, University of Texas at Dallas). Mankiw, N., and Zeldes, S., (1991), "The Consumption of Stockholders and Nonstockholders", Journal of Financial Economics, 29 pp. 97-112.
Forecasting in the Marcellino, M., Stock, J., and Watson, M., (2003), "Macroeconomic European Economic Euro Area: Country Specific Versus Area-Wide Information", Review, 47, pp. l - 18.

Markowitz, H., (1959), "Portfolio Selection, Efficient Diversification of Investments", John Wiley and Sons, New York, [Cowles Foundation for Research in Economics at Yale university]

298

Available at: http://cowles. econ.yale.edu/P/cm/ml6/ml6-all. pdf. Final Access on 18 September2008. Merton, R., (1973), "An Intertemporal Capital Asset Pricing Model", Econometrica, 41, pp. 867-887. Merton, R., (1974), "On the Pricing of Corporate Debt: The Risk Structure of Interest Rates", Journal of Finance, 29, pp. 449-470.
Merton, R., (1987), "A Simple Model of Capital Market Equilibrium Information", Journal of Finance, 42, pp. 483-510. with Incomplete

Miller, E., (1977), "Risk, Uncertainty, and Divergence of Opinion", Journal of Finance, 32, pp. 1151-1168. Miller, M., and Scholes, M., (1972), 'Rates and Return in Relation to Risk: A Reexamination of Some Recent Findings, ' in Studies in the Theory of Capital Markets, Eds., Jensen,M., Praeger,New York, pp. 47-78. (cited by Malkiel and Xu (2006)) Mnch, E., (2004), "Explaining the Cross-Section of Returns with Macroeconomic Factors", Unpublished Manuscript (Federal Reserve Bank of New York40)
Mnch, E., (2006), "Conditional Asset Pricing with a Large Information Paper series (Federal Reserve Bank of New Yorkal) Set", Working

Nagel, S., (2001), "Accounting Information Free of Selection Bias: A New UK Database 1953-1999", Working Paper series (London Business School) Nawrocki, D., (1999), "A Brief History of Downside Risk Measures". Journal of Investing, pp.9-26. Available at: http://www56. homepage. villanova. edu/david.nawrocki/nawrock3. htm Final Access on 19 September 2008. Newey, W., and West, K., (1987), "A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix", Econometrica, 55, pp. 703-708. Olmo, J., (2007), "An Asset Pricing Model for Mean-Variance-Downside-Risk Averse Investors", Working Paper series no.07/01 (Department of Economics, City University). Pedersen,C., and Hwang, S., (2007), "Does Downside Beta Matter in Asset Pricing?", Applied Financial Economics, 17, pp.961-978. Perez-Quiros, G., and Timmermann, A., (2000), "Firm Size and Cyclical Variations in Stock Returns", Journal of Finance, 55, pp. 1229-1262.
40This new affiliation is noted on the request of author based on correspondenceswith him 41See the preceding footnote

299

Perron, P., (1997), "Further Evidence on Breaking Trend Functions in Macroeconomic Variables", Journal of Econometrics, 80, pp. 355-385. Petkova, R., (2006), "Do the Fama-French Factors Proxy for Innovations in Predictive Variables? ", Journal of Finance, 61, pp. 581 - 612.

Petkova, R., and Zhang, L., (2005), "Is Value Riskier than Growth?", Journal of Financial Economics, 78, pp. 187-202.
Post, T., and Vliet, P., (2005), "Empirical Tests of the Mean-Semivariance Working Paper series (Erasmus University Rotterdam). CAPM",

Post, T., and Vliet, P., (2006), "Downside Risk and Asset Pricing", Journal of Banking and Finance, 30, pp. 823-849. Price, K., Price, B., and Nantell, T., (1982), "Variance and Lower Partial Moment Measures of Systematic Risk: Some Analytical and Empirical Results", Journal of Finance, 37, pp.843-855.
Priestley, R., (1996), "The Arbitrage Pricing Theory, Macroeconomic and Financial Factors, and Expectations Generating Processes", Journal of Banking and Finance, 20, pp. 869-890.

Roll, R., (1977), "A Critique of the Asset Pricing Theory's Tests Part I: On Past and Potential Testability of the Theory", Journal of Financial Economics, 4, pp. 129-176. Roll, R. and Ross, S., (1980), "An Empirical Investigation of the Arbitrage Pricing Theory", Journal of Finance, 35, pp. 1073-1103. Ross, S., (1976), "The Arbitrage Theory of Capital Asset Pricing", Journal of Economic Theory, 13, pp.341-360. Sarnat, M., (1974), "A Note on the Implications of Quadratic Utility for Portfolio Theory", Journal of Financial and Quantitative Analysis, 9, pp.687-689. Schneider, M., and Spitzer, M., (2004), "Forecasting Austrian GDP Using the Generalized Dynamic Factor Model", Working paper 89 (Oesterreichische Nationalbank). Schwert, G., (2003), "Anomalies and Market Efficiency", in Handbook of the Economics of Finance, Eds. Constantinides, G., Harris, M., and Stulz, R., Elsevier Science B. V, Ch (15), pp.937-972, Available at http: //schwert. ssb.rochester.edu/hbfechI5. pdf, Final Access on 15 September2008.

300

Shanken, J, (1992), "On the Estimation of Beta-Pricing Models", Review of Financial Studies, 5, pp 1-33. Shanken, J. and Weinstein, M., (2006), "Economic Forces and the Stock Market Revisited", Journal of Empirical Finance, 13, pp. 129-144. Sharpe, W., (1964), "Capital Asset Prices: A Theory of Market Equilibrium Conditions of Risk", Journal of Finance, 19, pp. 425-442. under

Spiegel, M., and Wang, X., (2005), "Cross-Sectional Variation in Stock Returns: Liquidity and Idiosyncratic Risk", Working Paper series (Yale School of Management). Stock, J., and Watson, M., (1998), "Diffusion Indexes", NBER Working Paper no. 6702. Stock, J., and Watson, M., (2002a), "Macroeconomic Forecasting Using Diffusion Indexes", Journal of Business and Economic Statistics, 20, pp. 147- 162. Stock, J., and Watson, M., (2002b), "Forecasting Using Principal Components From a Large Number of Predictors", Journal of the American Statistical Association, 97, pp. 1167-1179.
Vassalou, M., (2003), "News Related to Future GDP Growth as a Risk Factor in Equity Returns", Journal of Financial Economics, 68, pp. 47-73.

Wallis, K., (1974), "Seasonal Adjustment and Relations Between Variables", Journal of the American Statistical Association, 69, pp. 18-31. Watson, M., (2001), "Macroeconomic Forecasting Using Many Predictors", Working paper ( Department of Economics and Woodrow Wilson School Princeton University Yogo, M., (2006), "A Consumption-Based Explanation of Expected Stock Returns", Journal of Finance, 61, pp. 539-580.

301

B. Wider Relevant Reading

Bai J., and Ng, S., (2006), "Confidence Intervals For Diffusion Index Forecasts and , Inference for Factor-Augmented Regressions", Econometrica, 74, pp. 1133-1150.

Balvers, R., Cosimano, T., and McDonald, B., (1990), "Predicting Stock Returns in an Efficient Market", Journal of Finance, 45, pp. 1109-1128. Banz, R., (1981), "The Relationship Between Return and Market Value of Common Stocks", Journal of Financial Economics, 9, pp. 3-18. Basu, S., (1983) "The Relationship Between Earnings' Yield, Market Value and Return , For NYSE Common Stocks Further Evidence", Journal of Financial Economics, 12 pp. , 129-156. Bawa, V., (1975), "Optimal Rules for Ordering Uncertain Prospects", Journal of Financial Economics, 2, pp. 95-121.
Berk, J, Green, R., and Naik, V., (1999), "Optimal Investment, Security Returns", Journal of Finance, 54, pp. 1553-1607. Growth Options, and

Berkelaar, A., Kouwenberg, R., and Post, T., (2004), "Optimal Portfolio Loss Aversion", Review of Economics and Statistics, 86, pp. 973-987. Bhandari, L., (1988), "Debt/Equity Ratio and Expected Empirical Evidence", Journal of Finance, 43,, pp. 507-528 Common

Choice Under

Stock

Returns:

Campbell, J., and Shiller, R., (1987). "Cointegration and Tests of Present Value Models", Journal of Political Economy, 95, pp. 1062-1088. Campbell, J, and Shiller, R., (1988), "Stock Prices, Earnings, and Expected Dividends", Journal of Finance, 43, pp. 661-676. Campbell, J., and Vuolteenaho, T., (2004), "Bad Beta, Good Beta", American Economic Review, 94,1249-1275. Chan, L., Hamao, Y., and Lakonishok, J., (1991), "Fundamentals and Stock Returns in Japan", Journal of Finance, 46, pp. 1739-1764.
Business Cycle, Chordia, T., and Shivakumar, L., (2002), "Momentum, Varying Expected Returns", Journal of Finance, 57, pp. 985-1019. and Time

Cochrane, J., (1991), "Production-Based Asset Pricing and the Link Between Stock Returns and Economic Fluctuations", Journal of Finance, 46, pp. 209-237

302

Cochrane, J., (2005)42, "Financial Markets and the Real Economy", Foundations and Trends in Finance, 1, pp. 1-101, Available at:
http: //www. nowpubl ishers. com/product. aspx? product=FIN&doi=0500000001. Final Access on 15 September 2008.

Cochrane, J., (2007) 43,"Financial Markets and the Real Economy", in Handbook of the Equity Risk Premium, Eds. Mehra, R., Elsevier By, Ch (7), pp. 237-330. Available at: http: //faculty. chicago s b_edu/john. cochrane/research/Papers/financial and real pro ofs aug 07.pdf. Final Access on 15 September 2008 Cuthbertson, K., (1988), "Expectations, Learning and the Kalman Filter", Manchester School, 56, pp. 223-246. D'Agostino, A., and Giannone, D., (2006), "Comparing Alternative Predictors Based on Large-Panel Factor Models", Working Paper series no. 680 (European Central Bank). Dickey, D., and Fuller, W., (1979), "Distribution of the Estimators for Autoregressive Time Series With a Unit Root", Journal of the American Statistical Association, 74, pp.427-431.
Dickey, D., and Fuller, W., (1981), "Likelihood Ratio Statistics for Autoregressive Series with a Unit Root", Econometrica, 49, pp. 1057-1072. Time

Dittmar R., (2002), "Nonlinear Pricing Kernels, Kurtosis Preference, and Evidence from the Cross Section of Equity Returns", Journal of Finance, 57, pp.369-403. Fama, E., (1981), "Stock Returns, Real Activity, Economic Review, 71,545-565. Inflation, and Money", American

Forni, M., and Lippi, M., (2001), "The Generalized Dynamic Representation Theory", Econometric Theory, 17, pp. l 113-1141.

Factor

Model:

Ghysels, E., Santa-Clara, P., and Valkanov, R., (2005), "There is a Risk-Return Trade-off After All", Journal of Financial Economics, 76, pp. 509-548.

Gibbons, M., Ross, S., and Shanken, J., (1989), "A Test of the Efficiency of a Given Portfolio", Econometrica, 57, pp. 1121-1152. Gomes, J., Kogan, L., and Zhang, L., (2003), "Equilibrium Cross Section of Returns", Journal of Political Economy, 111, pp.693-732. Greene, W., (2003), Econometric Analysis, 5th ed. N. J. Prentice Hall,

42Another version of Cochrane (2006) that is cited in the thesis 4' Another version of Cochrane (2006) that is cited in the thesis

303

Harvey, C., and Siddique, A., (1999), "Autoregressive Conditional Skewness", Journal of Financial and Quantitative Analysis, 34, pp.465-487. Harvey, C., and Siddique, A., (2000), "Conditional Skewness in Asset Pricing Tests", Journal of Finance, 55, pp. 1263-1295. Li, Q., Vassalou, M., and Xing, Y., (2006), "Sector Investment Growth Rates and the Cross Section of Equity Returns", Journal of Business, 79, pp. 1637-1665. Markowitz, H., (1952), "Portfolio Selection", Journal of Finance, 7, pp. 77-91 Michou, M., Mouselli, S., and Stark, A., (2006), "Estimating the Fama and French Factors in the UK - an Empirical Review", Manchester Business School Working Paper Series (University of Edinburgh Management School; Manchester Business School; Manchester Business School) Pagan, A., (1984), "Econometric Issues in the Analysis of Regressions with Generated Regressors", International Economic Review, 25, pp. 221-247. Perron, P., (1988), "Trends and Random Walks in Macroeconomic Time Series Further Evidence From a New Approach", Journal of Economic Dynamics and Control, 12, pp. 297-332. Perron, P., (1989), "The Great Crash, the Oil Price Shock, and the Unit Root Hypothesis", Econometrica, 57, pp. 1361-1401. Schwert, G., (1989), "Tests for Unit Roots: A Monte Carlo Investigation", Journal of Business and Economic Statistics, 7, pp.5-17. Sims, C., (1980), "Macroeconomics and Reality", Econometrica, 48, pp. 1-48. Zhang, L., (2005), "The Value Premium", Journal of Finance, 60, pp.67-103.

'",.N F /Q C 11

k c
304