Performance Evaluation and Attribution of Security Portfolios

Table of Contents

Cover image

Title Page

Introduction to the Series

Copyright

Preface

Section 1: Performance Evaluation

Chapter 1. An Introduction to Asset Pricing Models

1.1 Historical Asset Pricing Models

1.2 The Beginning of Modern Asset Pricing Models

1.3 Efficient Markets

1.4 Studies That Attack the CAPM

1.5 Does proving the CAPM wrong = Market inefficiency? Or, do efficient markets = the CAPM is correct?

1.6 Small Capitalization and Value Stocks

1.7 The Asset Pricing Models of Today

1.8 Chapter-End Problems

References

Chapter 2. Returns-Based Performance Evaluation Models

2.1 Introduction

2.2 Goals, Guidelines, and Perils of Performance Evaluation

2.3 Returns-Based Analysis

2.4 Chapter-End Problems

References

Chapter 3. Returns-Based Performance Measures

3.1 Introduction

3.2 Luck vs. Skill

3.3 The Ultimate Goal of Performance Measures

3.4 Two Non-Regression Approaches

3.5 Regression-Based Performance Measures

3.6 Chapter-End Problems

References

Chapter 4. Portfolio-Holdings Based Performance Evaluation

4.1 Introduction

4.2 Unconditional Holdings-Based Performance Measurement

4.3 Conditional Holdings-Based Performance Measurement

4.4 Chapter-End Problems

References

Chapter 5. Combining Portfolio-Holdings-Based and Returns-Based Performance Evaluation (and the Return Gap)

5.1 Introduction

5.2 Performance-Decomposition Methodology

5.3 Application to U.S. Domestic Equity Mutual Funds

5.4 Empirical Results for U.S. Domestic Equity Mutual Funds

5.5 Results for U.S. Domestic Corporate BOND Mutual Funds

5.6 Appendix A

5.7 Appendix B

5.8 Chapter-End Problems

References

Chapter 6. Performance Evaluation of Non-Normal Portfolios

6.1 Introduction

6.2 Bootstrap Evaluation of Fund Alphas

6.3 Data

6.4 Results for U.S. Equity Funds

6.5 Sensitivity Analysis

6.6 Performance Persistence

6.7 Chapter-End Problems

References

Chapter 7. Fund Manager Selection Using Macroeconomic Information

7.1 Introduction

7.2 A Dynamic Model of Managed Fund Returns

7.3 Empirical Example: U.S. Domestic Equity Fund Data

7.4 Empirical Example: Results for U.S. Domestic Equity Funds

7.5 Chapter-End Problems

Appendix A Description of Mutual Fund Database

Appendix B Investments when fund risk loadings and benchmark returns may be predictable

Appendix C Investments when skills may be predictable

References

Chapter 8. Multiple Fund Performance Evaluation: The False Discovery Rate Approach

8.1 Introduction

8.2 The Impact of Luck on Managed Fund Performance

8.3 An Empirical Example: U.S. Domestic Equity Mutual Funds

8.4 An Empirical Example: Results for U.S. Domestic Equity Funds

References

Chapter 9. Active Management in Mostly Efficient Markets: A Survey of the Academic Literature

9.1 Introduction

9.2 Some Caveats

9.3 Does Active Management Add Value?

9.4 Active Management and Mostly Efficient Markets

9.5 Identifying Superior Active Managers (‘SAM’s)

9.6 Conclusions

9.7 Chapter-End Problems

References

Section 2: Performance Analysis and Reporting

Chapter 10. Basic Performance Evaluation Models

10.1 Basis Formula for the Calculation of Returns

10.2 Geometric Linkage and Scaling of Returns

10.3 Internal Rate of Return

10.4 Time-Weighted Return

10.5 Comparison Between the Time-Weighted Return and the Internal Rate of Return

10.6 Approximation Methods for the Computation of the Time-Weighted Return

10.7 Active Return

10.8 Continuously Compounded Returns

Appendix A Equality between the Time-Weighted Return and the Internal Rate of Return

Appendix B Solving Polynomial Equations for the Determination of Internal Rate of Return

Appendix C Time-Weighted Return and the Unit Price Method

Chapter 11. Indices and the Construction of Benchmarks

11.1 Basic Concepts

11.2 Equity Indices

11.3 Bond Indices

11.4 Money Market Indices

11.5 Peer Group Comparisons and Fund Universes

11.6 Benchmarks for Portfolios Investing in Multiple Asset Classes

11.7 Chapter-End Problems

Chapter 12. Attribution Analysis for Equity Portfolios According to the Brinson Approach

12.1 Introduction to Attribution Analysis

12.2 Single-Period Attribution Analysis According to the Brinson Approach

12.3 Multi-period Attribution Analysis According to Brinson et al.

12.4 Attribution Analysis in a Geometric Form

12.5 Further Aspects of Attribution Analysis

Chapter 13. Attribution Analysis for Fixed Income Portfolios

Appendix: Duration Measures

13.5 Exercises for Chapter-End Problems

Chapter 14. Analysis of Multi-Asset Class Portfolios and Hedge Funds

14.1 Basic Considerations

14.2 Attribution Analysis on Two Levels

14.3 Attribution Analysis on Three Levels

14.4 Implementation in practice

14.5 Risk-Adjusted Attribution Analysis Based On the Systematic Risk

14.6 Risk-Adjusted Attribution Analysis Based on the Information Ratio

14.7 Special Aspects in the Analysis of Hedge Funds

14.8 Chapter-End Problems

Chapter 15. Attribution Analysis with Derivatives

15.1 Attribution Analysis with Derivative-Based Currency Management

15.2 Treatment of Futures and Forwards

15.3 Treatment of Options

15.4 Swaps

15.5 Chapter-End Problems

Chapter 16. Global Investment Performance Standards (GIPS)

16.1 Background

16.2 Definition of Firm

16.3 Creation of Composites

16.4 Determination of Composite Return

16.5 Further Disclosure Requirements for Composite Structure and Sample Presentations

16.6 Maintenance of Composites

16.7 Independent Verification of Compliance with the Standards

16.8 Measurement of the Homogeneity of the Investment Process

16.9 Presentation of Risks according to the GIPS

16.10 Chapter-End Problems

Index

Introduction to the Series

The aim of the Handbooks in Economics series is to produce Handbooks for various branches of economics, each of which is a definitive source, reference, and teaching supplement for use by professional researchers and advanced graduate students. Each Handbook provides self-contained surveys of the current state of a branch of economics in the form of chapters prepared by leading specialists on various aspects of this branch of economics. These surveys summarize not only received results but also newer developments, from recent journal articles and discussion papers. Some original material is also included, but the main goal is to provide comprehensive and accessible surveys. The Handbooks are intended to provide not only useful reference volumes for professional collections but also possible supplementary readings for advanced courses for graduate students in economics.

KENNETH J. ARROW and MICHAEL D. INTRILIGATOR

Copyright

Academic Press is an imprint of Elsevier

The Boulevard, Langford Lane,

Kidlington, Oxford OX5 1GB, UK

225 Wyman Street, Waltham, MA 02451, USA

First edition 2013

Copyright © 2013 Elsevier, Inc. All rights reserved.

SOLNIK, BRUNO, McLEAVEY, DENNIS, GLOBAL INVESTMENTS 6th Edition, © 2009, Reprinted by permission of Pearson Education, Inc., Upper Saddle River, NJ.

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher

Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email: permissions@elsevier.com. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material

Notice

No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is availabe from the Library of Congress

ISBN–13: 978-0-08-092652-0

For information on all Academic Press publications visit our web site at store.elsevier.com

Printed and bound in the US

12 13 14 15 16 10 9 8 7 6 5 4 3 2 1

Preface

This book is intended to be the scientific state-of-the-art in performance evaluation—the measurement of manager skills—and performance attribution—the measurement of all of the sources of manager returns, including skill-based. We have attempted to include the best and most promising scientific approaches to these topics, drawn from a voluminous and quickly expanding literature.

Our objective in this book is to distill hundreds of both classic and the best cuttingedge academic and practitioner research papers into a unified framework. Our goal is to present the most important concepts in the literature in order to provide a directed study and/or authoritative reference that saves time for the practitioner or academic researcher. Sufficient detail is provided, in most cases, such that the investment practitioner can implement the approaches with data immediately, without consulting the underlying literature. For the academic, we have provided enough detail to allow an easy further study of the literature, as desired.

We have contributed in two dimensions in this volume—both of which, we believe, are missing in currently available textbooks. Firstly, we provide a timely overview of the most important performance evaluation techniques, which allow an accurate assessment of the skills of a portfolio manager. Secondly, we provide an equally timely overview of the most important and widely used performance attribution techniques, which allow an accurate measure of all of the sources of investment returns, and which are necessary for precise performance reporting by fund managers.

We believe that our text is timely. An estimated $71.3 trillion was invested in managed portfolios worldwide, as of 2009 (source: www.thecityuk.com). Managing this money, thus, is a business that draws perhaps $700 billion per year in management fees and other expenses for asset managers, in addition to a perhaps similar magnitude in annual trading costs accruing to brokers, market makers, and other liquidity providers (i.e., Wall Street and other financial centers). Our book is the first comprehensive text covering the latest science of measuring the main output of portfolio managers: their benchmarkrelative performance (alpha). Our hope is that investors use these techniques to improve the allocation of their money, and that portfolio management firms use them to better understand the quality of their funds’ output for investors.

We intend this book to be used in at least two ways:

First, as a useful reference source for investment practitioners—who may wish to read only one or a few chapters. We have attempted to make chapters self-contained to meet this demand. We have also included chapter-end questions that both test the reader’s understanding and provide examples of applications of each chapter’s concepts. The audience for this use includes (at least) those studying for the CFA exams; performance analysts; mutual fund and pension fund trustees; portfolio managers of mutual funds, pension funds, hedge funds, and fund-of-funds; asset management ratings companies (e.g., Lipper and Morningstar); quantitative portfolio strategists, regulators, financial planners, and sophisticated individual investors.

Second, the book serves as an efficient way for mathematically advanced undergraduate, masters, or Ph.D. students to undertake a thorough foundation in the science of performance evaluation and attribution. After reading this book, students will be prepared to handle new developments in these fields.

We have attempted to design each chapter of this book to contain enough detail to bring the reader to a point of being able to apply the concepts therein, including the chapter-end problems. In cases where further detail may be needed, we have cited the most relevant source papers to allow further reading.

We have divided our book into two sections:

Part 1 of the book covers the area of performance evaluation.

Chapter 1 provides a short overview of the basics of empirical asset-pricing as applied to performance assessment, including basic factor models, the CAPM, the Fama-French three-factor model and the research on momentum, and the characteristic-based stock benchmarking model of Daniel, Grinblatt, Titman, and Wermers.

Chapter 2 provides an overview of returns-based factor models, and the issues involved in implementing them. Chapter 3 discusses the issue of luck vs. skill in generating investment returns, and presents the fundamental performance evaluation measures, including those based on the Chapter 2 factor models. In addition, extensions of these factor models are introduced that contain factors that capture the ability of portfolio managers to time the stock market or to time securities over the business cycle.

Chapter 4 presents the latest approaches to using portfolio holdings to more precisely measure the skill of a portfolio manager. Chapter 5 provides a complete system for evaluating the skills of a portfolio manager using her portfolio holdings and net returns.

Many managed portfolios generate non-normal returns. Chapter 6 shows how to apply bootstrap techniques to generate more precise estimates of the statistical significance of manager skills in the presence of non-normal returns and alphas.

Chapter 7 covers a very new topic: how to capture the time-varying abilities of a portfolio manager (as briefly introduced in Chapter 3). Specifically, this chapter shows how to predict which managers are most likely to generate superior alphas in the current economic climate.

Chapter 8 also covers a very recent topic in performance evaluation: the assessment of the proportion of a group of funds that are truly skilled using only their net returns. This approach is very useful in assessing whether the highest alpha managers are truly skilled, or are simply the luckiest in a large group of managers.

Finally, Chapter 9 is a capstone chapter, in that it provides an overview of the research findings that use the principles outlined in the first 8 chapters. As such, it is a very useful summary of what works (and what does not) when looking for a superior asset manager (a SAM) and trying to avoid an inferior asset manager (an IAM).

Part 2 of the book primarily concerns performance attribution and related topics.

Since attribution analysis has become a crucial component within the internal control system of investment managers and institutional clients, ample space is dedicated to a thorough treatment of this field. The focus in this part lies on the practical applications rather than on the discussions of the various approaches from an academic point of view. This (practitioner’s) approach is accompanied by a multitude of examples derived from practical experience in the investment industry. Great emphasis was also put on the underlying mathematical detail, which is required for an implementation in practice.

Chapter 10 provides an overview of the basic approaches for the measurement of returns. In particular, the concepts of time-weighted return and internal rate of return, as well as approximation methods for these measures are discussed in detail.

Attribution analysis, in practice, requires a deep understanding of the benchmarks against which the portfolios are measured. Chapter 11 provides an introduction to the benchmarks commonly used in practice, and their underlying concepts.

Chapter 12 covers fundamental models for the attribution analysis of equity portfolios developed by Gary Brinson and others. Furthermore, basic approaches for the treatment of currency effects and the linkage of performance contributions over multiple periods are considered.

Chapter 13 contains an introduction to attribution analysis for fixed income portfolios from a practitioner’s point of view. The focus lies on a methodology that is based on a full valuation of the bonds and the option-adjusted spread. In addition, various other approaches are described.

Based on the methodologies for equity and fixed income portfolios, Chapter 14 presents different methodologies for the attribution analysis of balanced portfolios. This chapter also illustrates the basic approaches for a risk-adjusted attribution analysis and covers specific aspects in the analysis of hedge funds.

Chapter 15 describes the various approaches for the consideration of derivatives within the common methodologies for attribution analysis.

The final chapter (Chapter 16) deals with Global Investment Performance Standards, a globally applied set of ethical standards for the presentation of the performance results of investment firms.

The authors are indebted to many dedicated academic researchers and tireless practitioners for many of the insights in this book. Professor Wermers wishes to thank the many investment practitioners that have provided data or insights into the topics of this book, including through their professional investment management activities: Robert Jones of Goldman Sachs Asset Management (now at System Two and Arwen), Rudy Schadt of Invesco, Scott Schoelzel and Sandy Rufenacht of Janus (now retired, and at Three Peaks Capital Management, respectively), Bill Miller and Ken Fuller of Legg-Mason, Andrew Clark, Otto Kober, Matt Lemieux, Tom Roseen, and Robin Thurston of Lipper, Don Phillips, John Rekenthaler, Annette Larson, and Paul Kaplan of Morningstar, Sean Collins and Brian Reid of the Investment Company Institute.

Professor Wermers also wishes to thank all of the classes taught on performance evaluation and attribution since 2001—at Chulalongkorn University (Bangkok); the European Central Bank (Frankfurt); the Swiss Finance Institute/FAME Executive Education Program (Geneva); Queensland University of Technology (Brisbane); Stockholm University; the University of Technology, Sydney; and the University of Vienna. Special thanks are due to students in that first class of the SFI/FAME program during those dark days in September 2001, 10 days after the 9-11 attacks.

Professor Wermers is also indebted to his loving family, Johanna, Natalie, and Samantha, for the endless hours spent away from them while preparing and teaching this subject. He gratefully acknowledges Thomas Copeland and Richard Roll of UCLA and Josef Lakonishok of University of Illinois (and LSV Asset Management) for early inspiration, as well as Wayne Ferson, Robert Stambaugh, Lubos Pastor, and Mark Carhart for their recent contributions to the field. In addition, he owes his career to the brilliant mentoring of Mark Grinblatt and Sheridan Titman at UCLA, pioneers in the subject of performance evaluation. This text would not have been possible from such humble beginnings without their selfl ess support and guidance.

Dr. Fischer is indebted to his colleagues at IDS GmbH—Analysis and Reporting Services, an international provider of operational investment controlling services. Over the past years he has greatly benefited from numerous discussions surrounding practical applications.

Thanks are also due to Dr. Fischer’s former team members at Cominvest Asset Management GmbH. The design and the implementation of a globally applicable attribution software from scratch, and the implementation of the Global Investment Performance Standards were exciting experiences which left their mark on the current treatise.

He also wishes to thank various colleagues (Markus Buchholz, Detlev Kleis, Ulrich Raber, Carsten Wittrock, and others), with whom he co-authored papers in the past. Several sections in this book are greatly indebted to the views expressed there.

Dr. Fischer is also indebted to the CFA institute and the Global Investment Performance Committee for formative discussions surrounding the draft of the GIPS in 1998/1999 and during his official membership term from 2000 to 2004.

Both authors wish to thank J. Scott Bentley of Elsevier, whose vision it was to create such a book, and whose patience it took to see it through.

To those whose contributions we have overlooked, our sincere apologies; such an ambitious undertaking as condensing a huge literature necessitates that the authors choose topics that are either most familiar to us or viewed by us as most widely useful. Surely, we have missed some important papers, and we hope to have a chance to create a second edition that expands on this one.

Finally, to the asset management practitioner: we dedicate this volume to you, and hope that it is useful in furthering your goal of providing high-quality investment management services!

Section 1: Performance Evaluation

Chapter 1 An Introduction to Asset Pricing Models

Chapter 2 Returns-Based Performance Evaluation Models

Chapter 3 Returns-Based Performance Measures

Chapter 4 Portfolio-Holdings Based Performance Evaluation

Chapter 5 Combining Portfolio-Holdings-Based and Returns-Based Performance Evaluation (and the Return Gap)

Chapter 6 Performance Evaluation of Non-Normal Portfolios

Chapter 7 Fund Manager Selection Using Macroeconomic Information

Chapter 8 Multiple Fund Performance Evaluation: The False Discovery Rate Approach

Chapter 9 Active Management in Mostly Efficient Markets: A Survey of the Academic Literature

Chapter 1

An Introduction to Asset Pricing Models

Abstract

This chapter provides a brief overview of asset pricing models, with an emphasis on those models that are widely used to describe the returns of traded financial securities. Here, we focus on various models of stock returns and fixed-income returns, and discuss the reasoning and assumptions that underlie the structure of each of these models.

Keywords

Asset Pricing Models, CAPM, Factor Models, Fama French three-factor model, Carhart four-factor model, DGTW stock characteristics model, Estimating beta,Expected return and risk.

1.1 Historical Asset Pricing Models

Individuals are born with a sense of the perils of risk, and they develop mental adjustments to penalize opportunities that involve more risk.¹ For example, farmers do not plant corn, which requires a great deal of rainfall (which may or may not happen), unless the expected price of corn at harvest time is sufficiently high. Currency traders will not take a long position in the Thai baht and short the U.S. dollar unless they expect the baht to appreciate sufficiently. In essence, the farmer and the currency trader are each applying a personal discount rate to the expected return of planting corn or investing in baht. The farmer’s discount rate depends on his assessment of the risk of rainfall (which greatly affects his total corn crop output) and the risk of a price change in the crop. The currency trader’s discount rate depends on the relative economic health of Thailand and the U.S., and any potential government intervention against currency speculation—both of which may carry large risks. Both economic agents" discount rates also depend on their personal aversion to risk, and, thus, may require very different compensations to take similar risks.², ³

Asset managers and investors also understand that some securities are less certain in their payouts than others, and make adjustments to their investment plans accordingly. Short-maturity bank certificates of deposit (CDs), while paying a very low annual interest rate, are attractive because they return the principal fairly quickly and guarantee (with insurance) a particular rate-of-return. Stocks, with not even a promise that they will pay the next quarterly dividend, provide much higher returns than CDs, on average. In general, greater levels of risk in a security or security portfolio—especially those risks that cannot be inexpensively insured—require compensation by risk-averse investors in the form of higher potential future returns.

The most basic approach to an asset pricing model that describes the compensation to investors for risk-taking simply ranks securities by the standard deviation of their periodic (say, monthly) returns, then conjectures a particular functional relation between this risk and the expected (average future) returns of securities.⁴ But, should the relation be linear or non-linear between standard deviation and expected return? Should there be any credit given to securities that have counter-cyclical risk patterns (i.e., high returns during recessions)? How can we account for offsetting risk patterns between a group of securities, even within a bull market (e.g., technology vs. utility stocks)? Should risk that can be diversified by holding many different investments be rewarded? These questions are the focus of modern asset pricing theory.

The foundations of modern asset pricing models attempt to combine a few very basic and simple axioms that appear to hold in society, including the following. First, that investors prefer more wealth to less wealth. Second, that investors dislike risk in the payouts from securities because they prefer smooth patterns of consumption of their wealth, and not feast or famine periods of time. And, third, that investors should not be rewarded with extra return for taking on risk that could be avoided through a smart and costless approach to mixing assets. Our next sections briefly describe the most widely used asset pricing models of today. In discussing these models, we focus on their application to describe the evolution of returns for liquid securities—chiefly, stocks and bonds.⁵ However, the usefulness of these models—with some modifications—goes far beyond stocks and bonds to other securities, such as derivatives and less liquid assets such as private equity and real estate.

1.2 The Beginning of Modern Asset Pricing Models

A great deal of work has been done, over the past 60 years, to advance the ability of statistical models to explain the returns on securities. Building on Markowitz’s (1952) seminal work on efficient portfolio diversification, Sharpe published his famous paper on the capital asset pricing model (CAPM) in 1964 (Sharpe, 1964).⁶ These two ideas shared the 1990 Nobel Prize in Economics.

, is a linear function of the systematic (or market-related) risk of a stock or portfolio, β:

(1.1)

where Rt =  security or portfolio return minus riskfree rate, RMRFtis a measure of correlation of the security or portfolio with the broad market portfolio.⁷

This relation is extremely simple and useful for relating the reward (expected return) that is required of a stock with its level of market-based risk. For instance, if market-based risk (β) is doubled, then expected return, in excess of the riskfree rate, must be doubled for the security or portfolio to be in equilibrium with the market. If T-bills pay 2%/year and a stock with a beta of one promises an average return of 7%, then a stock with a beta of two must promise an average of 12%.

Sharpe’s CAPM is simple and is an equilibrium theory, but it depends on several unrealistic assumptions about the economy, including:

1. All investors have the exact same information about possible future expected earnings and their risks at each point in time.

2. Investors are risk-averse and behave perfectly rationally, meaning they do not favor one type of security over another unless the calculated Net Present Value of the first is higher.

3. The cost of trading securities is zero.

4. Investors are mean-variance optimizers (it is sufficient, but not necessary, for this requirement that security returns are normally distributed).

5. All investors are myopic, and care only about one-period returns.

6. Investors are price-takers, meaning that their actions cannot influence prices of securities.

7. There are no taxes on holding or trading securities.

8. Investors can trade any amount of an asset, no matter how small or large.

Several of these assumptions may not fit real-world markets, and many papers have attempted, with some—but far from complete—success in extending the CAPM to situations which eliminate one or more of these assumptions. Among these papers are Merton’s (1973) intertemporal CAPM (ICAPM), which extends the CAPM to a multiperiod model (to address #5). A good discussion of these extended CAPMs can be found in several investments textbooks, such as Elton et al. (2009).

While there are many extensions of the CAPM that deal with dropping one assumption at a time, it is not at all clear that dropping several assumptions simultaneously still results in the CAPM being a good model that describes the relation of returns to risk in real financial markets. Because of this, recent work has focused on building practical models that work with data, even if they are not based on a particular theoretical derivation. Although many attempts have been made, with some success, at creating a new model of asset pricing, no theory has become as universally accepted as the CAPM once was. Hopefully, some future financial economist will create such a new model that reflects real financial markets well. In the meantime, we must rely on either empirical applications of the CAPM, or on other models that have no particular equilibrium theory supporting them.

1.2.1 Estimating the CAPM Model

, and β, so we must estimate them somehow from data. This is where a time-series version of the CAPM (also called the Jensen model (Jensen, 1968)) can be used on return data for a security or a portfolio of securities. The time-series version of the CAPM can be written as

(1.2)

while its application to real-world data can be similarly written as:⁸

(1.3)

where we estimate the parameters α (the model intercept) and β . (This model is more generally called the single-factor model, as it does not require that the CAPM is exactly correct to be implemented on real-world data.) A widely used method for doing this is ordinary least squares (OLS), which fits the data with estimated values of α and β, such that the sum of the squared residuals from the fitted OLS regression line is minimized. Note that Equation estimate. In this discussion, we’ll stick with the model including an intercept to accommodate such issues.

After we estimate the model, we write the resulting fitted model as

(1.4)

is just a temporary deviation, and we expect it to be zero in the future. Using this expectation, we can use this model to forecast future returns with:

(1.5)

. One simple, but not very precise, method of estimating this parameter is to use the average historical values over the past T periods:⁹

include using the average return forecast from professionals, such as security analysts, or deriving forecasts from index futures or options markets.

We can also estimate the risk of holding a stock or portfolio—as well as decomposing this risk into market-based and idiosyncratic risk—with this one-factor model by applying the rules of variances to Equation (1.2):

(1.6)

to estimate the future total risk:

(1.7)

Figure 1.1 and Tables 1.1 and 1.2 show an example of a fitted model using Chevron-Texaco (CVX) over the 2007–2008 period. Two approaches to fitting the model of Equation (1.3) using OLS are presented in the graph and in the tables: (1) the unrestricted model, and (2) the restricted model (where α is forced to equal zero):

Figure 1.1 CAPM Regression Graph for Chevron-Texaco.

Table 1.1 Unrestricted Ordinary Least Squares CAPM Regression Output for Chevron-Texaco

Table 1.2 Restricted Ordinary Least Squares CAPM Regression Output for Chevron-Texaco

, since we force the fitted regression line to pass through zero, as shown in the figure above.

In most cases, it is better to allow the intercept to be estimated, since it can be non-zero by the randomness in stock returns, as illustrated by the Apple example discussed previously.

Next, let’s model CVX over the following two years, 2009–2010, shown in Table 1.3.

Table 1.3 Unrestricted Ordinary Least Squares CAPM Regression Output for Chevron-Texaco, 2009–2010

have changed from their values during 2007–2008. Does this mean that these parameters actually change quickly for individual stocks? In most cases, no—these changes are the result of estimation error, which happens when we have a very noisy (volatile) y-variable, such as CVX monthly returns,¹⁰ due again to randomness.

Besides using the above regression output in the context of Equation (1.5) to estimate the expected (going-forward) return of CVX, we can also use the regression output to estimate risk for CVX going forward, using Equation (1.6).

The results from the above two regression windows point out an important lesson to remember: individual stock betas are extremely difficult to estimate precisely, which makes the CAPM very difficult to use in modeling individual stocks. There are several ways to attempt to correct these estimated betas while still using the CAPM. One important example is a correction for stocks that respond slowly to broad stock market forces, and might have a lag in their reaction due to their illiquidity. Scholes and Williams (1977) describe an approach to correct for the betas of these stocks by adding a lagged market factor to the CAPM regression,

(1.8)

(assuming rmrft has trivial serial correlation):

(1.9)

There are many other potential problems with estimated betas, and numerous approaches to dealing with them. However, none of these methods, many of which can be complicated to implement, fully correct for the problem of large estimation errors for individual securities, such as stocks.¹¹ As a result, one should always be very careful about modeling an individual security. When possible, form portfolios of securities, then apply regression models.

1.3 Efficient Markets

The notion of market prices efficiently reflecting all available (public) information is likely as old as the notion of capitalism itself. Indeed, if prices swing wildly in a way that is not consistent with the (unknown) expected intrinsic value of assets, then a case can be made for government intervention. Examples of this are the two rounds of quantitative easing (QE1 and QE2) that were implemented during 2009 and 2010, during and shortly after the financial crisis of 2008 and 2009.¹²

However, there are many shades of market efficiency, from completely informationally efficient markets to markets that are only somewhat informationally efficient.¹³ In the world around us, we can easily see that many forms of information are fairly cheap to collect (such as announcements from the Federal Reserve), while many other forms are expensive (such as buying a Bloomberg terminal with all of its models). In their seminal paper, Grossman and Stiglitz (1980) argued that, in a world of costly information, informed traders must earn an excess return, or else they would have no incentive to gather and analyze information to make prices more efficient (i.e., reflective of information). That is, markets need to be mostly but not completely efficient, or else investors would not make the effort to assess whether prices are fair. If that were to happen, prices would no longer properly reflect all available and relevant information, and markets would lose their ability to allocate capital efficiently. Thus, Grossman and Stiglitz advocate that markets are likely Grossman-Stiglitz efficient, which means that costly information is not immediately and freely reflected in prices available to all investors. Indeed, the idea of Grossman-Stiglitz efficient markets is a very useful way for students to view real-world financial markets.

Behavioral finance academics, such as John Campbell and Robert Shiller, have found evidence that markets do not behave as if investors are perfectly rational in some Adam Smith invisible hand sense—in fact, they believe the evidence makes the potential for efficient markets—Grossman-Stiglitz or other notions of efficiency—very improbable in many areas of financial markets. This evidence is somewhat controversial among academics, although investment practitioners seem to have accepted the idea of behavioral finance more completely than academics. While the field of behavioral finance has become immense, a full discussion of the literature is beyond the scope of this book.¹⁴ However, in the next section, we will discuss some research that documents return anomalies—potentially driven by investor misbehaviors—that are directly related to the models used to describe stock and bond returns today—so that the reader will have a better understanding of the origin of these models.¹⁵

1.4 Studies That Attack the CAPM

Many financial economists during the 1970s attempted, with some success, to criticize the CAPM as a model that doesn’t reflect the real world of stock returns and risk. The reader should note that no one doubted that the mathematics of the CAPM were correct, given its many assumptions. Instead, the model was attacked because it did not work well in the real world of stock, bond, and other security and asset pricing, which means its assumptions were not realistic.

A few of the many famous papers are described here. Most CAPM criticisms have focused on the stock market, mostly because stock price and return data have been studied extensively by academic researchers and such data are of high-quality (i.e., from the Center for Research in Security Prices–CRSP—at the University of Chicago).

First, Banz (1981) studied the returns of small capitalization stocks using the CAPM model. Banz found that a size factor (one that reflects the return difference between stocks with low equity capitalization—price times shares outstanding—and stocks with high equity capitalization) adds explanatory power for the cross-section of future stock returns above the explanatory power of market betas. He finds that average returns on small stocks are too high, even controlling for their higher betas, and that average returns on large stocks are too low, relative to the predictions of the CAPM.

Bhandari (1988) found a positive relation between financial leverage (debt to equity ratio) and the cross-section of future stock returns, even after controlling for both size and beta. Basu (1983) finds that the earnings-to-price ratio (E/P) predicts cross-sectional differences in future stock returns in models that include size and beta as explanatory variables. High E/P stocks outperform low E/P stocks.

Keim (1983) finds that about 50% of the size factor return, during 1963–1979, occurs in January. Further, over 50% of the January return occurs during the first week of trading, in particular, the first trading day. And, Reinganum (1983) finds similar results, and also finds that this January effect does not appear to be completely explained by investor tax-loss selling in December and repurchasing in January.

1.5 Does proving the CAPM wrong = Market inefficiency? Or, do efficient markets = the CAPM is correct?

Emphatically, no! This is often termed the joint hypothesis problem, since any empirical test of the CAPM, such as the above-cited studies, is jointly testing the validity of the model and whether violations to the model can be found. Often, students of finance believe in the CAPM so thoroughly (probably through the fault of their professors) that they equate the CAPM’s validity to the validity of efficient markets. However, there is no such tie. Markets can be perfectly efficient, and the CAPM model can simply be wrong—it’s just that it does not describe the proper risk factors in the economy. For instance, if two risk factors drive the economy, then the CAPM will not work.

If the CAPM is exactly correct, however, markets must be efficient—unless we use an expanded notion of the CAPM that has two versions: one version that is visible to everyone, and another that is visible only to the informed investors. The CAPM modeled by Sharpe, however, has no such duality—there is one market portfolio and one beta for each security in the economy. In Sharpe’s CAPM world, markets are perfectly efficient, and everyone has the same information.¹⁶

1.6 Small Capitalization and Value Stocks

In the early 1990s, Fama and French tried to settle the question of the usefulness of the CAPM in the face of all these apparent stock anomalies. In doing so, Fama and French (FF; 1992) declared that beta is dead, meaning that the CAPM was a somewhat useless model, at least for the stock market. Instead, FF promoted the use of two new factors to model the difference in returns of different stocks: the market capitalization of the stock (also called size) and the book-to-market ratio (BTM) of the stock—that is, the accounting book value of equity divided by the market’s value of the equity (using the traded market price).

FF used a clever approach to demonstate this argument. Most prior studies of the CAPM first estimate individual stock or stock portfolio betas from the one-factor regression of Equation (1.4), as we did for CVX above, then test whether these betas forecast future stock returns. FF argued that small capitalization stocks tend to have much higher betas than large capitalization stocks, so it might be that small stocks simply have higher returns than large stocks, regardless of their betas.

First, FF estimated each stock’s beta with five years (60 months) of past returns, using the one-factor regression model of Equation (1.3). Then, they ranked all stocks by their market capitalization (size), from largest to smallest, then cut these ranked stocks into 10 groups. The top decile group was the group of largest stocks, while the bottom decile was the small stock group.

Next, FF ranked stocks—within each of these decile groups—by the betas of the stocks that they had already computed. Then, FF took the highest 1/10th of stocks, according to their betas, from each of the 10 size deciles (that 1/10th was 1/100 of all stocks)—then, recombined these 10 high beta subportfolios into a high beta, mixed size portfolio. This was repeated for the 2nd highest 1/10th of stocks in each portfolio to form the 2nd highest beta subportfolio with mixed size. And, so on, to the lowest beta 1/10th of stocks to form the low beta subportfolio with mixed size. Finally, FF measured the equal-weighted returns of each of these newly constructed 10 portfolios—each of which had stocks with similar betas, but mixed size—during the following 7 years. The objective was to separate the influence of size from beta by mixing the size of stocks with similar betas. This procedure is depicted in Figure 1.2.

Figure 1.2 Fama-French’s Beta is Dead Slicing Test.

When FF regressed this 7-year future return, cross-sectionally, on the prior equal-weighted betas of these 10 portfolios, they found no significant relation, where the CAPM’s central prediction is a strong and positive relation between betas and returns. Thus, according to FF, beta was dead.¹⁷ Then, FF presented evidence that not only does size work well, but so does BTM ratio; together, they both worked well, so they appear to be measuring different risks. Finally, FF looked at the return-on-equity (ROE) of small stocks and stocks with a high BTM ratio, and found that the ROE of these stocks was quite low—indicating, perhaps, that they are under financial distress and are at risk of bankruptcy. While not proving anything, FF suggest that size and BTM may be a proxy for financial distress—small stocks with high BTM, for instance, are highly stressed—and this may underlie the usefulness of size and BTM. Simply put, investors demand higher returns for financially distressed stocks, as they are more likely to fail together during a recession.

The reception of the Fama French paper was one of controversy, which still exists today. Most reseachers have admitted that Fama and French are right about what works better in the real world of stocks, but they disagree about why. FF represent one camp with their rational investor, financial distress risk economic story. Another camp believes that investors exhibit behavioral tendencies that color their choice of stocks. Underlying this economic story is the fact that individuals tend to overreact to longer-term trends in the economic fortunes of a corporation, and that they believe that the fortunes of stocks that have become less profitable over the past several years will continue to become worse—thus, they put sell pressure on small stocks and value stocks (high BTM stocks). A third camp believes that small stocks and value stocks have simply gone through a lucky streak, and that we should not place too much importance on the experience of U.S. stocks in the past few decades.

In an attempt to further test the FF findings, Griffin (2002) studied size and book-to-market as stock return predictors in the U.S., Japan, the U.K., and Canada. He found evidence in all four countries that size and BTM forecast stock returns, consistent with FF’s findings in U.S. stocks. However, he also found that returns correlate poorly for size and BTM across these countries, which could be evidence that they are risk-based or that they are due to irrational investor behavior—and country stock markets are segmented, preventing investors from arbitraging across differences in these factor returns across countries.

1.6.1 Momentum Stocks

Notably, Fama and French did not quite find all the important factors that drive stock returns. Jegadeesh and Titman (JT; 1993) found that momentum, measured as the one year past return of a stock is an important predictive variable for the following year’s return. In fact, a simple sorting of stocks on their one-year past return, followed by an equal-weighted long position in the top 10% winners and a short position in the bottom 10% losers of last year provides an arbitrage profit of almost 1% per month (i.e., about 10% during the following year).¹⁸Figure 1.3 illustrates the profitability over numerous portfolios formed over the period 1965–1989. The monthly (not annualized) returns of the long-short portfolio over the 36 event months following the portfolio formation are shown first, followed by the cumulated monthly returns over the same 36 months.¹⁹

Figure 1.3 Monthly and Cumulative Momentum Long/Short Portfolio Returns, 1965–1989.

Further evidence supporting momentum in U.S. stocks was found during 1941–1964, although not quite as strong—shown in Figure 1.4.

Figure 1.4 Monthly and Cumulative Momentum Long/Short Portfolio Returns, 1941–1964.

However, JT found that the depression era did not support their momentum theory, and, instead, momentum stocks lost considerable money (see Figure 1.5).

Figure 1.5 Monthly and Cumulative Momentum Long/Short Portfolio Returns, 1927–1940.

JT explained that momentum likely did not work during the depression era because of inconsistent monetary policy that artificially created reversals of stock returns during that time. Specifically, when the stock market dropped, the Fed eased monetary policy, and when it boomed, the Fed strongly tightened. Nevertheless, Daniel (2011) has shown, more recently, that momentum stocks outperformed during 1989–2007, but underperformed (badly) during the financial crisis of 2008–2009.

Further research by Rouwenhorst (1998) found that momentum exists in stocks in Europe, but not in Asia. More recent research seems to find momentum even in Japan (see Asness(2011)).

Today, although the evidence is, at times, inconsistent, momentum is strong enough that most academic researchers appear to accept that it is a reality of markets. One economic explanation of momentum is that investors underreact to short-term news about companies, such as improving earnings or cashflows. Thus, a stock that rises this year has a bright future next year—again, not always, but on average.²⁰

Finally, Griffin et al. (GMJ; 2003) examined momentum in the U.S. and 39 other countries, and found evidence that these factors work well in these markets, but that momentum across different countries is only weakly correlated. Therefore, country-level momentum factors work better in capturing momentum, rather than a global momentum factor across all countries. This finding suggests that whatever economics are at play in the risk of stocks, they work a little differently in different countries, but with the same overall result: small stocks outperform large stocks, value stocks outperform growth stocks, and momentum stocks outperform contrarian stocks (all of this is for an average year, but the reverse can occur for any single year or subset of years—such as the superior growth stock returns of the technology boom during the 1990s). Finally, GMJ found that momentum profits tend to reverse in the countries over the following one to four years.

Next, we will describe models that attempt to capture the multiple sources of stock returns noted above. While academics and practitioners do not agree on whether these sources of additional return represent systematic risks or simply return anomalies, these models have been developed to better describe the drivers of stock returns, regardless of the source of the factors" power.²¹

1.7 The Asset Pricing Models of Today

The above studies have inspired researchers to add factors to the single-factor model of Equation (1.2) that is, itself, inspired by the CAPM theory. As opposed to this theory-inspired single-factor model, almost all recent models are empirically inspired, which means that they are chosen because they explain the cross-section and/or time-series of security returns while still making economic sense. This means that we don’t simply try lots of factors until we find some that work, as this can always be done (and often leads to a breakdown of the model when we try to use it with other data). We carefully examine past research for both economic and econometric guidance on the factors that might be used in a model. Fortunately, many researchers have already done this work for us. Almost all models are multifactor models, meaning that more than one x-variable (risk factors) is used to predict the y-variable (security or portfolio excess returns).

1.7.1 Introduction to Multifactor Models

A multifactor model can be visualized as a simple extension of a single factor model, such as the CAPM. However, by using multiple risk factors, we are implicitly rejecting the CAPM and its many assumptions about investors and markets.

The simplest multifactor model is a two-factor model. Let’s suppose that we believe that, in addition to the broad stock market, the risk-premium to investing in small stocks drives security returns.

Then, the time-series model would be:

(1.10)

is the exposure of a security, or portfolio, to the small-capitalization risk-factor. This regression for Chevron-Texaco, implemented using Excel during the 24-month period January 2009 to December 2010, results in the following output Table 1.4.

Table 1.4 Two-Factor Regression for CVX

from the single-factor regression of CVX excess returns on RMRF (from a prior section) is 0.51.²² Therefore, in this case, the addition of a small-cap factor—to which Chevron-Texaco is negatively correlated—does not matter much. However, since we have estimated the two-factor model, and since its t-statistic is relatively close to −1.645 (the two-tailed critical value for 10% significance), we’ll use it.

Once the model is fitted, the next-period estimated expected return is:

(1.11)

or, using the fitted regression from above,

(1.12)

, from the model-predicted values of Equation (1.12).

The next-period estimated total risk, which contains a term for the covariance between RMRF and SMB, is

(1.13)

and the next-period estimated systematic (risk-factor related) risk is:

(1.14)

Again, following the simple approach of using historical sample data to estimate the above expected returns and variances, the equations for expected return, total, and systematic-only risk become

(1.15)

(1.16)

(1.17)

,

.²³

1.7.2 Models of Stock Returns

Regression-Based Models Fama and French (1993) designed a widely used multifactor model which adds both the small-capitalization factor (SMB) and a value stock factor (HML) to the single-factor model of Equation (1.2).²⁴

(1.18)

However, the most widely used returns-based model for analyzing equities is the four-factor model of Carhart (1997),

(1.19)

to the three-factor model of Fama and French.²⁵

Let’s estimate the Carhart model for CVX, during 2009–2010 in Table 1.5.

Table 1.5 Four-Factor Regression for CVX

.

Also, the four-factor model shows that RMRF and UMD are the most statistically significant explanatory variables, with SMB close behind. HML has no significance, since its p-value is equal to 99% (meaning that the chances of observing a coefficient of |0.00028| or larger by pure randomness, when its actual value is zero, is 99%). So, we conclude that CVX, during 2009–2010, has a beta close to 1 (typical for a stock), is a very large capitalization stock (since its loading on SMB is very negative and statistically significant), and it has significant momentum (meaning the prior-year return is high over the period 2009–2010—consistent with increasing oil prices!). Note that, in general, coefficients in this model that are close to (or slightly exceed) one have a large exposure to that risk factor. However, even coefficients at the level of 0.2 or 0.3 indicate a substantial exposure to a certain risk factor.

A Stock Characteristic-Based Model Another approach to modeling stocks that is based on the findings noted above (i.e., that market capitalization, value, and momentum drive stock returns) uses the characteristics (observable features) of stocks to assemble them into groups or portfolios of stocks with similar characteristics. Daniel and Titman (1997) found empirical evidence that suggests that characteristics provide better ex-ante forecasts than regression models of the cross-sectional patterns of future stock returns. This evidence indicates that stock factors like equity book-to-market ratio at least partially relate to future stock returns due to investors having behavioral biases against certain types of stocks (e.g., those stocks with recent bad news, which pushes the BTM ratio up too much).

Following Daniel and Titman, in the characteristic benchmarking approach, the average return of the similar characteristic portfolio is used as a more precise proxy for the expected return of the stock during the same time period. Any deviation of a single stock from this expected return is the stock’s residual, or unexpected return. Daniel et al. (1997) developed such an approach for U.S. equities, and many other researchers have replicated their approach in other stock markets.

First, all stocks (listed on NYSE, AMEX, or Nasdaq) having at least two years of book value of equity information available in the Compustat database, and stock return and market capitalization of equity data in the CRSP database, are ranked, at the end of each June, by their market capitalization. Quintile portfolios are formed (using NYSE size quintile breakpoints), and each quintile portfolio is further subdivided into book-to-market quintiles, based on their most recently available fiscal year-end book-to-market data as of the end of June of the ranking year.²⁶ Here, we industry-normalize the book-to-market ratio, since we would like to classify stocks by how much they deviate from their industry norms.²⁷,²⁸ Finally, each of the resulting 25 fractile portfolios are further subdivided into quintiles based on the 12-month past return of stocks through the end of May of the ranking year. This three-way ranking procedure results in 125 fractile portfolios, each having a distinct combination of size, book-to-market, and momentum characteristics.²⁹ The three-way ranking procedure is repeated at the end of June of each year, and the 125 portfolios are reconstituted at that date.

Figure 1.6 illustrates this process.

Figure 1.6 Daniel, Grinblatt, Titman, and Wermers stock benchmarking procedure.

A modification of this procedure is to reconstitute these portfolios at the end of each calendar quarter, rather than only once per year on June 30, using updated size, BTM, and momentum data. While the annual sort is closer to an implementable strategy that is an alternative to holding a particular stock, the quarterly sort allows us to more accurately control for the changing characteristics of the stock. For example, the momentum, defined as the prior 12-month return of a stock, can change quickly.

Value-weighted returns are computed for each of the 125 fractile portfolios, and the benchmark for each stock during a given quarter is the buy-and-hold return of the fractile portfolio of which that stock is a member during that quarter. Therefore, the benchmark-adjusted return for a given stock is computed as the buy-and-hold stock return minus the buy-and-hold value-weighted benchmark return during the same quarter.

1.7.3 Models of Bond Returns

Fama and French (1993) found a set of five risk factors that worked well in modeling both stock and bond returns. This includes three stock market factors and two bond market factors:

,

,

3. value factor (high book-to-market stock return minus low BTM stock return) (HML),

, and

.

It is very important to note that Fama and French (1993) modeled the time-series of returns on stocks and bonds, where Fama and French (1992) modeled the cross-sectional (across-stock) differences in returns on stocks—which is why the stock market return is included in the above group, but not in the 1992 paper’s factors. In essence, the 1992 paper says that we can assume that beta=1 for all stocks (without a huge amount of error), and, therefore, beta only affects stock returns over time. There is no difference in different stock returns at the same period of time, since they all have assumed betas of one, according to Fama and French (1992).

Fama and French (1993) also find that stock and bond returns are linked together through the correlation of the stock market return with the return on the two bond factors. Interestingly, a large body of other research since then, including Kandel and Stambaugh (1996) has found that broad macroeconomic factors, including the two bond factors noted above, help to forecast the stock market return.

Gruber, Elton, Agrawal, and Mann (2001) find that the three stock risk factors above (1–3) are also useful in modeling corporate bonds—in addition to exposure to potential default and taxation of bond income. Finally, Cornell and Green (1991) find that stock market returns are even more important than government bond market yields in modeling high-yield (junk) bonds.

The above research on bond markets suggest that a five-factor model should be used to model bonds:

(1.20)

Note that there is no momentum factor for bond markets, although some recent papers have also challenged this.

1.8 Chapter-End Problems

1. Download the monthly returns for Exxon-Mobil (XOM) during 2009 and 2010 from CRSP, Yahoo Finance, or another source. Also, download the 30-day Treasury Bill return and the monthly factor returns for RMRF, SMB, HML, and UMD from Ken French’s website, http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/.

A. Using Excel or a statistics package, run a single-factor linear regression (ordinary least squares) for XOM (the y-variable is the excess return of XOM, which is the XOM return minus T-Bill return, while the x-variable is the monthly return on RMRF). How does your regression output compare with that of CVX shown in this chapter—what are the differences in the two stocks according to this output?

B. Repeat, using a two-factor model that includes RMRF and SMB. How does your regression output compare with that of CVX shown in this chapter—what are the differences in the two stocks according to this output?

C. Repeat, using the Carhart four-factor model. How does your regression output compare with that of CVX shown in this chapter—what are the differences in the two stocks according to this output?

2. Download monthly returns for Apple (AAPL) during 2009 and 2010, and run a single-factor regression on the S&P 500 as the market factor. What are the resulting alpha and beta?

3. Using the AAPL data from problem #2, run a four-factor model. What are the coefficients on each factor, and what do they tell you about Apple’s stock?

4. Starting with the model of Equation (1.2), derive the risk model shown by Equation (1.6).

5. Starting with the model of Equation (1.10), derive the risk model shown by Equation (1.13).

6. Describe the empirical tests that find violations of the CAPM in stock returns.

7. Describe the empirical approach that Fama and French (1992) used to find that beta is dead.

8. Discuss each of the assumptions of the CAPM. For each assumption, provide some brief evidence from financial markets that indicates that the assumption may not be correct.

9. Suppose that an institution holds Portfolio K. The institution wants to