Florida State University Libraries
Electronic Theses, Treatises and Dissertations The Graduate School
2007 Market Efficiency and Market Anomalies: Three Essays Investigating the Opinions and Behavior of Finance Professors Both as Researchers and as Investors Colbrin (Colby) A. Wright
Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] THE FLORIDA STATE UNIVERSITY
COLLEGE OF BUSINESS
MARKET EFFICIENCY AND MARKET ANOMALIES:
THREE ESSAYS INVESTIGATING THE OPINIONS AND BEHAVIOR OF FINANCE PROFESSORS BOTH AS RESEARCHERS AND AS INVESTORS
By
COLBRIN (COLBY) A. WRIGHT
A Dissertation submitted to the Department of Finance in partial fulfillment of the requirements for the degree of Doctor of Philosophy
Degree Awarded: Summer Semester, 2007
The members of the Committee approve the dissertation of Colbrin A. Wright defended on May 22, 2007.
______David Peterson Professor Directing Dissertation
______Michael Brady Outside Committee Member
______Gary Benesh Committee Member
______James Doran Committee Member
Approved:
______William Christiansen, Chair, Department of Finance
______Caryn L. Beck-Dudley, Dean, College of Business
The Office of Graduate Studies has verified and approved the above named committee members.
ii
To my wonderful wife, Misty, whose unremunerated and oft-times unrecognized work as a mother is both unequivocally more challenging and infinitely more important than any of my professional accomplishments. Thanks for being my rock when the winds and rains have beat upon me.
iii ACKNOWLEDGEMENTS
No great work in life is ever accomplished alone. While readily acknowledging that my dissertation is no “great work,” I would like to gratefully acknowledge the many individuals who have improved the quality of this dissertation and influenced me along the way. First, thanks to my dissertation chair, Dr. David Peterson, who was mercilessly forced to read literally hundreds of pages of my writing. His questions, candor, direction, and suggestions have been invaluable. He, more than anyone else, has taught me how to be a researcher. Also, I sincerely thank each of the other members of my dissertation committee. Dr. Gary Benesh provided the voice of reason and experience. His candid feedback on the survey instrument and on the volume of work I initially proposed proved invaluable and prophetic. Dr. Mike Brady has been my brightly burning torch in an otherwise dark room as I constructed, distributed, and analyzed the responses to the survey. Dr. James Doran became my reason to keep working. On the days of frustration and disappointment, which accompany all dissertations, he was the optimistic cheerleader and demanding coach pushing me to keep going. His suggestions tremendously improved the quality of the work. I also want to thank Dr. Bill Christiansen whose instruction, mentoring, advice, and friendship have meant more to me than he will ever know. In addition to those mentioned above, I wish to thank the following individuals for their insightful suggestions, questions, and interest in my work: Prithviraj Banerjee, Jim Brau, Ronnie Clayton, Dean Diavatopolous, Michael Ehrhardt, Campbell Harvey, Matthew Spiegel, Tom Noe, Jeff Rockwell, Brian Tarrant, seminar participants at Central Michigan University, all respondents to my survey (especially those who provided feedback), and the entire finance faculty at Florida State University. Dave Horowitz, Tim Munyon, and Andrew Wilson provided much appreciated guidance in executing the structural equation modeling testing in the dissertation. Also, surveyZ and Qualtrics.com saved me countless hours of work by generously allowing me to use their survey software free of charge. Jean Heck graciously gave me access to his database that also greatly expedited the completion of the dissertation. Lastly, I thank Misty Wright for her expert assistance in collecting the data for this study. All errors in this dissertation are mine.
iv TABLE OF CONTENTS
LIST OF TABLES...... vi LIST OF FIGURES ...... vii ABSTRACT...... viii
INTRODUCTION AND MOTIVATION ...... 1 How Efficient Do We Think Us Stock Markets Are And Does It Really Matter?...... 3 What Really Matters When Buying and Selling Stock?...... 5 So You Discovered an Anomaly…Gonna Publish It?...... 6 Outline of Dissertation...... 9 HOW EFFICIENT DO WE THINK US STOCK MARKETS ARE AND DOES IT REALLY MATTER?...... 10 Introduction...... 10 Background...... 11 Subjects, Surveys, and Response Rate...... 13 How Efficient are US Markets?...... 19 Assessing Views of Market Efficiency Based on Investing Objectives...... 23 Does Market Efficiency Even Matter? ...... 24 Conclusion ...... 35 WHAT REALLY MATTERS WHEN BUYING AND SELLING STOCKS? ...... 46 Introduction...... 46 Background...... 47 Surveys in Finance Literature ...... 52 Survey Subjects, Description, and Distribution...... 53 Results...... 56 Conclusion ...... 67 SO YOU DISCOVERED AN ANOMALY…GONNA PUBLISH IT?...... 84 Introduction...... 84 The Theory and Assumptions ...... 86 Model Implications ...... 92 Analysis of Empirical Implications – Data and Methods ...... 94 Analysis of Empirical Implications – Results...... 101 Conclusion and Discussion...... 110 CONCLUSION...... 125 APPENDIX...... 127 REFERENCES ...... 138 BIOGRAPHICAL SKETCH ...... 144
v LIST OF TABLES
1. Summary Statistics 37 2. Opinions About Market Efficiency by Rank 38 3. Market Efficiency Specialists’ Opinions About Market Efficiency 39 4. Respondents’ Propensities to Actively Invest by Rank 40 5. Respondents’ Propensities to Actively Invest by Specialty 41 6. The Congruence of Respondents’ Opinions and Investment Objectives 42 7. Respondents’ Investment Objectives as a Function of Their Opinions and Confidence 43 8. Explaining Investment Objectives – Ordered Probit Analysis 44 9. Summary Statistics Reproduced 69 10. Relative Importance of 14 Groups of Variables 70 11. Relative Importance of 43 Individual Variables 71 12. Investment Experience by Rank 74 13. Respondents Who Have No Experience 75 14. Investment Experience by Gender 76 15. Ordered Probit Analysis of Investment Experience 77 16. What Matters to Active Investors 78 17. What Groups of Variables Matter to Active Investors 80 18. What Matters to Active Traders 81 19. Anomalies and Authors 112 20. Differences in Means and Medians: Anomaly vs. Matched Authors 113 21. Singe-Variable Probit Analyses 115 22. Full Specification Probit Analyses 116 23. Multicollinearity Mitigated Probit Analyses 119 24. Preliminary Regression Analyses 120 25. Multicollinearity Mitigated Regression Analyses 121 26. Principal Components 123 27. Principal Components Regression Analyses 124
vi LIST OF FIGURES
I. Structural Equation Modeling Path Diagram 45 II. Holdings of Active Investors 82
vii ABSTRACT
I study the topics of market efficiency and anomalies to market efficiency by focusing on finance professors in their joint roles as both researchers and market participants. I ask three main research questions: (1) how efficient do finance professors believe US stock markets are and does their opinion of market efficiency influence their investing behavior, (2) what really matters to finance professors when they buy and sell stocks, and (3) why do finance professors publish market anomalies? Related to the first question, I discover that finance professors agree that US stock markets are weak form efficient but not strong form efficient. However, there is much disagreement about the semi-strong form efficiency of US stock markets. Their investing behavior, though, suggests that finance professors accept markets as semi-strong form efficient; twice as many finance professors passively invest than actively invest. Surprisingly, their opinion about market efficiency has very little to do with their investing behavior. Instead, their investing behavior seems primarily driven by their confidence in their own abilities to beat the market, regardless of how efficient they perceive US stock markets to be. Related to the second question, I present three main findings. First, traditional valuation techniques and asset-pricing models commonly used in research and taught in the classroom are universally unimportant to finance professors when they buy and sell stocks. Second, the most important information to finance professors when considering stock purchases and sales are firm characteristics (PE ratio and market capitalization) and momentum related information (the stock’s return over the past six to 12 months and 52- week high and low). Third, finance professors have less real-world investing experience than one might expect – the median professor has bought an individual stock between 10 and 19 times, and 14.5% have never done so. Related to the third question, I find that finance professors are, in fact, acting rationally when they publish market anomalies. The theory I develop suggests it is rational for researchers to publish market anomalies if they have relatively few previous publications or have lesser reputations. Accordingly, the theory implies that the likelihood of publishing an anomaly and the profitability of published anomalies should be inversely related to the authors’ previous publications and reputation. These
viii implications are empirically corroborated providing evidence for the theory and supporting the notion that researchers are behaving rationally when they publish. Sadly, this also suggests that it is very likely that profitable anomalies have been discovered but not published so that the discoverer can exploit the anomaly, which provides indirect evidence of market inefficiency.
ix CHAPTER 1
INTRODUCTION AND MOTIVATION
Finance professors are compensated to read and perform high-level research on the subject of investing. Further, they teach the subject to undergraduate and graduate students. In addition to teaching and researching the topic, they also participate as individual investors directing the allocation of their own personal portfolios. Some of them are also privileged to participate at an even higher level by acting as professional managers. Hence, finance professors occupy a unique and privileged position. They are both researchers and participants in the arena of investing. They are abundantly educated, well informed, and highly sophisticated individuals who have the opportunity to apply the knowledge and sophistication gained through their research and teaching activities by participating in the very market they study. Additionally, the research they perform actually influences the arena in which they participate. Finance professors, then, represent a strikingly unique group. Yet to date they have almost been entirely ignored in their unique joint role as both researchers and participants. I can locate only one article that acknowledges and studies the dual role of finance professors – Haddad and Redman (2005). They survey finance, accounting, and economics professors to study four areas related to academics as investors: (1) current asset allocation, (2) expected sources of retirement income, (3) expected retirement asset allocation, and (4) types of financial instruments used. Considering the fact that only one article has explicitly recognized the important dual role of finance professors and that this article had a relatively limited scope, I see ample opportunity for productive and useful research in this area. Similar to Haddad and Redman (2005), I begin by recognizing the unique positions of finance professors as researchers and participants in the arena of investing. The topic of my work, however, substantially diverges from theirs. My dissertation is a compilation of three essays studying the subject of market efficiency and market anomalies through an in depth
1 analysis of finance professors in their joint role both as researchers and as market participants. The first two essays are based on a survey distributed to all finance professors at accredited, four-year universities and colleges in the United States. In the first essay, I concentrate on the issue of market efficiency. I ask finance professors five questions to assess how efficient they believe US stock markets truly are. I also empirically test, and ultimately refute, the notion that an investor’s decision to actively or passively invest is based on his perception of the efficiency of the market in which he is considering investing. In the second essay, I ask finance professors to identify which asset-valuation techniques, asset-pricing models, market anomalies, and other information are most and least important when they invest. In the second essay, I also analyze how much real- world investing experience finance professors possess. I am not the first to survey finance professors to assess their opinions on these matters. Welch (2000) surveys financial economists to develop a consensus estimate of the equity premium over a set of future horizons. As a secondary (arguably tertiary) matter, he also asks them a set of questions about “issues that are commonly debated in the academic literature.” Included in his survey are two broad questions about market efficiency and the stationarity of certain firm characteristics. His cursory treatment of the topics of this dissertation underscore their importance and also motivate the need for an in-depth study such as the one I offer. In the third essay, I address a perplexing question: why do finance professors publish highly profitable market anomalies? Considerable evidence exists that the profitability of market anomalies substantially decreases and oft-times entirely disappears once the anomaly is published (see Dimson and Marsh (1999) and Marquering, Nisser, and Valla (2006)). Given their role as market participants, finance professors could easily choose not to publish highly profitable anomalies and instead use them in their own trading. It doesn’t seem economically rational for a finance professor to publish a market anomaly, and yet our literature is replete with articles introducing and analyzing market anomalies (see Russell and Torbey (2002)). In the third essay, I present and empirically test a theory that explains this ostensibly irrational behavior. I further motivate each of the three essays below.
2
How Efficient Do We Think Us Stock Markets Are And Does It Really Matter?
Market efficiency has been given tremendous attention in academic literature. One reason for this is that the actually efficiency of securities markets presumably greatly influences the strategies investors adopt. Burton G. Malkiel in his book A Random Walk Down Wall Street explained the matter very eloquently. He remarked that if securities markets are efficient, “a blindfolded monkey throwing darts at a newspaper’s financial pages could select a portfolio that would do just as well as one carefully selected by experts.” Clearly, understanding the efficiency of securities markets is valuable information. Unfortunately, the literature on the subject has not definitively identified how efficient securities markets really are. There is much empirical evidence demonstrating efficiency in our stock markets. This is demonstrated by the myriad event studies that report the impounding of information into prices in an impressively short time period. On top of the academic evidence, the well-known Wall Street Journal Dartboard Contest seemed to convincingly validated Malkiel’s statement (see Adams and Cyree (2004) for summary results). In direct opposition to the efficient-market hypothesis, there is a large body of literature documenting short-term and long-term return anomalies (see Schwert (2002) and Russell and Torbey (2002) for insightful surveys of the subject). Market anomalies suggest that markets are not purely efficient and that investors may be able to construct strategies that consistently earn abnormal returns. There is obviously room for debate on the subject of the efficiency of US stock markets. But in spite of the many published anomalies, the ability of most investors to earn consistent abnormal returns with real-world investment dollars seems dubious. Even professional money managers and other sophisticated and informed investors struggle to beat the market (see Adams and Cyree (2004), Gruber (1996), Carhart (1997), Roll (1994), and Wermers (2000)). In the face of mounting evidence that sophisticated investors struggle to beat the market, investors increasingly appear to be merely passively investing. Battacharya and
3 Galpin (2005) present evidence that stock picking is declining around the globe and particularly in America, suggesting that investors in general are accepting the argument that it is extremely difficult to beat the market on a consistent basis. Instead of attempting to do so, more and more investors are simply passively investing their money in an attempt to mirror market returns. Battacharya and Galpin (2005) also note, however, that as more and more investors passively invest, the Grossman-Stiglitz (1980) paradox suggests markets might become more inefficient. I.e., the more investors accept and act as if markets are efficient, the more inefficient they may become, creating opportunities for above-market returns without having to incur commensurate risk levels. So it could be that as more and more people accept markets as efficient, the markets drift toward less efficiency. My objective in the first essay is to determine the collective opinion of the researchers on the subject through the use of a comprehensive survey instrument. Further, I aim to discover whether an investor’s perception of market efficiency really is a fundamental determinant of his investment objectives as many of us presume it is? I offer a brief preview of the results. I find that finance professors generally strongly agree that US stock markets are weak form efficient but that they are not strong form efficient. There is much disagreement, however, about the semi-strong form efficiency of US stock markets. The investing behavior of respondents helps to clear up the disagreement, though. Roughly twice as many respondents passively invest than actively invest, which suggests finance professors generally accept markets as semi- strong form efficient. Surprisingly, however, I discover that a finance professor’s opinion regarding the efficiency of US stock markets has little to do with his investment behavior and objectives. Using robust methodologies, including ordered Probit analysis and structural equation modeling, I discover that a respondent’s confidence in his own abilities to beat the market drive his behavior, not his opinion about the efficiency of the markets in which he invests, which motivates the need for further work investigating the role of overconfidence in investing such as that by Barber and Odean (2001).
4 What Really Matters When Buying and Selling Stock?
Similar to market efficiency, the question of what really matters when buying and selling stock has been the subject of many articles in finance literature. In fact, any article that deals with market efficiency or market anomalies necessarily addresses this question. However, once again, there seems to be little consensus about what truly matters when buying and selling stock. There exists a broad array of valuation techniques, asset-pricing models, and anomalies to market efficiency , which all suggest that some unique variable or factor that is highly relevant when buying or selling stocks. But the collection of literature and lack of consensus on the matter can be dizzying. Is β really the only thing that matters, as CAPM suggests? Or perhaps a stock’s correlations with the Fama and French (1993) factors are also extremely important. What about Carhart’s (1995) momentum factor? Could a stock’s dividend-yield also be important (Fama and French (1998) and Shiller (1998))? Or should an investor also look at a stock’s market capitalization (Banz (1981)), PE ratio (Basu (1977)), book-to-market equity (Stattman (1980)), or 52-week high and low (George and Hwang (2004))? What about it’s return over the past six months (Jegadeesh and Titman (1993))? Past 12 months? Is it the stock’s past returns or the industry’s past returns that matter (Moskowitz and Grinblatt (1999))? And, what about analysts – recommendations, target prices, earnings forecasts, etc.? Should an investor care what they say? What’s an investor to do? What matters most? What matters least? My objective in the second essay is to discover what valuation techniques, asset- pricing models, market anomalies, firm characteristics, corporate events, seasonal variables, and other information are most and least important to finance professors when they are considering buying and selling a stock. In the process, I aim to uncover how much real-world investing experience finance professors possess. Surely those reading their research and listening to their lectures would be interested to know whether finance professors represent a vastly experienced group on the subject of investing. Or, are they largely a group of theorists in an ivory tower with little real-world investing experience?
5 I offer a brief preview of the results. The most surprising finding is that the traditionally accepted and most widely taught valuation techniques and asset pricing models (such as dividend valuation models and CAPM) are strikingly unimportant in the eyes of finance professors when they invest. Instead, respondents indicate that the most important information in making their stock purchase and sale decisions are firm characteristics (especially the PE ratio and market capitalization) along with momentum related information (returns over the past six and 12 months and a stock’s 52-week high and low). I also find, somewhat surprisingly, that finance professors have less investing experience than one might expect. The average respondent had purchased an individual stock between 10 and 19 times, and 14.5% of all respondents had never purchased an individual stock.
So You Discovered an Anomaly…Gonna Publish It?
The third essay addresses a single critical question: If finance faculty have the opportunity to transform their education and especially their research into abnormal returns for themselves and investors whose money they manage, might they be inclined to withhold their most significant findings from publication to preserve the possibility of capturing the abnormal returns as long as possible? Simple laws of economics suggest that any activity that provides abnormal returns will attract attention. As the attention grows and entry occurs, the abnormal profitability of the activity will gradually dissipate, until the activity offers only normal profits. Similarly, if an investor discovers a strategy that results in consistent abnormal profits, he might reasonably expect that his strategy will receive increased attention. As his strategy receives increased attention, he should expect to observe other investors mimicking his strategy, until the abnormal profits it once provided likewise dissipate. The dissipation of profits would be expedited further if the investor made his strategy readily available to the general public. This brings up a heretofore-overlooked conundrum in the academic circles of finance. Perhaps given the opportunities for personal financial enrichment, finance faculty may be withholding from publication advancements in asset pricing models or
6 profitable anomalies to market efficiency in order to prevent their abnormal profits from dissipating. Perhaps finance professors are keeping the best for themselves and only sharing (publishing) the moderately interesting but ultimately unprofitable models, anomalies, and empirical results from their research. I am unaware of any literature addressing this possibility, which further underscores the importance of this dissertation. It is informative that Roll and Ross Asset Management states the following on its website: Roll and Ross was founded with a simple philosophy: Manage both risk and return. To control for risk, the firm relies on its proprietary APT risk control technology.
It is no surprise that an asset management company started by Stephen Ross would use some form of the APT model, but I am struck by the fact that they use a “proprietary” variation of the model. I suppose many would shrug their shoulders and find nothing out of the ordinary about this situation – an asset management firm using “proprietary” models in its investing. What is interesting is that the father of the APT started this firm, which is using a proprietary version of his model. It’s important to recognize that one cannot infer from that statement alone whether Ross had developed the proprietary APT risk control technology when he was still actively publishing and chose not to publish that particular model or whether this proprietary model came after he had decided to start Roll and Ross Asset Management. LSV Asset Management similarly admits to the use of proprietary models:
LSV Asset Management (LSV) is a quantitative value equity manager providing active management for institutional investors through the application of proprietary investment models.
Again, one cannot infer whether Lakonishok, Shleifer, and Vishny developed their proprietary models during their peak publishing years and chose not to publish their proprietary models or if they developed the models after forming LSV Asset management. But these examples motivate a study into the possibility of academics deciding against publishing their most prized work in order to capitalize on it in the markets. Even though principles of economics advises against publishing valuable anomalies, the literature has been filled with articles discussing specific market anomalies
7 (see Schwert (2002) and Russell and Torbey (2002) for a survey of the subject of market anomalies and Roll (1983) and Jegadeesh and Titman (1993) for examples of typical market-anomaly articles), how profitable they are (see, for instance, George and Hwang (2004)), how persistent they are (for example, see Dimson and Marsh (1999), Jegadeesh and Titman (2001), and Schwert (2002)), and what does or does not explain them (see, for instance, Grundy and Martin (2001) and Cooper, Guitierrez, and Hameed (2004)), but there are no articles that address the question of why finance professors are willing to publish them in the first place. To me, what’s more perplexing than the existence of market anomalies is the fact that anyone is willing to publish them once they are discovered. Considering the fact that we rely upon finance professors to discover and report market anomalies and the possibility that publishing an anomaly may prevent its discoverer from profitably exploiting it, it seems logical and necessary to explore the reasons why finance professors publish anomalies and especially the conditions under which finance professors may be motivated not to publish the anomalies they discover. One of the significant contributions of my third essay is to propose a theory that helps explain when or why a professor would or would not publish a profitable market anomaly he discovers. It ultimately predicts that the most accomplished finance professors have economic incentive not to publish profitable anomalies to market efficiency and advances in asset pricing. As I develop the theory, I explain the necessary conditions and then discuss the empirical implications of the model. The results ultimately corroborate the theory and suggest that authors who publish anomalies have fewer publications, especially fewer top publications per year, than non- anomaly authors publishing in the same journal. Additionally, authors who publish anomalies have been in the field for a shorter period of time than their non-anomaly counterparts. The profitability of an anomaly is inversely related to the number of publications that an author has at the time of the publication of the anomaly and the number of years the author has been in the field. Moreover, the profitability of an anomaly is strongly inversely related to the first principal component of the previous publications of the author and the number of years the author has been in the field. If this principal component may be interpreted as reputation, I can conclude that authors with
8 lesser reputations are much more likely to publish highly profitable anomalies, a conclusion that is consistent with the theory outlined in the paper. This implies that some of the brightest minds in our field, some of the most widely published authors in finance, have the highest incentive not to publish any market anomalies they may find. So while finance professors appear to be rational utility maximizers, we should not expect to see the kingpins of our field publishing highly profitable anomalies, unless their utility functions are highly skewed toward reputation. The empirically supported implication that some researchers discover but do not publish market anomalies also provides indirect evidence of inefficiency in the markets.
Outline of Dissertation
The remainder of the dissertation proceeds as follows. Chapter 2 contains the first essay entitled, “How Efficient Do We Think US Markets Are And Does It Really Matter? Chapter 3 contains the second essay entitled, “What Really Matters When Buying and Selling Stock?” Chapter 4 contains the third essay entitled, “So You Discovered an Anomaly…Gonna Publish It?” Each of the three essays is designed to be a self- contained publication-ready document, however, I offer brief concluding remarks on all three of the essays in Chapter 5.
9 CHAPTER 2
HOW EFFICIENT DO WE THINK US STOCK MARKETS ARE AND DOES IT REALLY MATTER?
Introduction
One of the most fundamental questions related to investing is whether one should actively or passively invest. Surely, the first factor to consider in making the active vs. passive investing decision is the efficiency of the market in which one is considering investing. Actively investing in a perfectly efficient market will not yield consistent abnormal returns, so why bother? This partially explains why so many academic articles in finance address the subject of market efficiency. Unfortunately, the voluminous work on the subject has not conclusively determined the actual efficiency of US stock markets. In light of the empirical disagreement on the subject, I take a unique approach to assessing the current efficiency of US stock markets. I survey the experts in the field to assess their opinions on the subject. Specifically, I invite all finance professors at accredited, four-year universities in the US to respond to a survey in which I inquire about their opinions on the efficiency of US stock markets. I ultimately find that finance professors strongly agree that US stock markets are not strong form efficient. To a slightly lesser degree they also concur that US stock markets are weak form efficient. However, they show little agreement regarding the semi-strong form efficiency of the markets. In spite of their ostensible disagreement on the matter, their investing objectives suggest they generally lean towards the belief that markets are, in fact, semi-strong form efficient: twice as many finance professors passively invest than those who actively invest. Although my primary objective is to determine the collective opinions of finance professors on the actual efficiency of US stock markets, I come to a surprising conclusion: a finance professor’s opinion of the efficiency of US stock markets does not really influence whether he actively or passively invests. Instead of basing their investment decisions on their perception of the general efficiency of US stock markets,
10 finance professors base their investing decisions on their confidence in their own abilities to beat the market, not the general efficiency of the markets. This contradicts the fundamental notion that the active vs. passive decision starts with an assessment of the efficiency of US stock markets. This finding questions the practical relevance of the literature on market efficiency and provides motivation for further work on the role of overconfidence in investing. The remainder of the essay proceeds as follows. Section II provides background on the topic, including a review of relevant literature. Section III describes the subjects, survey, and response rate. Section IV presents finance professors’ explicit opinions about the efficiency of US stock markets. Section V analyzes professors’ opinions of market efficiency through an analysis of their investing objectives. Section VI explores the question of whether a respondent’s decision to actively or passively invest is related to his perceptions of the market’s efficiency. Section VII concludes.
Background
I am not the first to survey academics to gather a reading on their collective opinion on market efficiency. Welch (2000) surveyed academics to assess their prediction for the equity premium over a set of future horizons. As a secondary matter, he also took the opportunity to survey academics on a broad range of topics of interest. Included in his survey was a single sweeping question about the efficiency of stock markets, which led him to conclude, “financial economists feel that, by and large, financial markets are efficient.” The lack of a clear empirical consensus on the actual efficiency of US markets moved Welch to include a question on the matter in his survey, which was fundamentally about a very different subject. There is much empirical evidence supporting a large measure of efficiency in our stock markets. This is demonstrated by the myriad event studies that report the impounding of information into prices in an impressively short time period. With mounting evidence of efficiency, Burton G. Malkiel in his book A Random Walk Down Wall Street explained the argument: “a blindfolded monkey throwing darts at a newspaper’s financial pages could select a portfolio that would do just as well as one
11 carefully selected by experts.” Malkiel’s statement spawned the well-known Wall Street Journal Dartboard Contest, which convincingly validated his opinion (see Adams and Cyree (2004) for summary results). In direct opposition to the efficient-market hypothesis, there is a large body of literature documenting short-term and long-term return anomalies (see Schwert (2002) and Russell and Torbey (2002) for insightful surveys of the subject). Market anomalies suggest that markets are not purely efficient and that investors may be able to construct strategies that consistently earn abnormal returns. However, many of the anomalies that have given cause for optimism to investors seeking to beat the market have been discredited for a variety of reasons: the anomalies may simply be practically unprofitable (see Jensen (1978) and Roll (1994)); the anomalies may derive most of their hypothetical profits from the short-side of a zero- investment portfolio (Chan (2003)), which may be constrained in reality; the anomalies may have been identified and promoted using erroneous or incomplete methodologies that do not fully consider changes to risk or relevant explanatory variables (see Fama (1998), Mitchell and Stafford (2000), Brav, Geczy and Gompers (2000), Eckbo, Masulis, and Norli (2000), Boehme and Sorescu (2002), and many others)); or the anomalies may have dissipated or even disappeared since their initial identification (see Dimson and Marsh (1999), Schwert (2002), and Marquering, Nisser, and Valla (2006)). Clearly, there is room for debate on the subject of the efficiency of US stock markets. In spite of the many published anomalies, the ability of most investors to earn consistent abnormal returns with real-world investment dollars seems dubious. Even professional money managers and highly sophisticated and informed investors struggle to beat the market (see Adams and Cyree (2004), Gruber (1996), Carhart (1997), Roll (1994), and Wermers (2000)). Consistent with the increasing volume of work suggesting that it is difficult for investors to beat the market consistently on a risk-adjusted basis, Battacharya and Galpin (2005) present evidence that stock picking is declining around the globe and particularly in America. Leveraging on the theoretical insight from Lo and Wang (2000), they posit that if everyone in the world holds only a combination the risk-free asset and the market portfolio, the trading volume of a stock should be entirely explained by its market
12 capitalization. It follows then that 1 – R2 from the cross-sectional regression of volume on market capitalization represents a reasonable measure of deviation from the indexing philosophy suggested by the two-fund separation theorem. In more precise language, 1 – R2 from the said regression proxies for the degree of stock picking in a given market. They find that this measure has been decreasing steadily throughout the world, and in the US in particular. Their paper suggests that investors are increasingly accepting the argument that it is extremely difficult to beat the market on a consistent basis. Instead of attempting to do so, more and more investors are simply passively investing their money in an attempt to mirror market returns. Battacharya and Galpin (2005) also note, however, that as more and more investors passively invest, the Grossman-Stiglitz (1980) paradox suggests markets might become more inefficient. I.e., the more investors accept markets as efficient, the more inefficient they may become, creating opportunities for above-market returns without having to incur commensurate risk levels. So it could be that as more and more people accept markets as efficient, the markets drift toward less efficiency. My objective is to determine the collective opinion of the researchers on the subject through the use of a comprehensive survey instrument. Further, I aim to discover whether an investor’s perception of market efficiency really is a fundamental driver of his investment objectives?
Subjects, Surveys, and Response Rate
Surveys in Finance Literature The use of survey instruments has historically been infrequent in finance literature for three main reasons. First, as Friedman (1953) articulated and Brav, Graham, Harvey, and Michaely (2005) reiterated, economic models are not conditional on the underlying agents’ understanding why they do what the do. As long as statistical inference using measurable data is able to adequately support or refute economic models and hypotheses, surveying the agents is unnecessary. Second, finance is privileged with access to an abundance of archival data, which provides a number of statistically desirable benefits in formal hypothesis testing. Therefore, use of the archival data is generally preferable to
13 survey-based data. Third, as Welch (2000) articulated, almost all surveys have shortcomings and flaws, which create skepticism regarding inferences from survey data. However, recent articles in respected journals such as Journal of Finance, Journal of Financial Economics, and Journal of Accounting and Economics have employed survey instruments as critical components of their research methodologies. Specifically, Brau and Fawcett (2006) survey CFOs to determine the primary motives behind firms’ decisions to go public, Brav, et al. (2005) survey financial executives to assess the determinants of dividend and share repurchase decisions, and Graham, Harvey, and Rajgopal (2005) survey executives to illuminate the factors motivating reported earnings and disclosure decisions. Further, the Journal of Applied Finance published two survey articles (Hartikainen and Torstila (2004), and Jorgensen and Wingender (2004)) in a single 2004 issue. Earlier articles based on surveys include Pinegar and Wilbricht (1989), Trahan and Gitman (1995), Welch (2000), Graham and Harvey (2001), and Krigman, Shaw, and Womack (2001). So although surveys are infrequent in finance literature, they have been published in some of the top journals in the field, are gaining greater acceptance, and have provided useful insights on a variety of subjects, particularly in bridging the theory-practice gap.
The Subjects The subjects of my study are finance professors in the United States. To identify the professors for the survey, I use the list of all regionally accredited U.S. universities compiled by the University of Texas at Austin.1 For each four-year university or college, I hand collect the names and email addresses of all professors of finance by visiting the relevant academic college and department websites at the university or college.
The Surveys The survey contains questions that fall into the following five categories: conditioning variables (mostly comprised of demographic variables), indicators of opinion on market efficiency, opinion on market efficiency, propensity to passively
1 http://www.utexas.edu/world/univ/
14 invest, and investment strategies. This essay primarily focuses on the portions dedicated to opinion of market efficiency and propensity to passively invest.
Beta Testing An important part of successful survey construction is beta testing, in which initial drafts of the survey are administered to test groups to determine the intelligibility, reliability, and validity of the questions on the survey. I beta test the survey on my colleagues - PhD students in the College of Business at Florida State University. After analyzing results from the beta testing and receiving feedback from the beta subjects, I modify the survey in close consultation with my dissertation committee members.
Distribution Finance literature incorporating surveys have largely depended on hard-copy distributions of surveys to potential respondents. Brau and Fawcett (2006) carry out three waves of mailings, which include a personalized envelope, personalized signed cover letter, the survey, a postage-paid return envelope and glossary of terms. They also enter respondents’ names into a drawing in which one respondent would receive $1,000. Their response rate is 18%. Brav et al. (2005) administer hard copies of their survey in person at two conferences and also distribute electronic copies of their survey via e-mail. To stimulate response they offer an advanced copy of the results and entry into a drawing in which two respondents would receive $500. Their total response rate is 16% (8% for the electronic delivery mechanism). Graham and Harvey (2001) distribute hard copies of their survey in two waves of mailings and faxes with follow up phone calls and faxes from a team of MBA students. They offer an advanced copy of results as incentive and experience a response rate of 9%. Paper surveys, however, are largely being replaced by electronic surveys distributed and collected via the Internet. Electronic surveys offer a number of advantages compared with traditional paper surveys (see Wright (2005), Van Selm and Jankowski (2006), Medlin, Roy, and Chai (1999), and Schaefer and Dillman (1998)). The primary advantages of electronic distribution and receipt are efficiency, expediency, accuracy, and cost savings. Specifically, electronic distribution and receipt
15 allows for distribution to a greater number of potential respondents in a shorter period of time, with faster and more complete responses, and without the expenses of envelopes, paper, and stamps. Additionally, the data are collected in electronic format, which removes the need for hand coding data and, thus, substantially reduces the possibility of measurement error introduced by mistakes in the transcription process. The primary disadvantages of this method of distribution and receipt are unrepresentative sampling and lower response rates. The unrepresentative sampling issue arises when some portion of the population of interest does not have access to the survey or when some portion of the population is less inclined to respond to the survey due to the delivery mechanism. The lower relative response rates (see Crawford, Couper, and Lamias (2001), which discusses Kwak and Radler (2000), Guterbock, Meekins, Weaver, and Fries (2000), and Medlin and Chai (1999)) may be a more troubling issue as demonstrated by recent finance literature. Brau et al. (2006) achieve an admirable response rate of 18% using personalized envelopes and personalized cover letters, while Brav et al. (2005) achieve a response rate of only 8% using the impersonal e-mail delivery mechanism. But in spite of the above contrast, there is encouraging evidence that response rates to electronic surveys are, in fact, not significantly different from response rates to paper surveys (see Schaefer and Dillman (1998)). In light of the considerable advantages it offers, I choose the electronic method of survey distribution and collection. The unrepresentative sampling concern should not be a problem in this essay since virtually every professor in the United States has access to the Internet and has an email account that is checked regularly and since the sample is relatively homogeneous. And although there may be some group that is less inclined to respond to an electronic survey, this issue is not unique to the electronic delivery format. Further, Schaefer and Dilman (1998) demonstrate that multi-mode (a combination of electronic and paper) contact with respondents did not significantly increase response rates. I also take comfort in the fact that if I can duplicate an 8% response rate, the sample in this study should be sufficiently large for testing. I take further comfort in the finding of Brav et al. (2005) that the responses to their electronic survey did not differ from those obtained through the in-person delivery mechanism.
16 I choose to create and distribute my surveys electronically using surveyZ.com and qualtrics.com2. I incorporate strategies that have proven helpful in increasing response rates to electronic surveys. Schaefer and Dillman (1998) state on p. 380 that, “the most powerful determinant of response rates is the number of attempts made to contact a sample unit.” They also argue that personalization increases response rates, suggesting the need to send emails addressed to the potential respondent rather than to a mailing list. Crawford et al. (2001) suggest that, in the electronic environment, reminders subsequent to the initial invitation to participate in the survey are more effective when sent two days after the initial invitation as opposed to one week as suggested by Dillman (1978) for mail surveys. They also find that offering an accurate estimate (as opposed to an inaccurately low estimate) of the time it takes to complete the survey leads to a higher number of completed surveys.3 They further argue that including a progress bar in the survey is recommended.4 Bosnjak and Tuten (2003) provide evidence that of four possible incentive payment structures (pre-paid, post-paid, prize-drawing, and no incentive), the prize-drawing incentive structure is most effective in eliciting completed responses (Brau and Fawcitt (2006) and Brav et al. (2005) use this incentive structure). Following the suggestions of the above literature, I implement a series of emails inviting participation in the study. Two days before (day – 2) distributing the official invitation to participate electronically, I send an email, the “pre-mail,” to all potential respondents. The pre-mail explains (a) that they will be receiving an electronic invitation to respond to the survey, (b) that their responses will be strictly confidential, (c) the purpose of the survey, (d) when they can expect to receive the invitation email, (e) an estimate of how long it will take to complete the survey, (f) and the incentives for responding. The incentive for responding is entry into a drawing for $500. The premail
2 I express appreciation to surveyZ.com and qualtrics.com who provided their survey software free of charge. 3 Crawford et al. (2001) find that providing an inaccurately low estimate of the time requirement elicited more responses, but more of those respondents dropped out of the survey before completion. Providing an accurate estimate led to fewer respondents starting the survey, but more of those who started actually finished the survey. 4 They actually found that including a progress bar led to lower completion rates, but they argued their results were an artifact of the survey structure, which included many open ended questions in the beginning stages of the survey. In spite of their results, they endorse the inclusion of a progress bar.
17 also contains a link to the survey and gives the recipients the option to take the survey at that time if they prefer. Two days after the pre-mail, I send the official invitation email (day 0) including the link to the survey hosted at qualtrics.com and repeating the relevant information from the pre-mail. Two days after the initial invitation survey (day 2), I send a “post-mail” to remind potential respondents of the opportunity to fill out the survey. A transcript of the survey, along with copies of the pre-mail, invitation email, and post-mail are contained in Appendices A, B, C, and D respectively. To further increase response rates, emails are sent in a manner such they are addressed individually to each potential respondent. Also, a progress bar is included at the bottom of each page of the survey. The premail was sent on February 19, 2007, the invitation email was sent on February 21, 2007, and the postmail was sent on February 23, 2007. The survey was deactivated on February 26, 2007.
Response Rate Emails inviting participation in the survey were sent to 4,525 professors. 60 of the email addresses were invalid. 1,183 professors started the survey, which is a started response rate of 26.49%. 870 professors completed the survey – i.e., they answered the final question on the survey – which represents a completion rate of 73.54% and a completed response rate of 19.48%. In order to enter the final data set, I require a respondent to meet five criteria: 1) s/he must answer yes to the consent question, 2) s/he must be a finance professor, thus eliminating professors of law, economics, and other disciplines,5 3) s/he must hold a Ph.D. or DBA, 4) s/he must be of the rank of assistant professor, associate professor, full professor, endowed chair, or eminent scholar, and 5) s/he must answer at least one of the five questions on the last page of the survey. There were two exceptions to this final rule. Two respondents answered well over half the questions on the survey but did not answer the last five. I allowed these two respondents to enter the final data set. Aside from these
5 Some of the websites from which email addresses were collected made it impossible to distinguish finance professors from professors of other fields, such as business law and economics. Because of this, the survey was sent to and completed by some professors of other disciplines.
18 outliers, the respondents who failed to answer one of the last five questions universally answered less than half the questions on the survey, which casts doubt on the credibility of their responses to the questions they did answer. I also removed three professors whose responses contained glaringly inconsistent answers to questions.6 Table 1 presents the summary statistics on the responses to the survey. The final data set consists of 642 respondents. Of the 642, 197 are assistant professors, 197 are associate professors, 171 are full professors, 71 are endowed chairs, and 6 are eminent scholars. Almost 15% of respondents are female. Slightly more than 83% of respondents are married. The median age of respondents is between 40 and 49. The median 9-month salary of respondents is $110,000 to $119,999. Respondents have an average of 13.37 articles published in peer reviewed journals and an average of 1.84 articles published in the Journal of Business, Journal of Finance, Journal of Financial Economics, Journal of Financial and Quantitative Analysis, or Review of Financial Studies.
How Efficient are US Markets?
Although Welch (2000) asks a question similar in spirit to mine, my reading on finance professors’ opinion about market efficiency offers a richer assessment of the matter. It is important to acknowledge that assessing professors’ opinions of market efficiency was not his primary research interest. Hence, it was naturally given less treatment. My survey improves upon his in three primary ways. First, his survey was presented to a limited target population, which resulted in 226 total responses. My survey was sent to virtually every finance faculty member at accredited, four-year universities in the United States, which resulted in 642 useable responses. The increased sample size resulting from the broader distribution represents an opportunity for richer results.
6 For instance, one professor responded that he had published a grand total of one article in peer-reviewed journals, but he indicated later that he had published over 50 articles in peer-reviewed journals that support market efficiency.
19 Second, I ask respondents a number of demographic and conditioning questions that allow me to form groups of professors who may be more or less knowledgeable about market efficiency. For instance, I ask respondents to identify their specific areas of specialty, which allows me to zero in on the opinions of finance professors who specialize specifically in market efficiency. Welch asked similar questions but with much less specificity. These types of categorizing questions should provide another opportunity for richer results. Third and most importantly, I ask five questions related to market efficiency whereas Welch asks two. Welch asks respondents how strongly they agree or disagree with the following statements: 1. I believe that, by and large, public securities market prices are efficient 2. I believe that, by and large, public securities market prices offer arbitrage opportunities.
In order to gauge finance professors’ opinions of the efficiency of US stock markets, I ask respondents to indicate how strongly they agree or disagree with the following statements. The scale is from 1 (strongly agree) to 7 (strongly disagree): 1. It is possible to predict future returns to US stocks using only past returns. 2. It is possible to predict future returns to US stocks using only past returns and publicly available information. 3. It is possible to predict future returns to US stocks using only past returns, publicly available information, and private information. 4. Investment returns are solely a compensation for risk. 5. Investment strategies exist that consistently beat average market returns without taking above- average risk. It is worth noting at this point that no survey will meet everyone’s approval. Some respondents took issue with the wording of one or more of the above questions. For example, Questions 1 – 3 above are intended to represent inquiries about weak, semi- strong, and strong form market efficiency. One respondent, however, pointed out in an email that he believes future returns can be predicted using past returns merely by employing the CAPM, which is not a statement regarding the inefficiency of the market. This possible interpretation was discussed with my dissertation committee prior to distributing the survey, but we decided on the above wording. Our reasoning is simple: if we asked respondents how strongly they agreed or disagreed with the statement that US
20 markets are weak form efficient, we leave the term weak form efficient open to interpretation. Hence, we decided to use more concrete language. The point is regardless of how the questions are constructed, someone will always think it is poorly worded. This is precisely why we ask five questions about market efficiency, instead of just asking the catchall question – how strongly do you agree or disagree the statement that US stock markets are efficient. I believe there is rich information to glean from professors’ responses to the above five questions, in spite of possible wording and interpretation disputes.
Results Table 2 shows the mean responses to the five questions regarding market efficiency. The table reports overall means and medians as well as means and medians across rank. Finance professors generally agree that US stock markets are weak form efficient. Respondents disagreed (mean response of 5.3) with the statement that future stock returns could be predicted using only past returns. Conversely, finance professors strongly agree that US stock markets are not strong form efficient. Respondents noticeably agreed (mean response of 2.68) with the statement that future returns to US stocks could be predicted using only past returns, publicly available information, and private information. They seem much less decided on the matter of semi-strong form efficiency. Respondents were relatively neutral (mean response of 4.46) about the statement that future stock returns could be predicted using only past returns and publicly available information. Regarding the other two questions on the subject, respondents were relatively neutral. Mean responses to the statements that (1) investment returns are solely a compensation for risk and (2) investment strategies exist that consistently beat average market returns without taking above-average risk are 4.29 and 4.46, respectively. Some of the respondents may be more qualified than their peers to offer an opinion on the efficiency of US stock markets. Specifically, finance professors who specialize in market efficiency should be particularly well qualified to make a statement on the matter. As a part of my survey, I ask respondents to indicate their specific areas of specialty. One of the options is “market efficiency and anomalies to market efficiency.”
21 I expect those finance professors who specialize in market efficiency to be most the most informed on the subject. I report the responses of this group in Table 3. The table shows the number and percentage of respondents who specialize in market efficiency who responded 1 through 7 to the five questions. Finance professors who specialize in market efficiency overwhelmingly agree that US stock markets are not strong form efficient. 67% of these participants showed their agreement by responding either 1 or 2, compared to only 9% responding 6 or 7, to the statement that future stock returns could be predicted using past returns, publicly available information, and private information, indicating they clearly do not accept US stock markets as strong form efficient. Although not quite to the same degree, these experts also seem largely convinced that US stock markets are weak form efficient. Only 14% of this group responded 1 or 2, compared to 49% responding 6 or 7, to the statement that future returns could be predicted using only past returns, indicating their general disagreement with the statement. Their opinion on semi-strong form efficiency, however, is much less polarized. 29% responded 1 or 2, while 24% responded 6 or 7 to the statement that future returns can be predicted using past returns and publicly available information. This suggests the experts lean slightly to the side of US markets not being semi-strong form efficient, but mostly it suggests the experts are undecided or ambivalent about the semi-strong form efficiency of US markets. This is somewhat expected given the conflicting empirical evidence on the matter. Respondents who specialize in market efficiency also generally disagree that investment returns are solely a compensation for risk (35% responded 6 or 7 compared to only 15% who answered 1 or 2). However, they also generally disagree with the notion that strategies exist that consistently beat average market returns without taking above average risk (30% responded 6 or 7 compared to only 18% answering 1 or 1). Both overall results and results from focusing specifically on market efficiency experts indicate finance professors are fairly strongly convinced that markets are not strong form efficient. To a slightly lesser degree, they seem to largely agree that US stock markets are weak form efficient. However, they seem almost perfectly split in their opinions of semi-strong form efficiency.
22 The semi-strong form efficiency of a market in which one is considering investing is arguably the most important of the three forms to those faced with the active vs. passive decision. It is unfortunate, though not unexpected, that the experts are almost perfectly split on the matter. Perhaps an analysis of their investing goals and behavior will help to clarify where they truly stand on the issue.
Assessing Views of Market Efficiency Based on Investing Objectives
Considering the general lack of consensus regarding the semi-strong form efficiency of US markets, it may be informative to explore respondents’ investment behavior to infer information about their views on market efficiency. It seems reasonable to expect that finance professors who passively invest are more convinced of the efficiency of US stock markets than those who actively invest. I ask the following question to determine the propensity of finance professors to passively or actively invest. “Please indicate how strongly you agree or disagree with the following statement: When I invest, my goal is to beat the market.” The scale is from 1 (strongly agree) to 7 (strongly disagree).
Results Table 4 shows the responses of all participants to the question asking respondents to indicate how strongly they agreed or disagreed with the statement, “When I invest, my goal is to beat the market.” The table separates respondents by rank and shows the number and percentage of respondents in each rank who responded 1 through 7 to the question. The most salient feature of the table is the fact that 42% of all participants responded either 6 or 7, compared to only 18% responding 1 or 2 to the statement. This may be interpreted to mean that twice as many finance professors admit to passively investing than those who admit to actively investing. This trend largely holds across all ranks: almost twice as many respondents within each rank passively invest than those who actively invest. I interpret this to mean that although professors’ opinions appear largely undecided about the semi-strong form efficiency of US stock markets, their
23 investment objectives suggest they are more convinced of the markets’ efficiency than they admit. Again, I am particularly interested in the group of respondents who specialize in market efficiency. I report their responses to the statement, “When I invest, my goal is to beat the market” in Table 5. For comparison purposes, I also include five other specialties in the table: (1) asset pricing, (2) behavioral finance, (3) capital structure, (4) corporate governance, and (5) derivatives. I include capital structure and corporate governance, because these are distinctly corporate finance subspecialties. I am interested to see how this group behaves. I include the other three specialties because they represent respondents who may be more inclined to believe either (a) markets are inefficient or (b) regardless of market efficiency, they have the skills to beat the market. The primary result from Table 5 is that those who specialize in market efficiency behave much like the overall sample. Twice as many of these experts passively invest compared to those who actively invest (40% to 20%). Again, this group indicated earlier it is largely undecided about the semi-strong form efficiency of US markets, but their investment objectives suggest they accept US stock markets as more efficient than they are willing to admit. The professors who specialize in the other areas shown in the table generally differ from the overall sample in predictable ways. Respondents specializing in corporate finance (capital structure and corporate governance) show a much stronger propensity to passively invest than the overall sample. Three times as many professors in these fields passively invest than those who actively invest. Professors specializing in behavioral finance and derivatives, however, shower a stronger propensity to actively invest than the overall sample. The proportion of these professors actively investing is nearly identical to the proportion passively investing. Professors who specialize in asset pricing show a propensity to passively invest that is similar to the overall sample.
Does Market Efficiency Even Matter?
I asserted earlier that the first factor to consider in making the active vs. passive investing decision is the efficiency of the market in which one is considering investing.
24 If this statement holds, I expect a finance professor’s investment objectives to be strongly correlated with his opinion of the efficiency of US stock markets. To begin the analysis, in Table 6 I double sort respondents based on their opinions of market efficiency and their investment objectives. To represent a respondent’s opinion of market efficiency, I take the average of the participants’ responses to the three questions related to weak, semi-strong, and strong form efficiency. To represent their investment objectives, I use participants’ response to the statement, “When I invest, my goal is to beat the market.” Panel A reports raw numbers, while Panel B reports percentages. The portions of the table highlighted in gray represent respondents whose investment objectives are highly congruent with their opinions of market efficiency (those who believe markets are inefficient and are trying to beat the market or those who believe markets are efficient and are not trying to beat the markets). The portions highlighted in black represent respondents whose investment objectives are highly incongruent with their opinions of market efficiency (those who believe markets are efficient and yet are trying to beat the market or those who believe markets are inefficient but are not trying to beat the markets). The table demonstrates that the portion of respondents whose investment objectives are congruent with their opinions of efficiency is much higher than the portion whose objectives are incongruent. The investment objectives of 264 (42%) respondents are highly congruent with their opinions of market efficiency, while the investment objectives of only 137 (22%) are incongruent with their opinions of market efficiency. This simple table suggests that a respondent’s opinion of market efficiency is, in fact, highly correlated with his investment objectives. However, there is a considerable portion of the sample (22%) who behave in a manner that is seemingly incongruent with their opinion of the actual efficiency of the markets. The inconsistency between a professor’s beliefs about the efficiency of the markets and his stated investment objectives may be explained by his level of confidence in his own ability to beat the market. I.e., the inconsistencies may be explained by the response to a question on the survey in which I asked respondents to indicate how strongly they agreed or disagreed with the following statement, “Given sufficient time
25 and resources, I could implement an investing strategy that would consistently beat the market.” In effect, there are two factors at work that may heavily influence an investor’s propensity to passively or actively invest: (1) his perceptions of the general efficiency of the markets in which he will invest and (2) his confidence in his own abilities to beat the market. It is easy to conceive of an investor who believes markets are inefficient but who does not believe he can develop a strategy to capitalize on the inefficiencies. In this case, the investor would likely passively invest, which would create one of the perceived inconsistencies discussed above – an investor who believes markets are not efficient but who still passively invests. To explore this possibility, I double sort respondents based on their opinions of market efficiency and their belief in their own abilities to implement an investing strategy that would consistently beat the market. For each subgroup, I report the mean response to the statement, “When I invest, my goal is to beat the market.” I expect those whose believe markets are inefficient and who believe they could personally implement a market-beating strategy to be the most likely to try to beat the market, while I expect those who believe markets are efficient and who do not believe they could implement a market-beating strategy to be the most likely to passively invest. Table 7 reports this information. Panel A uses the response to the question about the weak form efficiency of US stock markets as the measure of each respondent’s opinion of market efficiency. Panel B uses the response to the question about semi- strong form efficiency to represent each respondent’s opinion of market efficiency. The table reveals a somewhat surprising result. A professor’s opinion on the general efficiency of US stock markets has much less influence on his investment objectives than his confidence in his own abilities to implement a market beating strategy. Within confidence groupings (within columns), there is no monotonic pattern as respondents’ opinion of market efficiency changes. However, within the opinion groupings (within rows), there is an obvious monotonic pattern as a respondent’s confidence in his own investing abilities changes. This suggests that one’s opinion of market efficiency has very little to do with the decision of whether to actively or passively invest. What matters is merely a person’s confidence in his own abilities.
26 To add statistical meaning to the pattern presented in Table 7, I estimate an ordered Probit model. The dependent variable is the response to the statement, “When I invest, my goal is to beat the market.” The independent variables of interest include responses to the three statements about weak, semi strong, and strong form efficiency and the response to the statement, “Given sufficient time and resources, I could implement an investing strategy that would consistently beat the market.” Control variables include a gender binary variable, a marital status binary variable, the respondents’ rank and age, and dummy variables indicating the respondent specializes in asset pricing, behavioral finance, capital structure, corporate finance, derivatives, or market efficiency. I estimate two iterations of the ordered Probit model. In the first iteration I include the independent variables as listed above. In the second iteration, I add three interaction terms in which a participant’s response to the statement about his confidence in his ability to implement a market beating strategy is interacted on his responses to the weak, semi-strong, and strong form efficiency questions. The results of the ordered Probit model estimations are reported in Table 8. Several points in the table stand out. First, an investor’s opinion of market efficiency has little impact on his investing objectives. A respondent’s opinion of the weak form efficiency of the markets has no statistically significant relationship to his investment objectives. A respondent’s opinion of the strong form efficiency of markets is inversely related to his investment objectives. I.e., the less one accepts markets as strong form efficient, the more inclined he is to simply passively invest. This is perhaps reasonable if the investor assumes most abnormal investing profits are driven by private information and if the investor does not consider himself privy to the necessary private information. While a respondent’s opinion of the semi-strong form efficiency of the markets is significantly directly related to his investment objectives in the first iteration, it becomes negative and insignificant when the interaction terms are added to the model, which suggests that all of the influence of a respondent’s opinion about semi-strong form efficiency on his investing objectives is dependent on his confidence in his abilities to beat the market. This leads to the second main point from the table. A respondent’s investment objectives are primarily driven by his confidence in his abilities to beat the market,
27 regardless of his opinion of market efficiency. In the first iteration, the magnitude and statistical significance of the coefficient on the confidence variable exceeds that of all other Likert scale variables. Importantly, the coefficient on the confidence variable remains strongly statistically significant even in the presence of the interaction terms, which suggests a person’s confidence in his own abilities to beat the market is a major driver of his investment objectives, regardless of his opinion of market efficiency. This is further supported by the fact that the magnitude and statistical significance of the coefficient on the confidence variable well exceeds those of any of the interaction terms. It’s not an investor’s opinion of market efficiency that matters; rather, it’s his confidence in his own abilities to beat the market, regardless of his opinion about market efficiency, that seems to matter. A third interesting result from the table is that of the six specialties included in the model, only derivatives is statistically significant. Specifically, a professor who specializes in derivatives is much more likely to actively invest than one who does not. The summary results to this point are as follows. Finance professors agree that markets are weak form efficient. They agree even more strongly that markets are not strong form efficient. Conversely, they are conflicted about the semi-strong form efficiency of US stock markets. Their investment objectives, however, suggest that finance professors do generally accept markets as semi-strong form efficient, since about twice as many of them passively invest than those who actively invest. But surprisingly, finance professors’ opinions about the efficiency of US stock markets have very little to do with their decision of whether to actively or passively invest. Regardless of their perception of market efficiency, their investment objectives are primarily driven by their confidence in their own abilities to implement a strategy that can beat the market. I began the paper by assuming that an investor’s opinion about the general efficiency of the markets in which he considering investing is fundamental to his decision of whether to actively or passively invest. This seems refuted. Instead, what matters is simply his confidence in his own abilities, regardless of how efficient he perceives the markets to be. To analyze the robustness of this surprising result, I use structural equation modeling, which is a methodology that has been widely employed in literature that commonly uses survey data.
28
Structure Equation Modeling – General Information SEM, although not new, is still uncommon in finance literature. Even the studies using surveys do not employ this technique. Therefore, I offer a brief explanation of SEM. SEM is a multivariate predictive technique that utilizes covariance matrices (correlation matrices may also be used, but this is not recommended since doing so changes the interpretation of the output and may result in biased parameter estimates), instead of observation matrices, and common estimation techniques, such as maximum likelihood, to simultaneously estimate a set of separate but related regression equations. It is particularly useful in estimating multiple dependent relationships where the dependent variable in a given model serves as an independent variable in a related model (Hair, Black, Babin, Anderson, and Tatham (2006)). Although SEM is not typical in finance literature, it is one of the foundational methodologies in disciplines that regularly use survey data.7 For example, consider the following system:
Y1 = γ11X1 + ε1
Y2 = γ22X2 + β21Y1 + ε2
Y3 = β31Y1 + β32Y2 + ε3 The multiple dependent relationships in this system make it a particularly good
candidate for SEM. The dependent variable Y1 in the first equation, which is a function
of the exogenous variable X1, serves as an explanatory variable in the second and third
equations. Similarly, the dependent variable Y2 in the second equation, which is a
function of the exogenous variable X2 and the endogenous variable Y1, also serves as an explanatory variable in the third equation. The third equation is a function of the
endogenous variables Y1 and Y2. OLS estimation assumes that the disturbance term of the dependent variable is uncorrelated with the explanatory variables. This is guaranteed when the explanatory
variables are non-stochastic (not random). Clearly however, Y1 and Y2 as explanatory variables in the above system are stochastic. If the system only included the equations
7 For instance, Brady, Calantone, Ramirez, and Voorhees report in an untitled working paper that over the years of 1996 – 2005 at least 283 articles in the top seven marketing journals employed structural equation modeling in their methodologies.
29 with Y1 and Y2 as dependent variables, the classic solution to this econometric problem is
two-stage least-squares estimation (2SLS), which would estimate Y1 using X1 and then
use the fitted value of Y1 as the explanatory variable in the second equation. This special case of the instrumental variables approach has become a workhorse in finance in resolving the above issue. The inclusion of the third equation introduces further difficulties. It would require estimation of Y2 using X2, and the fitted value of Y1. Then the fitted values of Y1 and Y2 would be used to estimate Y3. Instead of working through the system in stages as does the instrumental variables approach, SEM solves the system simultaneously. It does so through path analysis, which estimates the theorized relationships between the variables in a system by analyzing the correlations between the variables along all possible paths connecting the variables in the system. Its ability to simultaneously solve the system hinges on restrictions, usually zero restrictions, placed on the system. This typically means that in order for SEM to work, the researcher must establish that some of the variables in the system are unrelated. Each zero restriction adds a degree of freedom to the model, which increases statistical power. SEM also offers other advantages to traditional regression techniques. Classical linear regression (CLR) techniques are designed to model observed variables, whereas SEM is designed to model observed and latent variables by incorporating a level of confirmatory factor analysis. In the process, SEM also reduces measurement error by using multiple observed variables to model the true latent variable, whereas CLR typically uses a single observed variable to proxy for the true variable of interest. Further, SEM explicitly models measurement error, which if present in explanatory variables creates a downward bias in coefficient estimates in CLR (see Hausman (2001)). Also, although SEM shares many common assumptions with CLR, it is more flexible in its ability to handle violations of the CLR assumptions. Like CLR, SEM assumes normality in the disturbance term of the dependent variable, specifically, multivariate normality since there are multiple dependent variables. Violation of this assumption can be handled with bootstrapping estimation techniques. And although MLE is the primary estimation technique in SEM, it can also employ distribution-free estimation techniques to obtain estimates in the presence of violations of the multi-variate
30 normality assumption as long as the data set is sufficiently large. Like CLR, SEM also assumes a linear relationship between dependent and independent variables. Further, SEM can suffer from problems of multicollinearity similar to CLR. Similar to CLR, SEM is tested by analyzing the overall fit of the model and the significance of the individual parameter estimates. The individual parameter estimates from the model are reported with significance levels similar to CLR. But the individual parameter estimates and their significance are of little use if the overall goodness of fit of the model is inadequate. The goodness of fit tests for SEM fall into three categories: (1) absolute fit measures, which measure the degree to which the model predicts the observed covariance matrix, (2) incremental fit measures, which compare the proposed model to some baseline or benchmark model, and (3) parsimonious fit measures, which measure model fit while penalizing models for a lack of parsimony. Multiple test statistics exist to measure goodness of fit – χ2, GFI, RMSEA, RMR, and SRMR for absolute fit measures; TLI/NNFI, NFI, CFI, IFI/Δ2, and RNI for incremental fit measures; and NFI, PGFI, and AIC for parsimonious fit measures (some of the specific measures cross over into multiple fit categories). For brevity, I address below only those statistics I decide to use in my analysis. While each category of goodness of fit measure and specific test statistic within the categories offer advantages and disadvantages, the simple χ2 statistic is the most commonly reported goodness of fit measure. The lower the χ2 statistic, the better the model fit, which means lack of significance is desirable. This unfortunately means that the best a researcher can expect using the χ2 measure of model fit is to fail to reject their theoretical model. The χ2 statistic, however, can be misleading depending on the complexity of the model (the more complex the model, the better the fit), the size of the sample (the larger the sample, the poorer the fit), and violations of the multi-variate normality assumption. Perhaps the most conservative approach is to follow Hu and Bentler’s (1999) recommendation. They endorse use of multiple measures of fit. Specifically, they recommend use of SRMR with one of the following: NFI/TLI, BL89, IFI, RNI, CFI, Mc, or RMSEA. Since the TLI measure seems highly endorsed (see Tucker and Lewis
31 (1973), Bollen (1986), Baumgartner and Homberg (1996)), I choose to report the SRMR and TLI measures of the goodness of fit. I also report the ubiquitous χ2 statistic and the CFI, which Gerbing and Anderson (1992) praise for its stability in the face of differing sample sizes. I briefly explain SRMR, TLI, and CFI below. SRMR stands for Standardized Root Mean Square Residuals. It is the average difference between the predicted and observed variances and covariances in the model, based on standardized residuals. A model with better fit will have a lower SRMR statistic. SRMR statistics approaching 0.08 are desirable (Hu and Bentler (1999)). TLI stands for Tucker Lewis Index. It measures the incremental goodness of fit between two models (proposed vs. null) and is calculated thusly:
2 2 (χ null / df null ) − (χ proposed / df proposed ) TLI = 2 (χ null / df null ) −1 Since a smaller χ2 statistic represents better model fit, it is desirable for the numerator to be large. Hence, a TLI statistic greater than 0.9 and approaching 0.95 is recommended (Hu and Bentler (1999)). The TLI statistic is typically between 0 and 1, but one can see that mathematically it can exceed 1 if the χ2 from the proposed model is less than the degrees of freedom from the proposed model. CFI stands for Comparative Fit Index. It is bounded by 0 and 1 and similarly measures the goodness of fit between a proposed and null model. It is calculated thusly:
2 max(χ proposed − df proposed )0, CFI = 1− 2 2 max(χ proposed − df proposed , χ null − df null )0,
Again, since a smaller χ2 statistic represents better model fit, the ideal result would be a numerator approaching zero – indicating good model fit – and a large 2 denominator (the χ null – dfnull term being large), which would result in a CFI statistic approaching 1. Hence, a CFI statistic approaching 0.95 is generally considered preferable (Hu and Bentler (1999)).
Structural Equation Modeling Testing The primary objective of my use of structural equation modeling is to test the robustness of the finding that one’s confidence in his own abilities to beat the market, not
32 his perceptions of the efficiency of the market, is the fundamental driver of his propensity to actively invest. I, therefore, structure the model such that a respondent’s opinion of market efficiency is reflected in his responses to the two questions about the weak and semi-strong efficiency of US stock markets. In other words, a respondent’s opinion is a latent variable manifested through his observed responses to these two questions. I include a respondent’s confidence in his own abilities to beat the market as an observed explanatory variable in the model. The dependent variable in the model is a respondent’s propensity to actively invest. For robustness, I define this differently than I did in the ordered Probit model. Instead of using a participant’s response to the statement, “When I invest, my goal is to beat the market,” I use a latent variable that is reflected in a participant’s response to questions about how frequently he trades certain investment assets. Specifically, a respondent’s propensity to actively invest is a latent variable manifested through the frequency with which he buys stocks, sells stocks, and sells ETFs.8 In the model, then, the latent dependent variable (propensity to actively invest), which is manifested through the frequency with which one buys and sells stocks and sells ETFs is a function of the latent variable representing his or her opinion on the efficiency of US stock markets and the observed explanatory variable indicating one’s confidence in his or her own ability to beat the market, without regard to the efficiency of the markets. If the previous finding is robust, I expect the latent variable representing a respondent’s opinion on market efficiency to be insignificant. Conversely, his or her confidence in his or her own abilities should be strongly significant.
Structural Equation Modeling Results I present the path diagram, which graphically depicts the model, in Figure I. Figure I also includes the parameter estimates from the model, along with p-values for each of the parameter estimates. Additionally, the bottom of the figure also contains the overall goodness of fit measures discussed above.
8 These three behaviors seem particularly associated with active investing. I.e., more frequent purchases and sales of stocks and more frequent sales of ETFs seem to be reasonably representative of active investing.
33 Before discussing the individual parameter estimates of the model, it is important to determine whether the overall fit of the model is adequate and whether the latent variables are reliably reflected by the observed variables selected to manifest them in the model. The χ2 from the model is 6.704 with 7 degrees of freedom, which is insignificant (p = 0.460). The SRMR from the model is 0.0131. The TLI and CFI are 1.001, and 1.00 respectively. Considering the SRMR should be below 0.08, the TLI and CFI measures should approach 0.95, and an insignificant χ2 is desirable, the overall model fit is adequate.9 The scales used in the model are reliable. Cronbach’s alpha measures the reliability of the manifest variables as they relate to the latent construct they reflect and should be above 0.7. The Cronbach’s alphas in the model are 0.853 and 0.789, respectively, for the opinion and propensity to actively invest latent constructs. The scales also demonstrate what is known as adequate convergent validity. The average variance extracted for the scales, which should exceed 0.5 (Fornell and Larcker (1981)), are 0.75 and 0.61, respectively, for the opinion and propensity to actively invest latent constructs. Further, discriminant validity is supported by the fact that the average variance extracted for each of the scales exceeds the shared variance between the two latent variables (the average variance extracted for each latent construct exceeds the squared correlation between the two constructs). Having shown the overall goodness of fit of the model and the reliability of the latent constructs to be adequate, it is now appropriate to discuss the individual parameter estimates. Figure I demonstrates quite clearly that, again, a respondent’s opinion of market efficiency has little, if anything, to do with his propensity to actively invest. The coefficient for opinion in the model is 0.013 and is insignificant (p = 0.689). However, the coefficient for confidence is 0.147 and is significant at the 1% level. Indeed, the results of the structural equation modeling corroborate those of the ordered Probit model
9 The TLI of 1.001 and CFI of 1.00 are unusually high. These unusually high global fit measures are primarily a function of the fact that the χ2 from the model is very small compared to the degrees of freedom. Marsh, Balla, and McDonald (1988) point out that in the “true” model, the χ2 should equal the degrees of freedom. So as the χ2 from a model approaches the degrees of freedom, the model approaches the “true” model, which leads to high global fit measures. This essentially means the fit of my model is extraordinarily good, which is largely due to the simplicity of the model. Other global fit measures are below 1 (for instance, NFI = 0.995 and RFI = 0.986).
34 earlier: a finance professor’s opinion regarding the general efficiency of US stock markets has no influence on his propensity to actively invest. Instead, his investment behavior is best explained solely by his confidence in his own abilities to beat the market, regardless of his views of the efficiency of the markets. A relatively new vein of finance literature has begun to explore the importance of human behaviorial biases as they relate to investing (see Hirshleifer (2001) for a summary). There seems to be compelling evidence that confidence (or more accurately overconfidence) significantly influences investor behavior (see Barber and Odean (2001) for instance). My results confirm the notion that an investor’s level of confidence substantially influences his investing decisions and underscore the need for further work on the subject.
Conclusion
I use a comprehensive survey distributed to virtually every finance professor in the United States to assess our field’s collective opinion on the actual efficiency of US stock markets. The 642 respondents in the final sample seem to agree on two points: (1) US markets are not strong form efficient and (2) US stock markets are weak form efficient. However, they seem largely conflicted in their opinions about the semi-strong form efficiency of US stock markets. Opinions of respondents who specialize in market efficiency are generally in line with those of the overall sample. Acknowledging the cliché that actions speak louder than words, I also analyze respondents’ investment objectives and behavior to gain a deeper understanding of how they truly perceive the efficiency of US stock markets. I find that twice as many (roughly 40% compared to 20%) respondents passively invest than actively invest, which suggests that although they may be conflicted about how efficient US markets are, they generally behave in a way that suggests they accept markets as semi-strong form efficient. Again, respondents who specialize in market efficiency manifest similar investment objectives as that of the overall sample. However, derivatives specialists and behavioral finance specialists show a stronger propensity to actively invest, while corporate finance specialists show an even stronger propensity to passively invest.
35 Respondents’ investment objectives seem somewhat correlated with their perceptions of the efficiency of US markets. However, deeper consideration reveals that investment behavior has very little to do with a respondent’s opinions about the efficiency of US stock markets. Instead, investment objectives and behavior are statistically significantly driven by individuals’ confidence in their own abilities to beat the market, regardless of their opinions about the efficiency of US stock markets. I began the analysis with what I thought was an indisputable assumption, almost a stylized fact, that an investor’s belief about the efficiency of US stock markets is fundamental to his decision of whether to actively or passively invest. This notion, however, seems refuted, which suggests that individual investors don’t base their behavior and decisions on their perceptions of the efficiency of the markets in which they invest. Instead, investors base their decision to actively or passively invest on their confidence in their own abilities to beat the market. This is surprising and presents interesting ramifications. If investors, particularly finance professors, don’t care about market efficiency when they establish their investment goals and make their investment decisions, who does care about market efficiency? Is it possible that market efficiency is purely an academic construct with little real-world meaning to those who actually invest? One thing is certain: the findings presented here suggest that the vein of behavioral finance exploring the role of overconfidence in investing has merit and deserves further attention.
36 Table 1 Summary Statistics
This table presents summary statistics of the responses to the survey and the demographics of the respondents to the survey. The number of responses are presented in the first column. The second column contains the breakdown of useable responses by rank, gender, and marital status. The third through seventh columns presents the responses in columns one or two as a percentage of emails sent, valid emails, useable responses (columns five and six), and subgroup total, respectively.
Responses as a Percentage of Emails Valid Useable Subgroup Descriptive Stats Raw Number Sent Emails Responses Total Emails Sent 4,525 Valid Emails 4,465 98.7% Surveys Started 1,183 26.1% 26.5% Surveys Completed 870 19.2% 19.5% Total Useable Responses 642 14.2% 14.4% Rank (# of Responses) 642 100.0% Assistant 197 30.7% 30.7% Associate 197 30.7% 30.7% Full 171 26.6% 26.6% Endowed Chair 71 11.1% 11.1% Eminent Scholar 6 0.9% 0.9% Gender 631 98.3% Female 92 14.3% 14.6% Male 539 84.0% 85.4% Marital Status 632 98.4% Single - Never Married 60 9.3% 9.5% Single - Previously Married 43 6.7% 6.8% Married 529 82.4% 83.7% Median Age Range 40 - 49 $110K - Median Salary Range $120K Mean # of Total Pubs 13.37 Mean # of Top Pubs 1.84
37 Table 2 Opinions About Market Efficiency by Rank
This table reports the overall mean and median responses and mean and median responses across rank to the five questions in the survey related to the actual efficiency of US stock markets. Participants in the survey were presented five statements regarding the efficiency of US stock markets and asked to indicate how strongly they agreed or disagreed with the statements. The scale was from 1 (strongly agree) to 7 (strongly disagree).
Please indicate how strongly you agree or Overall Assistant Associate Full End. Ch. / Em. Sch. disagree with the following statements # (1=strongly agree to 7=strongly disagree) N Mean Med. N Mean Med. N Mean Med. N Mean Med. N Mean Med. 1 It is possible to predict future returns to US 635 5.30 6 196 5.02 6 197 5.48 6 167 5.55 6 75 5.01 6 stocks using only past returns 2 It is possible to predict future returns to US stocks using only past returns and publicly 636 4.46 5 196 4.08 4 197 4.59 5 168 4.74 5 75 4.44 5 available information 3 It is possible to predict future returns to US stocks using only past returns, publicly 634 2.68 2 195 2.28 2 196 2.84 2 168 3.07 3 75 2.48 2 available information, and private information 4 Investment returns are solely a compensation 633 4.29 5 196 4.19 4 196 4.24 4.5 167 4.37 5 74 4.51 5 for risk 5 Investment strategies exist that consistently beat average market returns without taking 638 4.46 5 196 4.29 4 196 4.53 5 170 4.54 5 76 4.55 4.5 above-average risk
7
38 Table 3 Market Efficiency Specialists’ Opinions About Market Efficiency
This table reports the number and percentage of respondents who specialize in market efficiency that answered 1 – 7 for each of the five questions in the survey related to market efficiency. Participants in the survey were presented five statements regarding the efficiency of US stock markets and asked to indicate how strongly they agreed or disagreed with the statements. The scale was from 1 (strongly agree) to 7 (strongly disagree).
Please indicate how strongly you agree or disagree with the following Strongly Strongly statements Agree Neutral Disagree Agree Neutral Disagree 1 2 3 4 5 6 7 Total Σ(1 - 2) Σ(3 - 5) Σ(6 - 7) It is possible to predict future returns to US stocks using only past returns Raw # 7 8 20 6 14 31 22 108 15 40 53 Percent 6% 7% 19% 6% 13% 29% 20% 14% 37% 49% It is possible to predict future returns to US stocks using only past returns and publicly available information Raw # 9 23 21 10 20 17 9 109 32 51 26 Percent 8% 21% 19% 9% 18% 16% 8% 29% 47% 24% It is possible to predict future returns to US stocks using only past returns, publicly available information, and private information Raw # 54 19 14 6 6 6 4 109 73 26 10 Percent 50% 17% 13% 6% 6% 6% 4% 67% 24% 9% Investment returns are solely a compensation for risk Raw # 1 15 21 6 28 21 17 109 16 55 38 Percent 1% 14% 19% 6% 26% 19% 16% 15% 50% 35% Investment strategies exist that consistently beat average market returns without taking above-average risk Raw # 9 11 25 12 19 22 11 109 20 56 33 Percent 8% 10% 23% 11% 17% 20% 10% 18% 51% 30%
39 Table 4 Respondents’ Propensities to Actively Invest by Rank
This table reports the number and percentage of respondents who responded 1 – 7 in across the various ranks (assistant professor, associate professor, full professor, and endowed chair or eminent scholar) to the question of the survey inquiring about respondents’ personal investment objectives. Participants in the survey were asked to indicate how strongly they agreed or disagreed with the following statement: “When I invest, my goal is to beat the market.” The scale was from 1 (strongly agree) to 7 (strongly disagree).
Please indicate how strongly you agree or disagree with the following statement: When I invest, my goal is to beat the Strongly Strongly market. Agree Neutral Disagree Agree Neutral Disagree Rank 1 2 3 4 5 6 7 Total Σ(1 - 2) Σ(3 - 5) Σ(6 - 7) Assistant Professor Raw Number 17 21 41 26 17 34 38 194 38 84 72 Percent 8.8% 10.8% 21.1% 13.4% 8.8% 17.5% 19.6% 20% 43% 37% Associate Professor Raw Number 12 19 26 31 17 39 52 196 31 74 91 Percent 6.1% 9.7% 13.3% 15.8% 8.7% 19.9% 26.5% 16% 38% 46% Full Professor Raw Number 17 16 24 23 22 30 37 169 33 69 67 Percent 10.1% 9.5% 14.2% 13.6% 13.0% 17.8% 21.9% 20% 41% 40% Endowed Chair or Emminent Scholar Raw Number 4 9 9 11 7 16 19 75 13 27 35 Percent 5.3% 12.0% 12.0% 14.7% 9.3% 21.3% 25.3% 17% 36% 47% Total Raw Number 50 65 100 91 63 119 146 634 115 254 265 Percent 7.9% 10.3% 15.8% 14.4% 9.9% 18.8% 23.0% 18% 40% 42%
40 Table 5 Respondents’ Propensities to Actively Invest by Specialty
This table reports the number and percentage of respondents who responded 1 – 7 across six specialties (asset pricing, behavioral finance, capital structure, corporate governance, and derivatives) to the question of the survey inquiring about respondents’ personal investment objectives. Participants in the survey were asked to indicate how strongly they agreed or disagreed with the following statement: “When I invest, my goal is to beat the market.” The scale was from 1 (strongly agree) to 7 (strongly disagree).
Please indicate how strongly you agree or disagree with the following statement: When I invest, my goal is to beat the Strongly Strongly market. Agree Neutral Disagree Agree Neutral Disagree Specialty 1 2 3 4 5 6 7 Total Σ(1 - 2) Σ(3 - 5) Σ(6 - 7) Asset Pricing 10 15 20 15 7 27 25 119 25 42 52 8% 13% 17% 13% 6% 23% 21% 21% 35% 44% Behavioral Finance 11 12 19 11 12 15 13 93 23 42 28 12% 13% 20% 12% 13% 16% 14% 25% 45% 30% Capital Structure 7 7 21 19 13 19 35 121 14 53 54 6% 6% 17% 16% 11% 16% 29% 12% 44% 45% Corporate Governance 9 13 21 22 13 23 42 143 22 56 65 6% 9% 15% 15% 9% 16% 29% 15% 39% 45% Derivatives 15 14 13 11 8 16 13 90 29 32 29 17% 16% 14% 12% 9% 18% 14% 32% 36% 32% Market Efficiency 9 13 17 12 14 21 22 108 22 43 43 8% 12% 16% 11% 13% 19% 20% 20% 40% 40%
41 Table 6 The Congruence of Respondents’ Opinions and Investment Objectives
This table reports the number and percentage of respondents in the sample who fall into subcategories created by a double sort of two questions on the survey. Respondents are sorted based on their responses to the statement, “When I invest, my goal is to beat the market,” and by their average response to the three statements on the survey regarding the weak form, semi-strong form, and strong form efficiency of US stock markets. Response scales to the statements are 1 (strongly agree) to 7 (strongly disagree). Respondents in the subcategories highlighted in gray are those whose investment objectives are highly congruent with their opinions of market efficiency, while respondents in the subcategories highlighted in black are those whose investment objectives are incongruent with their opinions of market efficiency.
Panel A: Raw Numbers Strongly Ave. response to the 3 statements about weak, semi-strong, and Strongly Agree strong form efficiency of US stock markets Disagree Σ (Gray) Σ (Black) 1 2 3 4 5 6 7 Total 264 137 Strongly Agree 1 3 10 11 6 12 5 3 50 2 2 15 8 20 13 4 2 64 3 3 14 26 31 15 7 3 99 4 2 10 13 23 28 9 6 91 5 2 7 4 25 14 9 2 63 my goal is to When I invest,
beat the market 6 3 11 16 27 40 19 2 118 Strongly Disagree 7 5 12 13 30 59 17 10 146 Total 20 79 91 162 181 70 28 631
Panel B: As Percentage of Total Respondents Strongly Ave. response to the 3 statements about weak, semi-strong, and Strongly Agree strong form efficiency of US stock markets Disagree Total Σ (Gray) Σ (Black) 1 2 3 4 5 6 7 42% 22% Strongly Agree 1 0% 2% 2% 1% 2% 1% 0% 8% 2 0% 2% 1% 3% 2% 1% 0% 10% 3 0% 2% 4% 5% 2% 1% 0% 16% 4 0% 2% 2% 4% 4% 1% 1% 14% 5 0% 1% 1% 4% 2% 1% 0% 10% my goal is to When I invest,
beat the market 6 0% 2% 3% 4% 6% 3% 0% 19% Strongly Disagree 7 1% 2% 2% 5% 9% 3% 2% 23% Total 3% 13% 14% 26% 29% 11% 4% 100%
42 Table 7 Respondents’ Investment Objectives as a Function of Their Opinions and Confidence
This table reports the mean response to the statement, “When I invest, my goal is to beat the market” for subcategories of respondents created by a double sort of two questions on the survey. Respondents are sorted based on their responses to the statement, “Given sufficient time and resources, I could implement a strategy that would consistently beat the market” and by their responses to the statements on the survey regarding the efficiency of US stock markets. Response scales to all the statements in the table are 1 (strongly agree) to 7 (strongly disagree). Panel A reports the mean responses for the subcategories created using participants’ responses to the question about weak form efficiency as a measure of their opinion regarding market efficiency. Panel B reports the mean responses for the subcategories created using participants’ responses to the question about semi-strong form efficiency as a measure of their opinion regarding market efficiency.
Panel A: Opinion = Weak form Efficiency Panel B: Opinion = Semi-Strong Form Efficiency Given sufficient time and Given sufficient time and resources, I could implement resources, I could implement Strongly a strategy that would Strongly Strongly a strategy that would Strongly Agree consistently beat the market Disagree Agree consistently beat the market Disagree 1 2 3 4 5 6 7 Total 1 2 3 4 5 6 7 Total Strongly Agree Strongly Agree 1 2.7 2.0 3.0 5.5 5.0 6.3 5.5 4.3 1 3.1 3.2 3.9 5.2 4.4 5.1 5.8 4.5 (Not Efficient) (Not Efficient) 2 4.0 3.0 4.4 5.5 3.0 5.8 6.0 4.1 2 2.7 3.2 3.5 4.7 5.2 5.3 5.9 4.7
3 1.5 2.8 3.3 5.6 4.0 4.0 4.4 3.6 3 1.3 2.8 3.4 3.5 4.7 5.0 6.4 4.3
4 1.6 3.4 3.3 4.3 5.0 6.4 6.6 4.3 4 1.0 3.8 3.2 3.8 4.8 4.8 6.1 4.6
information ilable 5 4.0 2.8 4.1 4.3 5.1 5.8 6.3 4.7 5 4.0 2.5 3.8 4.5 5.0 6.0 4.9
only past returns only past returns and 6 1.8 3.5 3.8 3.9 4.6 5.0 5.8 4.5 6 5.0 2.3 3.6 4.3 4.3 5.3 5.8 4.7 returns to US stocks using returns to US stocks using returns It is possible to predict future It is possible to predict future It is possible publicly ava Strongly Strongly Disagree 7 3.2 3.9 3.4 4.1 5.2 5.1 6.1 5.1 Disagree 7 1.0 3.7 4.0 5.3 5.8 7.0 5.0 (Efficient) (Efficient) Total 2.6 3.3 3.6 4.4 4.8 5.1 6.0 4.6 Total 2.6 3.3 3.6 4.4 4.8 5.1 6.0 4.6
43 Table 8 Explaining Investment Objectives – Ordered Probit Analysis
This table reports the results from estimating two iterations of an ordered Probit model. The dependent variable in both iterations is a participant’s response to the statement, “When I invest, my goal is to beat the market.” The independent variables are the participant’s responses to the three questions on the survey about the weak form, semi-strong form, and strong form efficiency of US stock markets; the participant’s response to the question about his confidence in his own abilities to beat the market; three terms that interact the participant’s confidence on his opinions about market efficiency; gender and marriage binary variables (1 = male and 1 = married); the respondent’s rank and age; and binary variables indicating the respondent specializes in asset pricing, behavioral finance, capital structure, corporate governance, derivatives, or market efficiency. In the first iteration, the interaction variables are omitted. In the second iteration, they are included in the model.
Ind. Variable Est. p Est. p Opinion of Weak form Efficiency -0.049 0.207 0.135 0.138 Opinion of Semi Strong Form Efficiency 0.130 0.002 -0.101 0.333 Opinion of Strong Form Efficiency -0.076 0.009 -0.191 0.019 Confidence in Own Ability to Beat Market 0.326 <.0001 0.253 0.001 Confidence x Weak form Efficiency -0.041 0.032 Confidence x Semi Strong Form Efficiency 0.052 0.010 Confidence x Strong Form Efficiency 0.025 0.113 Gender (0 = Fem / 1 = Male) 0.064 0.614 0.069 0.588 Marital Status (0 = Single, 1 = Married) -0.006 0.957 0.009 0.940 Rank 0.077 0.172 0.076 0.181 Age -0.052 0.324 -0.060 0.263 Asset Pricing Binary 0.112 0.356 0.133 0.275 Behavioral Finance Binary -0.117 0.369 -0.142 0.288 Capital Structure Binary 0.148 0.199 0.171 0.140 Corporate Governance Binary 0.063 0.562 0.052 0.634 Derivatives Binary -0.424 0.001 -0.403 0.002 Market Efficiency Binary 0.135 0.282 0.155 0.217
44 This figure presents the path diagram, along with standardized parameter estimates and p-values, for the structural equation modeling employed to test what drives a respondent’s propensity to actively invest. The relationships are modeled such that a respondent’s propensity to actively invest is a latent construct that is reflected by the participant’s responses to the questions inquiring how frequently he buys stocks, sells stocks, and sells ETFs. This latent construct is a function of one observed independent variable (a respondent’s confidence in his own ability to beat the market) and one latent independent variable (a respondent’s opinion regarding the efficiency of US stock market), which is reflected by his responses to the two statements about the weak form and semi- strong form efficiency. Overall measures of fit, including χ2, SRMR, TLI, and CFI, are presented below the path diagram.
Frequency with which respondent Opinion on Weak buys stocks form Efficiency of 0.830 (<0.001) Opinion on 0.917 US Stock Markets <0.001 Efficiency of 0.020 US Stock (0.689) Markets Propensity to 0.901 Frequency with <0.001 which respondent 0.896 Actively Opinion on Semi- (<0.001) Invest sells stocks strong form Efficiency of US Stock Markets 0.301 <0.001 0.45 Confidence in <0.001 Frequency with own abilities to which respondent beat the market sells ETFs
2 χ = 6.704, dfχ2 = 7, pχ2 = 0.460 SRMR = 0.0131 TLI = 1.001 CFI = 1.00
Figure I Structural Equation Modeling Path Diagram
45 CHAPTER 3
WHAT REALLY MATTERS WHEN BUYING AND SELLING STOCKS?
Introduction
What really matters when considering whether to buy or sell a specific stock? Although this subject has received tremendous attention in academia, we seem to have reached little consensus about what truly matters and what does not when valuing securities. My objective is to survey a broad target population of finance professors in America to determine our profession’s collective opinion regarding what really matters most and what matters least when buying and selling stocks. Specifically, I aim to discover what valuation techniques, asset-pricing models, market anomalies, firm characteristics, corporate events, seasonal variables, and other information are most and least important when finance professors are considering buying and selling a stock. The question of what matters when valuing stocks has been either explicitly or implicitly the subject of many, many papers in finance. Any paper that deals with market efficiency, anomalies to market efficiency, or asset-pricing models necessarily addresses the subject. As a simple example, CAPM ultimately suggests that a stock’s β – its covariance with the market divided by the market’s variance – is the only firm-specific variable that really matters. However, given the broad array of asset-pricing models and anomalies to market efficiency – each ostensibly suggesting some unique variable or factor that is highly relevant when buying or selling stocks – what is an investor to do? Is β really the only thing that matters? Or perhaps a stock’s correlations with the Fama and French (1993) factors are also extremely important. What about Carhart’s (1995) momentum factor? Could a stock’s dividend-yield also be important (Fama and French (1998) and Shiller (1998))? Or should an investor also look at a stock’s market capitalization (Banz (1981)), PE ratio (Basu (1977)), book-to-market equity (Stattman (1980)), or 52-week high and low (George and Hwang (2004))? What about it’s return over the past six months (Jegadeesh and Titman (1993))? Past 12 months? Is it the stock’s past returns or
46 the industry’s past returns that matter (Moskowitz and Grinblatt (1999))? And, what about analysts – recommendations, target prices, earnings forecasts, etc.? Should an investor care what they say? What’s an investor to do? What is to be made of the jumble of literature on the subject? What matters most? What matters least? I see no better source for answers to these questions than finance professors themselves. They are the ones who read and perform the research on the subject. Further, they are also market participants themselves, investing their own money and, in some cases, acting as professional money managers. Additionally, part of their responsibilities includes teaching undergraduate and graduate students the principles of investing, which presumably includes pointing students toward the variables that are most relevant when making decisions regarding the purchase and sale of stocks. As I survey finance professors about their opinions regarding what matters most when considering whether to buy or sell a stock, I also assess how much practical investing experience finance professors have with stocks, margin trading, short selling, options, and futures. Considering their roles and influence as researchers and especially as teachers, it seems pertinent to inquire whether finance professors represent a vastly experienced group on the subject of investing. Or, are they largely a group of theorists in an ivory tower with little real-world investing experience. Surely, those reading their research and listening to their lectures would be interested to know. The remainder of the essay proceeds as follows. Section II offers further motivation including a review of the literature on subject and possible bases for finance professors’ investing strategies. Section III provides a brief review of the use of surveys in finance. I describe the subjects, surveys, beta testing, and distribution in Section IV. Results are presented in Section V. Section VI concludes.
Background
Literature on the Subject My work is most closely related to Welch (2000) and Haddad and Redman (2005). Welch (2000) surveys financial economists to develop a consensus estimate of the equity premium over a set of future horizons. As a secondary (arguably tertiary)
47 matter, he also asks them a set of questions about “issues that are commonly debated in the academic literature.” This set of questions covers a broad spectrum of topics, and only one of the questions relates directly to my objective. He asked respondents to indicate the degree to which they agreed or disagreed with the following statement, “I believe that size/ book-market/ price-earnings / momentum factors are stationary enough, so that they will work well in the future in explaining cross-sectional expected return differences.” On a scale from 1 (Strongly Agree) to 5 (Strongly Disagree), the mean response was 2.77. Haddad and Redman (2005) survey finance, accounting, and economics professors. Their study focuses on four areas related to professors as investors: (1) current asset allocation, (2) expected sources of retirement income, (3) expected retirement asset allocation, and (4) types of financial instruments used. The only portion of their survey that is directly related to my work is the questions they ask about respondents’ experience with various investments, which is my secondary topic of interest here. My work differs from and improves upon Welch (2000) and Haddad and Redman (2005) in three major ways. First and most importantly, neither of the previous authors concentrated on the primary research question of my paper. Welch asks a single question on the subject, while Haddad and Redman (2005) ask no questions on the topic. In actuality, then, mine is the first work to offer an in depth analysis of the question of what finance professors collectively opine are actually most and least important when investing. Second, my sample size and target populations are much broader than the previous authors’. Welch’s (2000) initial survey was sent to professors at 11 universities and made available on his personal website. He also made a later version of the survey available on the website of the Journal of Finance. His final sample is 226. Haddad and Redman (2005) survey finance, accounting, and economics professors. The total number of useable responses in their sample is 571, of which only 201 are finance academicians. In contrast to these previous works, an invitation to participate in my survey is sent to virtually every finance professor at accredited, four-year universities in the United States. My useable sample is 642, all of which are finance professors. I effectively triple the
48 sample size of finance professors compared to these previous studies. Hence, any questions from my survey that coincidentally overlap with theirs still contain new information given the significant increase in the sample size. Third, I ask substantially more conditioning questions. Previous authors ask some of the standard demographic questions and a few questions that separate finance professors by areas of specialization. My survey contains a fairly comprehensive section of demographic and other conditioning questions. The most useful of these conditioning questions will be discussed as they are used later in the paper.
Possible Decision Variables of Finance Professors A critical part of my efforts is to identify the list of relevant valuation techniques, asset-pricing models, market anomalies, and other potentially important information. To do this I survey the literature on asset pricing, equity valuation, and anomalies to market efficiency. Each of the following has at some point been identified in academic literature as potentially aiding an investor who seeks to maximize his returns and minimize his risk. Virtually all have been both supported and refuted in subsequent literature. I offer only a brief explanation of each. There are a number of stock valuation procedures that have become commonplace in investing textbooks. They include zero-growth, constant-growth, and variable-growth dividend valuations models and numerous multiples valuation models including the price-to-earnings (PE), price-to-cash-flow (P/CF), price-to-sales (P/S), and price-to-book value (P/BV) multiples models. Additionally, there are a number of asset pricing models that have been proposed in academia over the past 45 years. Some of the more common models include the capital asset pricing model (CAPM) (Sharpe (1963 and 1964) and Lintner (1965)), arbitrage pricing theory (APT) (Ross (1976) and Chen, Roll, and Ross (1986)), the Fama and French (1993) three-factor model, and the Carhart (1997) four-factor model. The commonality among all asset-pricing models is that stock returns are a compensation for rewarded risk. The only way to increase average returns, then, is to invest in assets that load more heavily on the risk factors the model suggests are rewarded. Accordingly,
49 investors accepting a particular model will seek stocks that load more heavily on the risk factors of the model in an attempt to maximize return. Aside from asset-pricing models, there have been numerous papers indicating certain firm characteristics or events may help investors to develop abnormal return generating strategies. Examples of these firm characteristics or events include market capitalization (Banz (1981) and Reinganum (1981)), book-to-market equity (Stattman (1980) and Rosenberg, Reid, and Lanstein (1981)), dividend-yield (Fama and French (1988) and Campbell and Shiller (1988)), price-earnings (Basu (1977 and 1983)), returns over the past six to twelve months (Jegadeesh and Titman (1993)), returns over the past three to five years (DeBondt and Thaler (1985)), returns to a stock’s industry over the past six to twelve months (Moskowitz and Grinblatt (1999)), the 52-week high (George and Hwang (2004)), and dividend increases and initiations (Healy and Palepu (1988) and Benartzi, Michaely, and Thaler (1995)). Authors have also identified seasonal trends in stock returns. French (1980) concludes that returns to stocks on Monday are lower than returns to stocks on other days of the week. Keim (1983), Reinganum (1983), and Roll (1983) show that firms, especially small firms, that experience negative returns during a given year experience a higher return around the turn of the year than do firms that experience positive returns during the year, while Ogden (2003) finds that the majority of annual stock returns occur from October to March. According to this vein of the literature, investors may need to pay particular attention to the day of week or month of the year when they are considering buying and selling stocks. There is also evidence of drift subsequent to earnings announcements – the stock prices of firms issuing earnings announcements that are surprisingly high drift upwards and vice versa (see Ball and Brown (1968), Bernard and Thomas (1990), and Chan (2003)). In light of this finding, investors may analyze the most recent earnings announcement relative to expectations to make investment decisions. It is not uncommon for investors to rely on the opinions of other informed investors to make their decisions. Jegadeesh, Kim, Krische, and Lee (2004) report that changes in consensus analyst recommendations are useful in predicting stock returns. Brav and Lehavy (2003) find that changes to analysts’ target prices can be informative
50 about future price movements even when recommendations are unchanged. Asquith, Mikhail, and Au (2005) conclude that, in addition to changes in recommendations and target prices, the qualitative content of the analyst report can also be useful in predicting stock price movement. Perhaps then, investors pay particular attention to changes in analyst recommendations and target prices and to the content of their reports. Diether, Malloy, and Scherbina (2002) report that dispersion in analyst earnings forecasts is inversely related to future returns. Perhaps, then, investors should consider not only analyst recommendations and target price revisions, but should also consider the dispersion among the analyst opinions before buying or selling a stock. Corporate events have also been identified that seem to lead to either unusually high or low performance. Evidence indicates firms who split their stock or repurchase their stock subsequently experience higher performance (see Ikenberry, Lakonishok, and Vermaelen (1995) and Ikenberry, Rankine, and Stice (1996)). Investors accepting these findings will prefer stocks that have recently split or repurchased their own stock. Alternatively, firms that carryout IPOs and SEOs, firms that switch their listings to the NYSE or AMEX, and firms that make stock-funded acquisitions experience relatively poor performance subsequent to the events (see Loughran and Ritter (1995), Speiss and Affleck-Graves (1995), Dharan and Ikenberry (1995), Asquith (1983)). Investors familiar with these findings may shy away from firms involved in such corporate events. Recently, authors have begun to explore the relationship between investor sentiment and stock returns. Baker and Wurgler (2004) report that beginning-of-period sentiment measures can help predict returns to stocks whose valuations are highly subjective or difficult to arbitrage. Specifically, returns to small stocks, young stocks, high volatility stocks, unprofitable stocks, non-dividend paying stocks, extreme growth stocks, and distressed stocks are inversely related to the level of investor sentiment at the beginning of the period. Investors aware of their findings may factor sentiment measures, such as consumer overconfidence, into their investment decisions. I include each of the above valuation techniques, asset-pricing models, market anomalies, ratios, firm characteristics, corporate events, and information in my survey to
51 determine which of these finance professors consider the most useful when deciding whether to buy or sell a stock.10
Surveys in Finance Literature
The use of survey instruments has historically been infrequent in finance literature. There are three primary reasons for this. First, as Friedman (1953) articulated and Brav, Graham, Harvey, and Michaely (2005) reiterated, economic models are not conditional on the underlying agents’ understanding why they do what they do. As long as statistical inference using measurable data is able to adequately support or refute economic models and hypotheses, surveying the agents is unnecessary. Second, finance is privileged with access to an abundance of archival data, which provides a number of statistically desirable benefits in formal hypothesis testing. Therefore, use of archival data is generally preferable to survey-based data. Third, as Welch (2000) articulated, almost all surveys have shortcomings and flaws, which create skepticism regarding inferences from survey data. However, recent articles in respected journals such as Journal of Finance, Journal of Financial Economics, and Journal of Accounting and Economics have employed survey instruments as critical components of their research methodologies. Specifically, Brau and Fawcett (2006) survey CFOs to determine the primary motives behind firms’ decisions to go public, Brav, et al. (2005) survey financial executives to assess the determinants of dividend and share repurchase decisions, and Graham, Harvey, and Rajgopal (2005) survey executives to illuminate the factors motivating reported earnings and disclosure decisions. Further, the Journal of Applied Finance published two survey articles (Hartikainen and Torstila (2004), and Jorgensen and Wingender (2004)) in a single 2004 issue. Earlier articles based on surveys include Pinegar and Wilbricht (1989), Trahan and Gitman (1995), Welch (2000), Graham and Harvey (2001), and Krigman, Shaw, and Womack (2001). So although surveys are infrequent in finance
10 Obviously, the above list is not exhaustive. For the sake of obtaining a high response rate, we kept the survey as brief as possible, which necessitated the exclusion of some other potentially relevant information when considering buying and selling stocks such as turnover, institutional ownership, and implied volatility to name a few.
52 literature, they have been published in some of the top journals in the field, are gaining greater acceptance, and have provided useful insights on a variety of subjects, particularly regarding the practice of finance. More importantly, without access to the private brokerage accounts of finance professors, a survey instrument is the only way to address the research objective of this essay.
Survey Subjects, Description, and Distribution
The Subjects The subjects of the study are finance professors in the United States. To identify the professors for the survey, I use the list of all regionally accredited U.S. universities compiled by the University of Texas at Austin.11 For each four-year university or college, I collect the names and email addresses of all professors of finance by visiting the relevant academic college and department websites at the universities and colleges on my list.
The Surveys The survey contains questions that fall into the following five categories: (1) conditioning variables, (2) indicators of opinion on market efficiency, (3) opinion on market efficiency, (4) propensity to passively invest, and (5) investment strategies and experience. The segment of the survey on investment strategies and experience is the primary focus of this paper. A copy of the full survey is contained in Appendix A.
Beta Testing Successful survey construction involves beta testing, in which initial drafts of the survey are administered to test groups to determine the intelligibility, reliability, and validity of the questions on the survey. I beta test the survey on my colleagues - PhD students in the College of Business at Florida State University. After analyzing results from the beta testing and receiving feedback from the beta subjects, I modify the survey in close consultation with my dissertation committee members.
11 http://www.utexas.edu/world/univ/
53
Distribution Although most surveys in finance implement a hard-copy distribution of their surveys and achieve a respectable response rate (Brau et al. (2006) – 18%; Brav et al. (2005) – 16%; and Graham and Harvey (2001) – 9%)12, paper surveys are largely being replaced by electronic surveys distributed and collected via the Internet. Electronic surveys offer a number of advantages compared with traditional paper surveys including: (1) efficiency, (2) expediency, (3) accuracy, and (4) cost savings (see Wright (2005), Van Selm and Jankowski (2006), Medlin, Roy, and Chai (1999), and Schaefer and Dillman (1998)). The primary disadvantages of electronic distribution are unrepresentative sampling and lower response rates. In light of the considerable advantages it offers, I choose the electronic method of survey distribution and collection. The unrepresentative sampling concern should not be a problem since virtually every professor in the United States has access to the Internet and has an email account that is checked regularly. And although there may be some group that is less inclined to respond to an electronic survey, this issue is not unique to the electronic delivery format. Further, Schaefer and Dilman (1998) demonstrate that multi-mode (a combination of electronic and paper) contact with respondents did not significantly increase response rates. Also, my response rate is at the high end of the typical range in finance, and my sample size is actually relatively large. I create and distribute the survey electronically using surveyZ.com and qualtrics.com13. I incorporate strategies that have proven helpful in increasing response rates to electronic surveys. Schaefer and Dillman (1998) state on p. 380 that, “the most powerful determinant of response rates is the number of attempts made to contact a sample unit.” They also argue that personalization increases response rates, suggesting the need to send emails addressed to the potential respondent rather than to a mailing list. Crawford et al. (2001) suggest that, in the electronic environment, reminders subsequent to the initial invitation to participate in the survey are more effective when sent two days after the initial invitation as opposed to one week as suggested by Dillman (1978) for
12 Some have implemented a combination of hard-copy and electronic distribution methods. 13 I express appreciation to surveyZ.com and qualtrics.com who provided their survey software free of charge.
54 mail surveys. They also find that offering an accurate estimate (as opposed to an inaccurately low estimate) of the time it takes to complete the survey leads to a higher number of completed surveys.14 They further argue that including a progress bar in the survey is recommended.15 Bosnjak and Tuten (2003) provide evidence that of four possible incentive payment structures (pre-paid, post-paid, prize-drawing, and no incentive), the prize-drawing incentive structure is most effective in eliciting completed responses (Brau and Fawcitt (2006) and Brav et al. (2005) use this incentive structure). Following the suggestions of the above literature, I implement a series of emails inviting participation in the study. Two days before (day – 2) distributing the official invitation to participate electronically, I send an email, the “pre-mail,” to all potential respondents. The pre-mail explains (a) that they will be receiving an electronic invitation to respond to the survey, (b) that their responses will be strictly confidential, (c) the purpose of the survey, (d) when they can expect to receive the invitation email, (e) an estimate of how long it will take to complete the survey, (f) and the incentives for responding. The incentive for responding is entry into a drawing for $500. The premail also contains a link to the survey and gives the recipients the option to take the survey at that time if they prefer. Two days after the pre-mail, I send the official invitation email (day 0) including the link to the survey hosted at qualtrics.com and repeating the relevant information from the pre-mail. Two days after the initial invitation survey (day 2), I send a “post-mail” to remind potential respondents of the opportunity to fill out the survey. Copies of the pre- mail, invitation email, and post-mail are contained in Appendices B, C, and D, respectively. To further increase response rates, emails are sent in a manner such they are addressed individually to each potential respondent. Also, a progress bar is included at the bottom of each page of the survey.
14 Crawford et al. (2001) find that providing an inaccurately low estimate of the time requirement elicited more responses, but more of those respondents dropped out of the survey before completion. Providing an accurate estimate led to fewer respondents starting the survey, but more of those who started actually finished the survey. 15 They actually found that including a progress bar led to lower completion rates, but they argued their results were an artifact of the survey structure, which included many open ended questions in the beginning stages of the survey. In spite of their results, they endorse the inclusion of a progress bar.
55 The premail was sent on February 19, 2007, the invitation email was sent on February 21, 2007, and the postmail was sent on February 23, 2007. The survey was deactivated on February 26, 2007.
Results
Response Rates Emails inviting participation in the survey were sent to 4,525 professors. 60 of the email addresses were invalid. 1,183 professors started the survey, which is a started response rate of 26.49%. 870 professors completed the survey – i.e., they answered the final question on the survey – which represents a completion rate of 73.54% and a completed response rate of 19.48%. In order to enter the final data set, I require a respondent to meet five criteria: 1) s/he must answer yes to the consent question, 2) s/he must be a finance professor, thus eliminating professors of law, economics, and other disciplines,16 3) s/he must hold a Ph.D. or DBA, 4) s/he must be of the rank of assistant professor, associate professor, full professor, endowed chair, or eminent scholar, and 5) s/he must answer at least one of the five questions on the last page of the survey. There were two exceptions to this final criterion. Two respondents answered well over half the questions on the survey but did not answer the last five. I allowed these two respondents to enter the final data set. Aside from these outliers, the respondents who failed to answer one of the last five questions universally answered less than half the questions on the survey, which casts doubt on the credibility of their responses to the questions they did answer. I also removed three professors whose responses contained glaringly inconsistent answers to questions. 17 Table 9 (Table 1 Reproduced) presents summary statistics on the responses to the survey. The final useable data set consists of 642 respondents. Of the 642, 197 are
16 Some of the websites from which email addresses were collected made it impossible to distinguish finance professors from professors of other fields, such as business law and economics. Because of this, the survey was sent to and completed by some professors of other disciplines. 17 For instance, one professor responded that he had published a grand total of one article in peer-reviewed journals, but he indicated later that he had published 53 articles in peer-reviewed journals that support market efficiency.
56 assistant professors, 197 are associate professors, 171 are full professors, 71 are endowed chairs, and 6 are eminent scholars. Almost 15% of respondents are female. Slightly more than 83% of respondents are married. The median age of respondents is between 40 and 49. The median 9-month salary of respondents is $110,000 to $119,999. Respondents have an average of 13.37 (median = 10) articles published in peer reviewed journals and an average of 1.84 (median = 0) articles published in the Journal of Business, Journal of Finance, Journal of Financial Economics, Journal of Financial and Quantitative Analysis, or Review of Financial Studies.
What Matters Most I now turn to the primary objective of the paper – to determine what finance professors believe is most important when contemplating stock purchases and sales. To assess this information, I posed questions to respondents in the following general form: “When you are considering buying or selling a stock, how important are the following?” The respondents were then presented a list of valuation techniques, asset-pricing models, and other potentially useful information. The scale was 1 (not important at all) to 7 (extremely important). Before analyzing the relative importance of the individual valuation techniques, asset-pricing models, market anomalies, and other potentially relevant information, I first obtain a reading on what general categories of information are most important. Each of the individual variables included in the survey to which a finance professor may look before making a stock purchase or sale are grouped into one of the following 14 categories: dividend valuation models, other valuation models, asset pricing models, a stock’s correlation with the market, a stock’s correlation with other asset- pricing factors, firm characteristics, Jegadeesh and Titman (1993) momentum variables, Moskowitz and Grinblatt (1999) momentum variables, George and Hwang (2004) momentum variables, DeBondt and Thaler (1985) reversal variables, analyst information, corporate events, sentiment measures, and seasonality. I report the overall means and means across rank for these fourteen groups of potentially useful investment information in Table 10.
57 Not surprisingly, a stock’s correlation with the market tops the list. But the remainder of the list holds numerous surprises. Firm characteristics and momentum follow the stock’s correlation with the market at the top of the list, while traditional dividend valuation models, asset pricing models, and loadings on asset pricing factors are at the bottom of the list along with seasonality information such as the day of the week and month of the year. I next drill down to discover the specific ratios, characteristics, and other information that finance professors find most useful when investing. Table 11 presents the overall mean responses and mean responses across rank to all the questions on the survey dedicated to topic at hand. One striking result from Table 11 is that the most fundamental valuation techniques and asset-pricing models in finance are unimportant in professors’ decisions of whether to buy or sell a stock. The dividend growth valuation models (#1 and 2), Arbitrage Pricing Theory (#6), the Fama and French 3-factor (#7), and the Carhart 4- factor (#8) models are all noticeably unimportant. Each has a mean response of approximately two. The Capital Asset Pricing Model (#5) received a mean response of 2.67. Apparently, finance professors are comfortable teaching these core theories and models and using them in their research, but they don’t really believe they are of much practical use. It is worth noting, however, that only two questions received an average response greater than 3.5, one of which is the stock’s correlation with the market (#9). So even though CAPM doesn’t seem to have tremendous credibility in professors’ eyes, β still seems relevant. The day of the week (#42) and month of the year (#41) are also conspicuously unimportant. The mean responses to these two factors are 1.54 and 1.94, respectively. Regardless of the academic robustness of seasonal anomalies, finance professors do not seem to believe they are of much actual benefit in their personal investing. Aside from the stock’s correlation with the market, other factors that appear relatively more important to finance professors overall include (with mean responses in parentheses): #16 – the stock’s PE ratio (3.71), #13 – the stock’s market capitalization (3.38), #37 – mergers and acquisitions activity (3.31), #17 – the stock’s return over the past six months (3.2), #19 – the stock’s return over the past year (3.19), and #14 – the
58 stock’s book-to-market ratio (3.17). The preliminary indications of the data, then, are that firm characteristics, such as PE, market capitalization, and book-to-market ratios, along with momentum related information are most important to finance professors when investing in stocks, while traditional valuation models, asset-pricing models, and seasonal anomalies are the least important. A striking result, however, from Table 11 is that even the most important variables do not show a great deal of significance. The most important variable (the stock’s correlation with the market) received a mean response of only 3.82. Since the scale is from 1 (not important at all) to 7 (extremely important), this means that even the most important variable leans more towards the “not important at all” side of the scale. This implies that the factors at the top of the list really are not “most important;” rather, they are merely the least unimportant. There are at least three possible explanations for this. First, perhaps the variables on the list truly are of relatively little importance to finance professors when they invest. Second, finance professors likely have differing opinions on the relative importance of the variables in the survey, so that there is simply no consensus that any of the variables is extremely important in the eyes of all the respondents when they invest. This is clearly plausible given the voluminous and oft-times contradictory evidence in our literature. The third possible explanation, however, seems most likely. As several respondents pointed out in emails, many professors merely passively invest. For those who passively invest, very few of the items on my list will be of much significance when they buy or sell a stock, since passive investors likely invest mostly in mutual funds instead of buying and selling individual stocks. A sizeable portion of the respondents embracing a passive investment philosophy would explain why even the highest mean response in Table 11 is only 3.82. To investigate the issue further, I explore the lifetime investment experience of finance professors.
Lifetime Investing Experience Table 12 reports the mean responses of participants to the questions about their lifetime investing experience. It contains both overall means and medians and means and medians across rank. The median finance professor has purchased an individual stock
59 only between 10 and 19 times over his or her investing lifetime. Not surprisingly, the median is higher for professors of higher ranks than for assistant professors, which may be primarily a function of age. Aside from buying stocks, the median professor has never engaged in any of the other investing activities included in the survey. I am struck by the relative inexperience of finance professors in Table 12, especially outside of simply buying and selling stocks. To view the issue from another angle, I focus on the percentage of respondents that have no experience with the various assets and investing techniques. Table 13 reports this data and shows that 14.5% of finance professors have never purchased a stock. Outside of stocks, 57.4% have never purchased an ETF, 72.6% have never margin traded, 78.6% have never engaged in short selling, 65.4% have never purchased a call option, 73.0% have never purchased a put option, 79.9% have never written a call option, 87.0% have never written a put option, and 84.5% have never entered into a futures contract. Finance professors obviously largely stick to the basics of simply buying and selling stocks. Still, the median professor has only purchased a stock between 10 and 19 times over his or her lifetime. It is reasonable that, with such limited stock experience, a finance professor would find very few, if any, of the variables in the survey to be of much importance when buying and selling stock. What can possibly be important when buying and selling stocks if you don’t actually buy and sell stocks? As a secondary matter, this analysis also suggests that finance professors may often find themselves in the undesirable position of teaching a subject such as short selling, margin trading, or derviatives with which they have very little personal experience, which is unfortunate both for the teacher and the student. In light of the provocative results here, I investigate finance professors’ relative inexperience a little further before returning to the primary focus of the paper. Previous studies have shown a difference between the genders when it comes to investing. For instance, Barber and Odean (2001) demonstrate that men tend to me more overconfident than women in the realm of investing. Perhaps, then, men have more investing experience than women. Table 14 shows the mean and median responses to the questions about investing experience sorted by gender. It also reports the statistical significance of the differences between the two genders (parametric t-tests of the
60 differences in means and nonparametric Wilcoxon tests of the differences in medians). The table demonstrates that men on average do have significantly more lifetime investing experience than do females with all of the investment vehicles and techniques in the survey, which is consistent with Barber and Odean (2001). However, the gender differences may be entirely related to other demographic variables that are correlated with gender. For clarification, I conduct a set of ordered Probit models to determine which of the demographic variables are significantly related to the dispersion in investing experience. The lifetime investing experience with each of the assets or techniques in the survey are the dependent variables. The independent variables in the ordered probit models include gender, age, rank, marital status, total number of publications, salary, and binary variables representing specialization in asset pricing, derivatives, financial engineering, or market efficiency. My priors are the following (with the expected sign of the parameter estimate in parentheses). Men have more investing experience (gender = positive). Age and rank are positively correlated with investing experience (age and rank = positive). Single professors will have more experience as Barber and Odean (2001) demonstrated that his group of single investors, specifically single men, are more overconfident in the realm of investing (marital status = negative).18 Professors with high publications and higher salaries should have more experience as they are likely to be older and more knowledgeable (number of publications and salary = positive). I also expect that professors specializing in asset pricing, derivatives, and financial engineering, should have more experience (positive for all these binary variables). I am unsure about the direction of the estimate for the binary variable representing the group specializing in market efficiency. I also include as independent variables the responses to two other questions on the survey: Please indicate how strongly you agree or disagree with the following statements (1 = strongly agree to 7 = strongly disagree) – (1) given sufficient time and resources, I could implement an investing strategy that would consistently beat the market and (2) when I invest, my goal is to beat the market. It seems reasonable to believe that professors who are more confident in their abilities to beat the market and who are
18 This may be negated by the fact that single professors are likely younger than married professors.
61 actively trying to beat the market will have more experience with the various investments and techniques. Because of the scales of the questions, this would be manifest by parameter estimates that are negative and significant in the ordered Probit models. Table 15 reports the results of the ordered Probit testing. The relationships in the table are generally consistent with expectations. Focusing specifically on professors’ experience investing in stocks, significant variables include: gender, age, total articles, salary, specialization in derivatives, and trying to beat the market. The coefficients of each of these variables, except salary, are of the predicted sign. A professor’s experience investing in stocks increases if the professor is a man and as the professor’s age and total articles increase. The professor’s experience with stock investing also increases if his or her specialty is derivatives and if he or she is actively trying to beat the market when investing. His or her experience with investing in stock decreases with salary. The independent variables that show widespread significance in relation to professors’ experience with other investments and techniques include gender, age, specialization in derivatives, confidence that one can beat the market, and actually trying to beat the market. Men continue to show more experience with the various assets and techniques than women, even when controlling for the other variables. Again, age is positively related to experience. Also, specialization in derivatives is related to more experience. Finally, believing that one can beat the market and actually trying to beat the market are both associated with more investing experience. In brief summary, finance professors have less experience with stocks, margin trading, short selling, futures, options, and commodities than one might suspect. Their inexperience with stock purchases and sales may help explain the earlier findings that none of the variables appear particularly important to finance professors when they invest. Outside of stocks, finance professors are especially inexperienced with the other investments and techniques. Further, men have more investing experience than women. Also, age, confidence in one’s ability to beat the market, actually trying to beat the market, and a specialization in derivatives are positively associated with investing experience.
What Matters to Active Investors
62 The analysis of finance professors lifetime investing experience suggests there may be a large number of respondents who are mostly passively investing. Anticipating this possibility, I asked respondents to the survey to indicate the degree to which they agree or disagree with the following statement on a scale of 1 (strongly agree) to 7 (strongly disagree): “When I invest my goal is to beat the market.” No question on a survey is perfect. Clearly, this question is open to interpretation, but I believe it provides a simple reading on the degree to which respondents passively or actively invest. Related to the primary objective of the paper, the responses to this question allow me to categorize finance professors as either passive or active investors, which should allow for more information regarding what finance professors believe is most and least important when investing. Specifically, the opinions of the active traders should be of more interest than those of passive investors. I categorize professors based on their responses to the above conditioning question. Those answering 1 -3 are categorized as “active investors,” while those answering 5 – 7 are categorized as “passive investors.”19 Those who responded with a 4 are considered neutral and are dropped. Table 16 presents the factors in the survey ranked by the average responses of the active investors and offers a reading of what the professors who are most actively trying to beat the market believe are most important when investing. After making the active-passive distinction, some of the variables begin to push towards the “extremely important” side of the axis. The top ten most important factors to active investors when considering buying and selling a stock are (with mean responses of active investors in parentheses): (1) the stock’s PE ratio (4.29), (2) the return to the stock over the past six months (4.04), (3) the stock’s correlation with the market (3.95), (4) mergers and acquisitions activity (3.92), (5) the return to the stock over the past year (3.87), (6) the stock’s market capitalization (3.86), (7) the stock’s 52-week high (3.81), (8) the stock’s 52-week low (3.79), (9) the stock’s most recent earnings announcement
19 For robustness, I also classified respondents as active if their response was either 1 or 2 and passive if their response was either 6 or 7. Those answering 3, 4, or 5 were considered neutral and dropped. This more polarized classification of active vs. passive investors, obviously, reduced the number of respondents qualifying for each classification (118 for active vs. 253 for passive).
63 compared to analyst expectations (3.78), and (10) the return to the stock’s industry over the past six months. 20 Considering five (#2, 5, 7, 8, and 10) out of the top ten are related to momentum, it appears that finance professors actively trying to beat the market may be implementing some layer of momentum investing based on Jegadeesh and Titman (1993), Moskowitz and Grinblatt (1999), or George and Hwang (2004). Momentum is one of the most robustly supported anomalies in academic literature, and apparently finance professors accept it as more than an artifact of the data. Further, firm characteristics do matter to finance professors, specifically the stock’s PE ratio (#1) and market capitalization (#6). Also, book-to-market ratio was the eleventh most important factor. The ten least important factors, starting with the absolute least important, are: (1) the day of the week (1.74), (2) Carhart’s four-factor model (1.92), (3) Arbitrage Pricing Theory (2.03), (4) the constant growth dividend valuation model (2.04), (5) the stock’s correlation with Carhart’s momentum factor (UMD) (2.08), (6) Fama and French’s three- factor model (2.1), (7) the month of the year (2.21), (8) listing switches (2.29), (9) the variable growth dividend valuation model (2.31), and (10) Fama and French’s value factor (HML) (2.34).21 The results at the bottom of the pile are consistent with the overall means identified above – the traditional valuation techniques and asset-pricing models are not important to finance professors when actually buying and selling stocks. (CAPM was the 13th least important factor). The message, then, repeats: finance professors teach these fundamental valuation methods and asset-pricing models and employ them in their research, but they don’t think they are of much practical use. This begs the question: if these valuation techniques and asset-pricing models are not practically valuable, how much merit should they be given in the classroom and in research? I am also interested to know the relative importance of the fourteen groups of potentially useful information outlined earlier across the groups of active and passive
20 The top-10 most important factors in the eyes of the robust definition of active investors described in the previous footnote were the same as those presented in Table 16. The order for a few of the factors changed, but the factors comprising the top 10 remained consistent, as did the number one factor and number two factors – PE ratio and the stock’s return over the past six months, respectively. 21 Again, the bottom-10 least important factors in the eyes of the robust definition of active investors described in the previous footnote were the same as those presented in Table 16. Similarly, the order for a few of the factors changed, but the factors comprising the bottom 10 remained consistent.
64 investors. Table 17 presents the relative importance of the 14 groups of potentially useful investment information ranked based on the mean responses of the active investors. It also presents the differences between the two categories of investors along with the statistical significance of the differences. The most important grouping of information in the eyes of active investors is the collection of Jegadeesh and Titman (1993) momentum variables (3.96). The list of the top 5 most important groups is rounded out by: the stock’s correlation with the market, George and Hwang momentum variables (52-week high and low), firm characteristics, and Moskowitz and Grinblatt (1999) momentum variables. Firm characteristics and momentum seem to be relatively important to these active investors. The bottom five are: seasonality variables, asset-pricing models, dividend valuation models, the stock’s correlation with or loading on asset-pricing factors other than the market, and sentiment measures. This continues to intrigue: traditional valuation techniques and asset-pricing models are not important to these active investors.
What Matters to Active Investors Who Trade Stocks at Least Monthly? To add further clarification to the subject of what matters most to finance professors when buying and selling stocks, I zoom in on a special group of respondents. I posed the following conditioning question early in the essay: “How frequently do you buy the following investment vehicles? - Individual or Foreign Stocks?” The scale is 1 (more than once a day), 2 (daily), 3 (weekly), 4 (monthly), 5 (yearly), and 6 (less than once a year). I carve out of the active investors group those professors who responded 1 – 4 to this question. I, therefore, have a segment of finance professors who both are trying to beat the market and who trade stocks monthly or more frequently (“active traders” hereafter). The responses of this group of finance professors should be particularly meaningful. The median active trader has purchased an individual stock more than 50 times over his or her life compared with a median of only 10 – 19 times for those professors who are not active traders. I present the top 10 and bottom 10 investment factors of active traders in Table 18. I also show the percentage of respondents who
65 believed the item was highly important (responses = 6 or 7) and highly unimportant (responses = 1 or 2). The story is consistent but magnified. The least important factors to this group again relate to the traditional valuation techniques and asset-pricing models. The list at the bottom is familiar – the dividend valuation models and the traditional asset pricing models – and, again, their mean responses hover around 2. Further, the percentage of active traders that think they are highly important are almost universally below 10%, while the percent who think they are highly unimportant are almost universally above 65%. On the flip side, the importance of firm characteristics and momentum are even stronger for active traders. PE ratio (4.86) and market capitalization (4.35) are the number one and number two most important factors to this group. 39.2% and 25%, respectively, believe these firm characteristics are highly important. The number three, four, five, and six most important factors to this group all relate to momentum. Further, the mean responses of almost all of the top 10 variables are above 4 – leaning towards the “extremely important” side of the axis. The percentage of respondents who think they are highly important is almost universally above 20%. This all suggests these firm characteristics, especially PE ratio, and momentum variables are, indeed, most important, not just the least unimportant. Since this group of active traders is so informative, it is useful to compare their asset allocation to that of other respondents to the survey. In the survey I ask respondents about the allocation of their wealth. Figure II compares the holdings of the active traders to the holdings of all other respondents to the survey. For expositional purposes the wealth categories are condensed into the following: (1) fixed-income investments, (2) stocks (US and foreign), (3) mutual funds and ETFs, (4) real estate, (5) hedge funds, and (6) commodities and derivatives. The portfolios of active traders noticeably differ from the portfolios of other respondents. Active traders have roughly twice as much of their total wealth invested in stocks than non active traders (40% to 20%). Also, active traders have about a third less of their total wealth in mutual funds and ETFs (40% to 60%). Also, active traders show a
66 stronger propensity to invest in hedge funds, derivatives, and commodities, though their holdings in these assets are still relatively small. In brief summary, the factors that finance professors believe are most important when considering stock purchase or sale are somewhat surprising. The traditional dividend valuation models and asset-pricing models that have become so prevalent in text books and research are of little import in the eyes of professors. Conversely, firm characteristics, especially PE ratio and market capitalization, along with momentum variables, are the most important factors in the eyes of finance professors.
Conclusion
Employing a comprehensive survey that resulted in 642 useable responses, I offer a consensus reading of what valuation techniques, asset-pricing models, firm characteristics, corporate events, and other information are most important to finance professors when they are considering buying and selling stock. I also explore how much real-world investing experience finance professors have. At first blush, nothing seems to be particularly important to finance professors when they buy and sell stocks. Traditional asset-pricing models and valuation techniques are conspicuously unimportant from the outset. Alternatively, a stock’s correlation with the market and PE ratio show glimpses of significance, but even they are not compellingly important. However, this general lack of importance seems explained by the fact that many finance professors merely passively invest. This is supported by an investigation of finance professors’ real-world investing experience. The median professor in the survey has purchased a stock only 10 – 19 times over the course of his or her investing lifetime, and 14.5% of all respondents have never purchased a stock. This, perhaps, explains why they don’t seem to find anything too important when they buy and sell stocks: they simply don’t buy and sell stocks at all. Further, the median professor in our study has no experience with the other investments and techniques specified in the survey. After separating active from passive investors, and specifically after identifying active investors who buy or sell stocks at least monthly, the list of the 10 most and 10
67 least important factors becomes more polarized and more informative. Still, the traditional asset-pricing models and valuation techniques are of very little importance to finance professors who actively invest when they buy and sell stocks. Specifically, dividend-based valuation models, CAPM, APT, and the Fama and French (1993) and Carhart (1997) models are all glaringly unimportant in the eyes of these finance professors. Also, seasonal anomalies based on the day of the week or month of the year such as the Monday effect (Fama (1980)) and the January effect (see Keim (1983)) are equally unimportant to finance professors. The factors that do seem to matter to finance professors who actively invest and buy and sell stocks at least monthly can be broken into three general categories: (1) firm characteristics, (2) momentum data, and (3) mergers and acquisitions activity. A stock’s PE ratio seems very important to these finance professors. Market capitalization is also noticeably important. Similarly, multiple factors related to momentum – specifically, the stock’s return over the past six months and one year, along with the stock’s 52-week high and low – are consistently important to respondents. This suggests that the robustness of momentum in academic literature has made believers out of a sizeable portion of finance professors. Also, respondents indicated that mergers and acquisitions activity was an important factor in their decisions to buy and sell stocks. What really matters, then? In the final analysis, none of the traditional valuation techniques and asset-pricing models matter. Instead, what matters most are two firm characteristics – PE ratio and market capitalization – and momentum (a stock’s past returns and 52-week high and low, along with its industry’s past returns). As a final summary point for consideration, I pose a question that is yet unanswered in my own mind. If the traditional asset-pricing models (CAPM, APT, Fama French, and Carhart) are so glaringly unimportant to finance professors when they buy and sell stocks, why do we use them so ubiquitously in our research? What does it mean to say that we used the Fama-French (1993) model to adjust for risk and found that a mutual fund or a portfolio of stocks offers “no risk-adjusted” returns if we don’t believe the model itself is important at all when an individual is actually considering whether to buy or sell a stock? Perhaps, the results of this survey may need to be considered more deeply regarding how we should actually adjust for risk in our research.
68 Table 9 Summary Statistics Reproduced
This table presents summary statistics of the responses to the survey and the demographics of the respondents to the survey. The number of responses are presented in the first column. The second column contains the breakdown of useable responses by rank, gender, and marital status. The third through seventh columns presents the responses in columns one or two as a percentage of emails sent, valid emails, useable responses (columns five and six), and subgroup total, respectively.
Responses as a Percentage of Emails Valid Useable Subgroup Descriptive Stats Raw Number Sent Emails Responses Total Emails Sent 4,525 Valid Emails 4,465 98.7% Surveys Started 1,183 26.1% 26.5% Surveys Completed 870 19.2% 19.5% Total Useable Responses 642 14.2% 14.4% Rank (# of Responses) 642 100.0% Assistant 197 30.7% 30.7% Associate 197 30.7% 30.7% Full 171 26.6% 26.6% Endowed Chair 71 11.1% 11.1% Eminent Scholar 6 0.9% 0.9% Gender 631 98.3% Female 92 14.3% 14.6% Male 539 84.0% 85.4% Marital Status 632 98.4% Single - Never Married 60 9.3% 9.5% Single - Previously Married 43 6.7% 6.8% Married 529 82.4% 83.7% Median Age Range 40 - 49 $110K - Median Salary Range $120K Mean # of Total Pubs 13.37 Mean # of Top Pubs 1.84
69 Table 10 Relative Importance of 14 Groups of Variables
This table presents the relative importance of 14 categories of potentially useful information. Each of the questions on the survey related to the factors that respondents analyze when considering buying or selling stocks is categorized into one of the following 14 categories: dividend valuation models, other valuation models, asset pricing models, correlation with the market, correlation with other asset-pricing factors, firm characteristics, Jegadeesh and Titman (1993) momentum, Moskowitz and Grinblatt (1999) momentum, George and Hwang (2004) momentum, DeBondt and Thaler (1985) reversal, analyst information, corporate events, sentiment, and seasonality. The overall mean responses and mean responses across groups for these fourteen groups are presented below in descending order.
Rank Categories of Potentially Useful Information when Investing End. Ch. / # Overall Assist. Assoc. Full Em. Sch.
Scale = 1 (Not Important at All) to 7 (Extremely Important) N N Mean N Mean N Mean N Mean 0 Mean 1 The Stock's Correlation with the Market 610 3.82 184 4.06 189 3.86 162 3.60 75 3.64 2 Firm Characteristics (BM, PE, Div. Yield, Market Cap.) 591 3.33 181 3.51 184 3.25 154 3.31 72 3.13 3 Jegadeesh and Titman Momentum (stock's return over past 6 months and 1 year) 604 3.20 184 3.45 189 3.12 161 3.16 70 2.79 4 George and Hwang Momentum (stocks' 52-week high and low) 605 3.02 181 3.13 189 3.06 162 3.04 73 2.63 5 Other Valuation Models (PE, Sales, Revenue based) 610 3.02 181 3.23 191 2.95 165 3.11 73 2.45 6 Moskowitz and Grinblatt Momentum (industry's return over past 6 months and 1 year) 597 2.98 184 3.36 185 2.88 159 2.90 69 2.46 7 DeBondt and Thaler Reversal (stock and industry returns over past 3 years) 594 2.80 182 3.03 184 2.82 157 2.72 71 2.37 8 Analyst Information 573 2.66 173 2.90 181 2.56 151 2.79 68 2.05 9 Corporate Events 587 2.57 181 3.02 181 2.29 155 2.56 70 2.17 10 Sentiment Measures 604 2.42 183 2.69 186 2.42 163 2.26 72 2.10 11 Other Asset Pricing Model Loadings and Correlations (SMB, HML, UMD) 590 2.13 182 2.65 184 1.90 154 1.86 70 1.98 12 Asset Pricing Models (CAPM, APT, Fama and French, Carhart) 601 2.11 183 2.47 188 1.92 158 1.94 72 2.09 13 Dividend Valuation Models 609 1.96 182 2.09 189 1.77 163 2.01 75 2.01 14 Seasonality (day of week and month of year) 601 1.73 182 1.86 188 1.67 158 1.70 73 1.65
70 Table 11 Relative Importance of 43 Individual Variables
This table reports the overall mean responses and mean responses by rank to the questions on the survey related to the factors that respondents analyze when considering buying or selling stocks. Respondents were asked the relative importance in making their investment decision of a list of numerous possible factors one might analyze when buying and selling stocks. For instance, “When you are considering buying or selling stock, how important in making your decision are the following valuation models – The Constant Growth Dividend Valuation Model?” The response scale was: 1 = “Not Important at All” to 7 = “Extremely Important.” Since there were only 6 eminent scholars who responded to the survey, they are combined with the endowed chairs for reporting purposes.
Rank When you are considering buying or selling stock, how important in En. Ch./Em. # making your decision Overall Assist. Assoc. Full Sch. 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean N Mean N Mean N Mean 1... are the following stock valuation models?-The Constant Growth Dividend Valuation 615 1.85 184 1.98 192 1.65 164 1.91 75 1.93 Model 2... are the following stock valuation models?-The Variable Growth Dividend Valuation 613 2.09 182 2.19 190 1.92 166 2.16 75 2.08 Model 3... are the following stock valuation models?-The PE Multiple Valuation Model 617 3.06 184 3.24 193 2.94 166 3.24 74 2.47 4... are the following stock valuation models?-Other Multiples Valuation Models (such as 613 2.97 182 3.22 192 2.93 165 2.99 74 2.45 Price-to-Cash Flow and Price-to-Sales) 5... are the following asset-pricing and return-explaining models?-Capital Asset Pricing 616 2.67 185 2.76 192 2.54 166 2.62 73 2.90 Model (CAPM) 6 ... are the following asset-pricing and return-explaining models?-Arbitrage Pricing Theory 609 1.91 184 2.22 190 1.73 161 1.78 74 1.88
7... are the following asset-pricing and return-explaining models?-Fama and French's 3 - 614 2.13 184 2.53 191 1.90 166 1.99 73 2.05 Factor Model 8 ... are the following asset-pricing and return-explaining models?-Carhart's 4 - Factor Model 605 1.80 183 2.39 189 1.50 160 1.56 73 1.62
9... is the stock's correlation with or loading on the following factors?-The Market 610 3.82 184 4.06 189 3.86 162 3.60 75 3.64 10... is the stock's correlation with or loading on the following factors?-Fama and French's 599 2.32 182 2.77 186 2.09 160 2.11 71 2.27 Size Factor (SMB) 11... is the stock's correlation with or loading on the following factors?-Fama and French's 605 2.22 185 2.69 188 2.00 160 1.96 72 2.14 Value Factor (HML)
71 Table 11 Continued
Rank When you are considering buying or selling stock, how important in En. Ch./Em. # making your decision Overall Assist. Assoc. Full Sch. 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean N Mean N Mean N Mean 12... is the stock's correlation with or loading on the following factors?-Carhart's Momentum 603 1.87 185 2.48 188 1.65 156 1.49 74 1.69 Factor (UMD) 13... are the following financial ratios and firm characteristics in making your decision?-Market 607 3.38 183 3.60 189 3.32 161 3.27 74 3.24 Capitalization 14... are the following financial ratios and firm characteristics in making your decision?-Book- 602 3.17 183 3.52 188 2.99 158 3.01 73 3.05 to-Market Ratio 15... are the following financial ratios and firm characteristics in making your decision?- 605 3.06 184 3.05 189 3.05 158 3.14 74 2.95 Dividend Yield 16... are the following financial ratios and firm characteristics in making your decision?-Price- 606 3.71 184 3.88 188 3.67 161 3.77 73 3.27 to-Earnings Ratio 17... are past returns to each of the following over the period indicated?-The stock over the 608 3.20 186 3.51 190 3.07 162 3.15 70 2.84 past six months 18... are past returns to each of the following over the period indicated?-The stock's industry 603 3.03 185 3.41 187 2.91 161 2.93 70 2.60 over the past six months 19... are past returns to each of the following over the period indicated?-The stock over the 607 3.19 184 3.40 189 3.19 163 3.17 71 2.70 past year 20... are past returns to each of the following over the period indicated?-The stock's industry 603 2.95 185 3.31 186 2.85 161 2.89 71 2.42 over the past year 21... are past returns to each of the following over the period indicated?-The stock over the 600 2.89 184 3.09 188 2.90 157 2.82 71 2.46 past three to five years 22... are past returns to each of the following over the period indicated?-The stock's industry 601 2.73 184 3.00 184 2.73 161 2.61 72 2.33 over the past three to five years 23 ... are the following in making your decision?-Analyst buy and sell recommendations 609 2.51 186 2.74 189 2.39 162 2.64 72 1.92
24... are the following in making your decision?-Analyst earnings estimates 605 2.69 184 2.85 187 2.65 162 2.79 72 2.18 25... are the following in making your decision?-Analyst target prices 604 2.42 183 2.67 189 2.29 161 2.55 71 1.82 26... are the following in making your decision?-Changes to analyst buy and sell 603 2.60 184 2.90 186 2.52 161 2.67 72 1.89 recommendations
72 Table 11 Continued
Rank When you are considering buying or selling stock, how important in En. Ch./Em. # making your decision Overall Assist. Assoc. Full Sch. 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean N Mean N Mean N Mean 27 ... are the following in making your decision?-Changes to analyst earnings estimates 601 2.68 183 2.92 188 2.57 160 2.80 70 2.09
28... are the following in making your decision?-Changes to analyst target prices 604 2.45 184 2.77 187 2.28 161 2.58 72 1.76 29 ... are the following in making your decision?-The qualitative content of analyst reports 605 2.79 185 2.89 189 2.69 160 3.00 71 2.27
30... are the following in making your decision?-The dispersion in analyst earnings 601 2.50 184 2.74 187 2.50 159 2.50 71 1.87 forecasts or target prices 31... are the following corporate events in making your decision?-Stock Splits 605 2.05 185 2.44 186 1.89 162 2.01 72 1.54 32... are the following corporate events in making your decision?-Stock Repurchases 602 2.91 184 3.16 185 2.69 161 3.06 72 2.47 33 ... are the following corporate events in making your decision?-Initial Public Offerings 600 2.55 183 3.13 184 2.17 162 2.51 71 2.15 34 ... are the following corporate events in making your decision?-Seasoned Equity Offerings 600 2.46 182 2.98 184 2.13 163 2.47 71 1.97
35... are the following corporate events in making your decision?-Listing Switches 600 1.99 183 2.49 186 1.70 160 1.90 71 1.63 36... are the following corporate events in making your decision?-Dividend Increases or 604 2.89 184 3.27 185 2.57 163 2.99 72 2.50 Initiations 37 ... are the following corporate events in making your decision?-Mergers and Acquisitions 605 3.31 184 3.67 187 3.12 162 3.31 72 2.88
38... are the following in making your decision?-The stock's 52-week high 609 3.03 183 3.14 190 3.06 163 3.04 73 2.64 39... are the following in making your decision?-The stock's 52-week low 605 3.01 181 3.10 189 3.06 162 3.03 73 2.62 40... are the following in making your decision?-The stock's most recent earnings 606 3.11 181 3.29 190 3.03 162 3.18 73 2.74 announcement compared to analyst expectations 41... are the following in making your decision?-The month of the year 607 1.94 184 2.02 189 1.88 161 1.94 73 1.86 42... are the following in making your decision?-The day of the week 602 1.54 182 1.70 188 1.47 159 1.47 73 1.44 43... are the following in making your decision?-Investor sentiment measures (e.g., 604 2.42 183 2.69 186 2.42 163 2.26 72 2.10 consumer confidence)
73 Table 12 Investment Experience by Rank
This table reports the overall mean and median responses and mean and median responses by rank to the questions on the survey related to the respondents’ experience with the investment assets and techniques specified in the survey. The question was posed as follows: “Approximately how many times have you done the following over your investing lifetime?” The response scale was: 1 = Never, 2 = 1 - 9 times, 3 = 10 - 19 times, 4 = 20 - 29 times, 5 = 30 - 39 times, 6 = 40 - 49 times, 7 = more than 50 times. Since there were only 6 eminent scholars who responded to the survey, they are combined with the endowed chairs for reporting purposes.
Approximately how many times have Rank you done the following over your End. Ch. / # Overall Assist. Assoc. Full investing lifetime? Em. Sch. 1 = Never, 2 = 1 - 9 times, 3 = 10 - 19 times, 4 = 20 - 0 29 times, 5 = 30 - 39 times, 6 = 40 - 49 times, 7 = more than 50 times N Mean Med N Mean Med N Mean Med N Mean Med N Mean Med 1 Purchased an individual stock 633 3.88 3 194 3.26 2 194 4.01 4 169 4.38 4 76 4.05 4 2 Purchased an exchange traded fund (ETF) 625 1.85 1 193 1.84 1 191 1.79 1 167 2.02 1 74 1.66 1 Purchased an individual stock or ETF on the 3 628 1.79 1 194 1.63 1 192 1.72 1 166 1.92 1 76 2.05 1 margin 4 Short sold an individual stock or ETF 625 1.50 1 193 1.58 1 191 1.37 1 166 1.47 1 75 1.73 1 5 Purchased an individual call option contract 622 1.79 1 192 1.66 1 190 1.81 1 166 1.95 1 74 1.74 1 6 Purchased an individual put option contract 619 1.61 1 187 1.47 1 191 1.67 1 166 1.72 1 75 1.56 1 7 Written a call option contract 618 1.56 1 189 1.28 1 189 1.57 1 164 1.86 1 76 1.61 1 8 Written a put option contract 621 1.38 1 188 1.21 1 191 1.39 1 166 1.55 1 76 1.42 1 9 Entered into a futures contract 620 1.39 1 190 1.19 1 191 1.43 1 163 1.47 1 76 1.61 1
74 Table 13 Respondents Who Have No Experience
This table reports the number and percentage of respondents who have no experience with the investment assets and techniques specified in the model. The question was posed as follows: “Approximately how many times have you done the following over your investing lifetime?” The response scale was: 1 = Never, 2 = 1 - 9 times, 3 = 10 - 19 times, 4 = 20 - 29 times, 5 = 30 - 39 times, 6 = 40 - 49 times, 7 = more than 50 times. The first column reports the number of total responses to the question. The second column reports the number of respondents who answered “1” to the question. Column three reports the percentage of respondents who answered “1” to the question.
Approximately how many times have you done # of Answer = the following over your investing lifetime? Responses Never % Purchased an individual stock 633 92 14.5% Purchased an exchange traded fund (ETF) 625 359 57.4% Purchased an individual stock or ETF on the margin 628 456 72.6% Short sold an individual stock or ETF 625 488 78.1% Purchased an individual call option contract 622 407 65.4% Purchased an individual put option contract 619 452 73.0% Written a call option contract 618 494 79.9% Written a put option contract 621 540 87.0% Entered into a futures contract 620 524 84.5%
75 Table 14 Investment Experience by Gender
This table reports the mean and median responses by gender to the questions on the survey related to the respondents’ experience with the investment assets and techniques specified in the survey. The question was posed as follows: “Approximately how many times have you done the following over your investing lifetime?” The response scale was: 1 = Never, 2 = 1 - 9 times, 3 = 10 - 19 times, 4 = 20 - 29 times, 5 = 30 - 39 times, 6 = 40 - 49 times, 7 = more than 50 times. The table also reports the differences in mean and median responses between the genders, along with the significance levels of the differences using paired t-tests (unequal variances) for the differences in means and using the non-parametric Wilcoxon test of the differences in medians. Since there were only 6 eminent scholars who responded to the survey, they are combined with the endowed chairs for reporting purposes.
Approximately how many times have you done the # following over your investing lifetime? N Mean Median 1 = Never, 2 = 1 - 9 times, 3 = 10 - 19 times, 4 = 20 - 29 times, 5 = 30 - 0 39 times, 6 = 40 - 49 times, 7 = more than 50 times Female Male Female Male Dif p Female Male Dif p 1 Purchased an individual stock 90 534 3.10 4.02 0.92 <.0001 2 4 2 0.0005 2 Purchased an exchange traded fund (ETF) 91 525 1.43 1.91 0.49 <.0001 1 1 0 0.0028 3 Purchased an individual stock or ETF on the margin 92 527 1.28 1.87 0.59 <.0001 1 1 0 <.0001 4 Short sold an individual stock or ETF 91 525 1.13 1.57 0.43 <.0001 1 1 0 0.0004 5 Purchased an individual call option contract 91 522 1.27 1.88 0.60 <.0001 1 1 0 0.0001 6 Purchased an individual put option contract 91 519 1.19 1.68 0.50 <.0001 1 1 0 0.0005 7 Written a call option contract 91 518 1.15 1.63 0.48 <.0001 1 1 0 0.0013 8 Written a put option contract 90 522 1.12 1.43 0.30 0.0009 1 1 0 0.0493 9 Entered into a futures contract 90 521 1.19 1.41 0.23 0.0242 1 1 0 0.0651
76 Table 15 Ordered Probit Analysis of Investment Experience
This table reports the parameter estimates and significance levels of estimating nine ordered Probit models. The dependent variable in the model is a respondent’s experience with one of the investment assets or techniques in the survey (e.g., how many times a professor has purchased an individual stock). The independent variables in the model are: gender (0 = female / 1 = male); age; rank; marital status (0 = single / 1 = married); the number of total and top articles the respondent has published; salary; binary variables indicating the respondent specializes in asset pricing, derivatives, financial engineering, and market efficiency: and the respondent’s answers to the following two questions: Please indicate how strongly you agree or disagree with the following statements (1 = strongly agree to 7 = strongly disagree) – (1) given sufficient time and resources, I could implement an investing strategy that would consistently beat the market and (2) when I invest, my goal is to beat the market. The estimates are reported in the first line, while the significance levels are reported immediately below the estimate.
Approximately how many times have you ------Independent Variables (estimates with p-values directly below)------done the following over your investing # lifetime? Marital Total Top Asset Fin. Mkt. Can Beat Trying to 0 1 (Never) to 7 (More than 50 times) Gender Age Rank Status Articles Articles Salary Pricing Deriv. Eng. Eff. Market Beat Mkt. 1 Purchased an individual stock 0.33 0.12 0.03 0.07 0.01 -0.01 -0.04 -0.10 0.27 -0.11 -0.04 -0.04 -0.15 p-value 0.01 0.04 0.67 0.61 0.01 0.76 0.02 0.42 0.08 0.66 0.74 0.12 <.0001 2 Purchased an exchange traded fund (ETF) 0.43 -0.07 0.01 -0.05 0.00 -0.03 0.02 0.15 0.26 0.29 0.22 -0.09 -0.08 p-value 0.01 0.28 0.86 0.74 0.91 0.13 0.37 0.27 0.11 0.24 0.10 0.00 0.01 3 Purchased an ind. stock or ETF on the margin 0.86 0.00 0.07 -0.28 0.01 0.00 0.01 0.02 0.33 0.23 0.18 -0.09 -0.14 p-value <.0001 0.97 0.46 0.09 0.37 0.94 0.80 0.92 0.06 0.39 0.21 0.01 <.0001 4 Short sold an individual stock or ETF 0.70 0.00 -0.07 -0.05 0.01 0.01 0.02 0.21 0.15 0.15 0.29 -0.06 -0.12 p-value 0.00 0.96 0.45 0.78 0.19 0.77 0.43 0.16 0.41 0.58 0.05 0.11 0.00 5 Purchased an individual call option contract 0.63 0.11 -0.05 -0.08 0.01 -0.02 0.01 0.20 0.54 0.27 -0.01 -0.08 -0.07 p-value 0.00 0.10 0.55 0.60 0.27 0.24 0.59 0.18 0.00 0.29 0.93 0.02 0.02 6 Purchased an individual put option contract 0.54 0.15 -0.04 -0.04 0.01 -0.04 0.03 0.05 0.60 0.21 -0.07 -0.09 -0.05 p-value 0.01 0.04 0.63 0.81 0.38 0.05 0.11 0.76 0.00 0.44 0.67 0.01 0.11 7 Written a call option contract 0.66 0.18 0.06 -0.05 0.00 0.00 -0.01 0.21 0.47 0.29 0.05 -0.11 -0.04 p-value 0.01 0.02 0.54 0.81 0.55 0.96 0.67 0.19 0.01 0.31 0.76 0.00 0.27 8 Written a put option contract 0.34 0.16 -0.01 -0.01 0.01 0.00 0.02 0.09 0.60 0.12 0.01 -0.09 -0.03 p-value 0.18 0.08 0.90 0.97 0.46 0.89 0.35 0.61 0.00 0.70 0.96 0.03 0.44 9 Entered into a futures contract 0.27 0.13 0.12 -0.15 0.01 -0.02 0.01 0.11 0.70 -0.32 -0.32 -0.18 -0.04 p-value 0.25 0.13 0.26 0.47 0.31 0.51 0.60 0.57 0.00 0.34 0.11 <.0001 0.29
77 Table 16 What Matters to Active Investors
This table reports the mean responses to the questions on the survey related to the factors that respondents analyze when considering buying or selling stocks. Respondents are classified into two groups depending on their response to the following question: “Please indicate how strongly you agree or disagree with the following statement - When I invest, my goal is to beat the market.” The response scale for this question was 1 (Strongly Agree) to 7 (Strongly Disagree). Respondents who answered “1” – “3” to this question are classified as “Active Investors,” while respondents answering “5” – “7” are classified as “Passive Investors.” Respondents in both categories were asked the relative importance in making their investment decision of a list of numerous possible factors one might analyze when buying and selling stocks. For instance, “When you are considering buying or selling stock, how important in making your decision are the following valuation models – The Constant Growth Dividend Valuation Model?” The response scale was: 1 = “Not Important at All” to 7 = “Extremely Important.” Mean responses of both groups are reported along with the differences in means and the significance of the differences using paired t-tests (unequal variances). The table is ranked in descending order based on the mean responses of Active Investors.
Investor Type When you are considering buying or selling stock, how important in making your decision... # Active Passive Difference 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean Dif p 1 ... are the following financial ratios and firm characteristics in making your decision?-Price-to-Earnings Ratio 213 4.29 303 3.21 1.08 <.0001 2 ... are past returns to each of the following over the period indicated?-The stock over the past six months 210 4.04 308 2.61 1.43 <.0001 3 ... is the stock's correlation with or loading on the following factors?-The Market 213 3.95 309 3.64 0.32 0.06 4 ... are the following corporate events in making your decision?-Mergers and Acquisitions 209 3.92 306 2.87 1.05 <.0001 5 ... are past returns to each of the following over the period indicated?-The stock over the past year 209 3.87 308 2.69 1.18 <.0001 6 ... are the following financial ratios and firm characteristics in making your decision?-Market Capitalization 211 3.86 306 2.98 0.89 <.0001 7 ... are the following in making your decision?-The stock's 52-week high 211 3.81 309 2.52 1.29 <.0001
8 ... are the following in making your decision?-The stock's 52-week low 208 3.79 309 2.52 1.28 <.0001 9 ... are the following in making your decision?-The stock's most recent earnings announcement compared to analyst expectations 209 3.78 308 2.61 1.17 <.0001
10 … are past returns to each of the following over the period indicated?-The stock's industry over the past six months 208 3.75 307 2.52 1.23 0.06 11 ... are the following financial ratios and firm characteristics in making your decision?-Book-to-Market Ratio 211 3.71 304 2.68 1.03 <.0001 12 ... are the following stock valuation models?-The PE Multiple Valuation Model 215 3.64 313 2.56 1.08 <.0001 13 ... are the following stock valuation models?-Other Multiples Valuation Models (i.e., Price-to-Cash Flow or Sales) 212 3.61 313 2.41 1.21 <.0001 14 ... are the following corporate events in making your decision?-Stock Repurchases 208 3.56 305 2.51 1.05 <.0001 15 ... are past returns to each of the following over the period indicated?-The stock's industry over the past year 206 3.55 308 2.53 1.03 <.0001 16 ... are the following financial ratios and firm characteristics in making your decision?-Dividend Yield 212 3.40 305 2.71 0.68 <.0001
78
Table 16 Continued
Investor Type When you are considering buying or selling stock, how important in making your decision... # Active Passive Difference 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean Dif p 17 ... are the following corporate events in making your decision?-Dividend Increases or Initiations 209 3.38 306 2.56 0.83 <.0001 18 ... are the following in making your decision?-The qualitative content of analyst reports 209 3.37 306 2.30 1.07 <.0001 19 ... are past returns to each of the following over the period indicated?-The stock over the past three to five years 206 3.36 305 2.50 0.86 <.0001 20 ... are the following in making your decision?-Changes to analyst earnings estimates 208 3.16 303 2.27 0.89 <.0001 21 ... are the following in making your decision?-Analyst earnings estimates 209 3.16 307 2.33 0.83 <.0001 22 ... are past returns to each of the following over the period indicated?-The stock's industry over the past three to five years 207 3.13 307 2.40 0.72 <.0001 23 ... are the following in making your decision?-Changes to analyst buy and sell recommendations 210 3.06 303 2.21 0.85 <.0001 24 ... are the following in making your decision?-Investor sentiment measures (e.g., consumer confidence) 208 3.02 307 1.98 1.04 <.0001 25 ... are the following corporate events in making your decision?-Initial Public Offerings 206 3.01 304 2.27 0.74 <.0001 26 ... are the following in making your decision?-Analyst buy and sell recommendations 210 3.00 309 2.09 0.90 <.0001 27 ... are the following corporate events in making your decision?-Seasoned Equity Offerings 207 2.94 304 2.14 0.80 <.0001 28 ... are the following in making your decision?-The dispersion in analyst earnings forecasts or target prices 209 2.91 304 2.10 0.81 <.0001 29 ... are the following in making your decision?-Changes to analyst target prices 209 2.90 305 2.04 0.86 <.0001 30 ... are the following in making your decision?-Analyst target prices 208 2.89 306 2.02 0.87 <.0001 31 ... are the following asset-pricing and return-explaining models?-Capital Asset Pricing Model (CAPM) 214 2.63 312 2.70 -0.06 0.68 32 ... are the following corporate events in making your decision?-Stock Splits 208 2.42 308 1.75 0.67 <.0001 33 ... is the stock's correlation with or loading on the following factors?-Fama and French's Size Factor (SMB) 207 2.38 304 2.21 0.17 0.25 34 ... is the stock's correlation with or loading on the following factors?-Fama and French's Value Factor (HML) 211 2.34 304 2.09 0.24 0.10 35 ... are the following stock valuation models?-The Variable Growth Dividend Valuation Model 215 2.31 310 1.86 0.45 0.00 36 ... are the following corporate events in making your decision?-Listing Switches 207 2.29 304 1.76 0.53 0.00 37 ... are the following in making your decision?-The month of the year 208 2.21 310 1.70 0.51 0.00 38 ... are the following asset-pricing and return-explaining models?-Fama and French's 3 - Factor Model 212 2.10 312 2.06 0.03 0.80
79 Table 16 Continued
Investor Type When you are considering buying or selling stock, how important in making your decision... # Active Passive Difference 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) N Mean N Mean Dif p 39 ... is the stock's correlation with or loading on the following factors?-Carhart's Momentum Factor (UMD) 208 2.08 305 1.70 0.37 0.01 40 ... are the following stock valuation models?-The Constant Growth Dividend Valuation Model 215 2.04 312 1.72 0.32 0.01 41 ... are the following asset-pricing and return-explaining models?-Arbitrage Pricing Theory 211 2.03 308 1.76 0.27 0.03 42 ... are the following asset-pricing and return-explaining models?-Carhart's 4 - Factor Model 208 1.92 310 1.69 0.24 0.07 43 ... are the following in making your decision?-The day of the week 207 1.74 307 1.34 0.40 <.0001
80 Table 17 What Groups of Variables Matter to Active Investors
This table presents the relative importance of 14 categories of potentially useful information to two categories of respondents. Respondents are classified into two groups depending on their response to the following question: “Please indicate how strongly you agree or disagree with the following statement - When I invest, my goal is to beat the market.” The response scale for this question was 1 (Strongly Agree) to 7 (Strongly Disagree). Respondents who answered “1” – “3” to this question are classified as “Active Investors,” while respondents answering “5” – “7” are classified as “Passive Investors.”
Respondents in both categories were asked the relative importance in making their investment decision of a list of numerous possible factors one might analyze when buying and selling stocks. For instance, “When you are considering buying or selling stock, how important in making your decision are the following valuation models – The Constant Growth Dividend Valuation Model?” The response scale was: 1 = “Not Important at All” to 7 = “Extremely Important.” Each of the questions on the survey related to the factors that respondents analyze when considering buying or selling stocks is categorized into one of the following 14 categories: dividend valuation models, other valuation models, asset pricing models, correlation with the market, correlation with other asset-pricing factors, firm characteristics, Jegadeesh and Titman (1993) momentum, Moskowitz and Grinblatt (1999) momentum, George and Hwang (2004) momentum, DeBondt and Thaler (1985) reversal, analyst information, corporate events, sentiment, and seasonality. Mean responses of both groups are reported along with the differences in means and the significance of the differences using paired t-tests (unequal variances). The table is ranked in descending order based on the mean responses of Active Investors.
0 Active Passive # Category of Potentially Useful Information N Mean N Mean Dif t p 1 Jegadeesh and Titman Momentum (stock's return over past 6 months and 1 year) 207 3.96 307 2.65 1.31 8.52 <.0001 2 The Stock's Correlation with the Market 213 3.95 309 3.64 0.32 1.86 0.064 3 George and Hwang Momentum (stocks' 52-week high and low) 208 3.81 309 2.52 1.29 7.79 <.0001 4 Firm Characteristics (BM, PE, Div. Yield, Market Cap.) 208 3.81 297 2.91 0.89 6.4 <.0001 5 Moskowitz and Grinblatt Momentum (industry's return over past 6 months and 1 year) 204 3.64 306 2.53 1.11 7.21 <.0001 6 Other Valuation Models (PE, Sales, Revenue based) 211 3.63 312 2.48 1.15 7.22 <.0001 7 DeBondt and Thaler Reversal (stock and industry returns over past 3 years) 204 3.23 303 2.46 0.77 4.92 <.0001 8 Analyst Information 194 3.22 293 2.21 1.01 7.33 <.0001 9 Corporate Events 200 3.06 300 2.25 0.82 6.21 <.0001 10 Sentiment Measures 208 3.02 307 1.98 1.04 6.86 <.0001 11 Other Asset Pricing Model Loadings and Correlations (SMB, HML, UMD) 204 2.25 298 1.99 0.26 1.95 0.052 12 Dividend Valuation Models 214 2.17 309 1.78 0.39 3.12 0.002 13 Asset Pricing Models (CAPM, APT, Fama and French, Carhart) 208 2.16 306 2.03 0.14 1.23 0.219 14 Seasonality (day of week and month of year) 206 1.97 307 1.52 0.44 4.11 <.0001
81
Table 18 What Matters to Active Traders
This table reports the mean responses to the questions on the survey related to the factors that respondents analyze when considering buying or selling stocks for single unique group of respondents – respondents who buy stocks at least monthly and who answered “1” – “3” to the following question for which the response scale was 1 (Strongly Agree) to 7 (Strongly Disagree): “Please indicate how strongly you agree or disagree with the following statement - When I invest, my goal is to beat the market.” This group was asked the relative importance in making their investment decision of a list of numerous possible factors one might analyze when buying and selling stocks. For instance, “When you are considering buying or selling stock, how important in making your decision are the following valuation models – The Constant Growth Dividend Valuation Model?” The response scale was: 1 = “Not Important at All” to 7 = “Extremely Important.” Only the ten most important and ten least important factors are reported in the table.
Response Response When considering buying or selling stock, how important in making your decision… N Mean # = 1 or 2 = 6 or 7 0 Scale = 1 (Not Important at All) to 7 (Extremely Important) # % # % 1 ... are the following financial ratios and firm characteristics in making your decision?-Price-to-Earnings Ratio 74 4.86 9 12.2% 29 39.2% 2 ... are the following financial ratios and firm characteristics in making your decision?-Market Capitalization 72 4.35 10 13.9% 18 25.0% 3 ... are past returns to each of the following over the period indicated?-The stock over the past six months 72 4.26 15 20.8% 19 26.4% 4 ... are the following in making your decision?-The stock's 52-week high 74 4.24 14 18.9% 22 29.7% 5 ... are the following in making your decision?-The stock's 52-week low 74 4.20 14 18.9% 22 29.7% 6 ... are past returns to each of the following over the period indicated?-The stock over the past year 71 4.10 19 26.8% 16 22.5% 7 ... are the following stock valuation models?-Other Multiples Valuation Models (i.e., Price-to-Cash Flow or Sales) 73 4.01 20 27.4% 20 27.4% 8 ... are the following stock valuation models?-The PE Multiple Valuation Model 74 4.00 20 27.0% 17 23.0% 9 ... is the stock's correlation with or loading on the following factors?-The Market 74 4.00 17 23.0% 12 16.2% 10 ... are the following corporate events in making your decision?-Mergers & Acquisitions 73 3.99 17 23.3% 15 20.5% … … … … … … … … … … … … … … 34 ... is the stock's correlation with or loading on the following factors?-Fama and French's Value Factor (HML) 73 2.45 48 65.8% 4 5.5% 35 ... are the following corporate events in making your decision?-Listing Switches 73 2.44 50 68.5% 8 11.0% 36 ... are the following stock valuation models?-The Variable Growth Dividend Valuation Model 73 2.38 45 61.6% 4 5.5% 37 ... are the following in making your decision?-The month of the year 73 2.25 54 74.0% 7 9.6% 38 ... are the following asset-pricing and return-explaining models?-Fama and French's 3 - Factor Model 72 2.22 52 72.2% 3 4.2% 39 ... is the stock's correlation with or loading on the following factors?-Carhart's Momentum Factor (UMD) 72 2.19 55 76.4% 5 6.9% 40 ... are the following stock valuation models?-The Constant Growth Dividend Valuation Model 73 2.10 51 69.9% 3 4.1% 41 ... are the following asset-pricing and return-explaining models?-Arbitrage Pricing Theory 72 2.01 51 70.8% 1 1.4% 42 ... are the following asset-pricing and return-explaining models?-Carhart's 4 - Factor Model 72 2.00 56 77.8% 3 4.2% 43 ... are the following in making your decision?-The day of the week 73 1.78 60 82.2% 1 1.4%
82
This figure compares the current wealth allocation of “Active Traders” to all other respondents. In order to qualify as an Active Trader, respondents had to buy stocks at least monthly and answer “1” – “3” to the following question for which the response scale was 1 (Strongly Agree) to 7 (Strongly Disagree): “Please indicate how strongly you agree or disagree with the following statement - When I invest, my goal is to beat the market.” In order to determine wealth allocation we asked respondents what percentage of their total wealth, excluding their primary residence, was currently invested in a short list of possible investments. For instance, “Excluding your primary residence, approximately what percentage of your total personal wealth is currently invested in the following investment vehicles – Bonds or Other Fixed-Income (Interest Bearing) Investments?”
Portfolio Allocation
70.00
60.00
50.00
40.00 Active Investors Passive Investors 30.00 Percent of Portfolio of Percent
20.00
10.00
0.00 Fixed Income US and Foreign Mutual Funds and Real Estate Hedge Funds Commodities and Investments Stocks ETFs Derviatives Asset Class
Figure II Holdings of Active Investors
83 CHAPTER 4
SO YOU DISCOVERED AN ANOMALY…GONNA PUBLISH IT? AN INVESTIGATION INTO THE RATIONALITY OF PUBLISHING MARKET ANOMALIES
Introduction
If a finance professor discovers a strategy, say a profitable anomaly to market efficiency, that yields consistent abnormal returns, why would he publish it? Basic principles of economics dictate that any economic activity that produces abnormal profits attracts attention, which eventually leads to new entry into the activity. The new competition then drives profits down until the activity offers only normal profits over the long run. Ackerman, McEnally, and Ravenscraft (1999) offer support for this notion in the realm of investing. They report that hedge funds that cease voluntarily reporting performance are either abysmally low or exceptionally high performers. The cessation of reporting by exceptional performers is consistent with the idea that they don’t want to attract attention and have their strategies reverse engineered into the disappointing oblivion of merely normal profits. A new anomaly to market efficiency that yields abnormal profits could easily be classified as an “economic activity that produces abnormal profits.” Any anomaly that produces abnormal returns should attract attention and lead to an increased number of investors trading on the anomaly. As the number of investors using the anomaly increases, the profitability of the anomaly should eventually dissipate until it yields profits exactly commensurate with the risk involved. This is expected even when the purveyor of the anomaly is as secretive as possible. Publishing it in some medium that would give a large number of investors immediate access to the anomaly would presumably tremendously expedite the dissipation of its profitability. Publishing a profitable market anomaly, then, seems to be absurdly irrational behavior. Yet in spite of the apparent irrationality, numerous articles have been published introducing profitable market anomalies (see Schwert (2002) and Russell and
84 Torbey (2002) for surveys of the subject).22 Indeed, we observe many articles introducing and discussing anomalies, but we have observed no articles seeking to explain why a rational finance professor would be willing to publish an anomaly in the first place. To me, what is more perplexing than the existence of market anomalies is the fact that anyone, and especially a finance professor who presumably understands how to invest and is interested in making money, is willing to publish them once discovered. Could it be that the group of individuals who arguably subscribe most ardently to the assumption of rational economic participants are themselves exhibiting irrational behavior in the act of publishing profitable market anomalies? The primary contribution of my work is to get to the bottom of this potentially troubling question. My specific contribution is to introduce and empirically test a theory that outlines conditions under which the publishing of a market anomaly is rational behavior. The model predicts that publishing an anomaly to market efficiency is rational behavior if the professor authoring the article has relatively few publications and little reputation in the field. The theory, however, poses a very interesting implication for the academic side of finance: it suggests that the most accomplished finance professors have the highest incentive not to publish profitable anomalies to market efficiency. In answering the above research question, this paper contributes both to the literature on market efficiency and anomalies to market efficiency. It also contributes to the field of metafinance, which Cooley (1994) defines as the study of the nature, structure, and behavior of finance. I present the theory and assumptions in Section II and discuss the empirical implications in Section III. I then analyze the primary empirical implication of the model, that anomalies to market efficiency are most likely to be published by professors who are relatively young in their careers and have not yet established a strong reputation in the field, in Sections IV and V. I close with a discussion and conclusions in Section VI.
22 There have also been articles about how profitable market anomalies are, how persistent they are, and what does or does not explain them (see, for instance, George and Hwang (2004), Dimson and Marsh (1999), Jegadeesh and Titman (2001), Schwert (2002), Grundy and Martin (2001) and Cooper, Guitierrez, and Hameed (2004)). But I can find no articles that address the fundamental question of why finance professors are willing to publish them in the first place.
85 The Theory and Assumptions
Foundational Assumptions I develop and propose the following model with the intent of arriving at a set of empirical implications and predictions from the model that can, in fact, be tested. Accordingly, as I define variables in the model, I am less concerned with how easily the variables can be quantified and more concerned with how accurately the variables represent the assumptions of the model, which I hope conform to reality as far as assumptions in economic models can be expected to, while still maintaining overall model tractability. Two primary assumptions buttress the model and deserve brief discussion prior to formal development of the model: • ASSUMPTION 1: Once an anomaly to market efficiency is published, its profitability diminishes so that the only anomalies a finance professor can profitably implement are those he does not publish. • ASSUMPTION 2: A finance professor’s opportunities to accept employment as a professional money manager or to start his own hedge or mutual fund are directly correlated with his reputation. The first assumption is based on the discussion of the basic principles of competitive free markets in the introduction. Empirical work has already been done corroborating this idea. Dimson and Marsh (1999) present evidence that both in the U.S. and the U.K. the size effect conspicuously dissipates or even reverses subsequent to the publicizing of the anomaly in both countries. The authors disclaim the intent and findings of their paper by stating, “…we are not suggesting that the reversal of the small- cap premium is a consequence of its discovery and dissemination. Instead, our evidence suggests that the reversal resulted from a change in fundamentals, not just a change in sentiment.” But the timing of the dissipation and reversal of the size effect in both countries reported in their paper adds credibility to the assumption that publishing an anomaly leads to its downfall. Marquering Nisser, and Valla (2006) take a less apologetic stance. They show that since initial publication, the weekend effect, the holiday effect, the time-of-the month effect, and the January effect have all disappeared. Their results suggest that publication does, indeed, lead to an anomaly’s demise.
86 Similarly, Schwert (2002) finds that the size effect, value premium, weekend effect, and dividend-yield effect all have dissipated since initial publication. Schwert notes two possibilities explaining the dissipation of the anomalies: either they were anomalously isolated to the sample period in which they were identified or they were exploited by practitioners causing their gradual disappearance. His second explanation is consistent with the foundational assumption of my model that publishing an anomaly contributes to the dissipation of its profitability.23 Related to the second foundational assumption, anecdotal evidence suggests reputation does, in fact, play a role in the probability of landing a lucrative position as a professional money manager. Examples of finance professors who have been or currently are employed as professional money managers or compensated advisors to professional money managers include Mark Carhart, Eugene Fama, Kenneth French, Sandy Grossman, Josef Lakonishok, Robert Merton, Richard Roll, Stephen Ross, Myron Scholes, Andrei Shleifer, Robert Vishney, and others still. While this list does nothing to statistically validate the assumption, it offers enough incentive to proceed with the analysis.
The Model I begin formal development of the theory with the assumption that a finance professor’s utility (U) is a Cobb-Douglas24 function of two variables: wealth (W) and reputation (R). U(W, R) = W α R β (1)
Researchers using the Cobb-Douglas utility function often impose the constraint that 0 ≤ α ≤ 1 and that β = 1 – α (see Douglas (1976)). Since these constraints ensure
23 Schwert (2002) reports that the January effect has persisted, though to a lesser degree, since initial publication. Further, momentum seems to have actually strengthened since initial publication. However, the majority of the evidence supports the assumption that anomalies seem to dissipate and even disappear after publication. 24 Tversky and Kahnemann (1991) list Cobb-Douglas as one of the commonly used utility functions.
87 that the utility function satisfies the general axioms of consumer choice, I also adopt them in my model. 25,26 I assume that a finance professor’s reputation is a log function of his publication record (P). The variable P takes into account both the quality and quantity of his published research and is on the scale of one to infinity (one meaning he has no 27 publications). P is also raised to the power of a positive constant parameter, λR, within the natural log.
R(P) = ln(PλR ) (2)
The log relationship, along with the positive constant parameter λR and lower bound on the value of P, ensures that the first derivative of reputation (R) with respect to publication (P) is positive while the second derivative is negative – indicating that the marginal impact of publication on reputation is positive but decreasing as P increases. This captures the notion that the first few top articles do more for one’s reputation than does the 30th. For example, how much would another Journal of Finance article increase Eugene Fama’s reputation? How much would a Journal of Finance article increase my reputation (as of this writing, I have none)? I argue it would marginally improve my
reputation more than it would his. The constant parameter λR allows flexibility in the magnitude of the influence of publication on a professor’s reputation.28 Next, I assume that wealth (W) is the sum of a professor’s salary (S), which is the compensation paid to the professor from the university or college, and outside income (O), which includes income from investing and compensation from service as a professional money manager.29
25 The constraint that β = α – 1 is not necessary for the function to satisfy the axioms of consumer choice. (Nor is it necessary for the development of the model in this paper.) The only constraints necessary to ensure satisfaction of the axioms of consumer choice are that α < 1 and β ≤ 1 or α ≤ 1 and β < 1. 26 The Cobb-Douglas function was first introduced in 1928 (Cobb and Douglas (1928)) to model production functions and has since been used extensively to model utility functions. 27 P is on the scale 1 – ∞ to prevent reputation from taking a negative value. 28 Since the form of equation (2) is used throughout the model with other variables, I graphically represent the relationship in Appendix F. 29 Actually, wealth is the integral of salary and outside income over the life of the individual. However, representing wealth as the integral of these two variables complicates the math that follows without contributing new insights. Therefore, we opt to use the simpler version of wealth for the sake of intelligibility to the reader.
88 W S,( O) = S + O (3)
I also assume salary is a linear function of reputation with the positive constant,
δS, as the slope.
S = δ S R (4)
One may argue that salary should be a log function of reputation to capture a positive but decreasing marginal relationship. This is unnecessary, however, since reputation is a log function of a single variable – publishing (P). So one may more accurately think of salary as a log function of publishing by substituting equation (2) into equation (4) to obtain equation (5).
λR S = δS ln(P ) (5)
Similar to the relationship between publication and reputation, equation (5) suggests that a professor’s first few top articles contribute more marginally to his salary than does his 30th. This is supported by Swidler and Goldreyer (1998) who report that the first top publication for an assistant professor increases the present value of his lifetime earnings by $33,754 on average. His second and third top publications increase the present value of his total earnings by $32,820 and $31,886, respectively. Their results validate the assumption of a positive but diminishing marginal relationship between publication (P) and salary (S). For tractability, I present salary as a linear function of reputation, instead of as a log function of publishing. The constant positive parameter, δS allows for flexibility in the magnitude of the influence of reputation on salary.30 I assume outside income is a function of reputation (R) and the profitability (π) of the professor’s unpublished research, which is on the scale of one to infinity (one meaning the unpublished strategy offers no profits).31 Specifically, his outside income is assumed to be a log function of the profitability of his unpublished work raised to a
30 I also impose the constraint that P λR > 1, which would ensure that the term ln( P λR ) is positive, thus guaranteeing salary (S) > 0. Since P is already constrained to be equal to or greater than 1, this amounts to imposing the constraint that λR > 1. 31 π is on the scale 1 – ∞ to prevent outside income from assuming a negative value.
89 constant positive parameter, λO. This log term is then multiplied by his reputation times a
constant positive parameter, δO.
λO O π )( = δO Rln(π ) (6)
The log function, along with the positive constant parameters δO and λO and lower bound on the value of π, serves the same role as outlined above – ensuring a positive first
derivative and negative second derivative. The constant positive parameters λO allows flexibility in the magnitude of the influence of the profitability of unpublished work on a
professor’s outside income, while the constant positive parameter δO allows flexibility in the magnitude of the influence of reputation on outside income. The inclusion of reputation in the function is critical and represents the foundational assumption that the possibility of a finance professor working as professional money manager is positively related to reputation. This has anecdotal support as discussed earlier. The collective intuition behind equation (6) is that a finance professor needs reputation (R) to get the opportunity to work as a professional money manager and he needs profitable strategies to substantially gain from that employment. According to the foundational assumption outlined earlier, the only plausible profitable strategies assumed in the model are those the professor has developed but not published (π). I next perform a few simple substitutions to represent the utility function in a workable form. I first substitute equations (4) and (6) into equation (3) to obtain equation (7) and arrange terms to arrive at equation (8).
λO W S,( O) = δ S R + δ O Rln(π ) (7)
λO W S,( O) = R[δ S + δ O ln(π )] (8)
Now I substitute equation (8) into equation (1) to arrive at equation (9) and perform some simple algebraic manipulations to obtain equations (10) and (11):
λO α β U(W, R) = {R[δ S + δ O ln(π )]} R (9)
λO α +βα U(W, R) = [δ S + δ O ln(π )] R (10)
90 U(W, R) = Aα R +βα (11)
λO where A = [δ S + δ O ln(π )] (12)
Equation (11) becomes the key utility function that will serve as the focus of the analysis. The primary question of interest is at what point a finance professor is more motivated to publish than he is not to publish. If this condition exists, it would be rational to publish a market anomaly. In mathematical terms, under what conditions does the marginal utility from publishing exceed the marginal utility from not publishing? ∂U ∂U < (13) ∂π ∂P
In order to answer this question, I need the partial derivative of utility with respect to both P and π, which I calculate in detail in Appendix G. For brevity, I present the final results below. Since a professor will choose to publish only when equation (13) holds, I substitute into equation (13) the partial derivatives obtained in Appendix B to obtain equation (14): ⎛ δ λ ⎞ ⎛ λ ⎞ ()αAα −1 ⎜ O O ⎟R +βα < Aα [()α + β R βα −+ 1]⎜ R ⎟ (14) ⎝ π ⎠ ⎝ P ⎠
I now rearrange the inequality to more clearly ascertain under what conditions equation (13) holds:
⎛ R + βα ⎞ ⎛ α ⎞⎛ δ λ ⎞ ⎛ Aα ⎞ ⎜ ⎟ P ⎜ ⎟⎜ O O ⎟ < ⎜ ⎟ π ⎜ βα −+ 1 ⎟()⎜ ⎟⎜ ⎟ ⎜ α −1 ⎟() (15) ⎝ R ⎠ ⎝ α + β ⎠⎝ λR ⎠ ⎝ A ⎠
After some simple algebraic reduction, equation (15) becomes: RPk < Aπ (16)
⎛ α ⎞⎛ δOλO ⎞ where k = ⎜ ⎟⎜ ⎟ (17) ⎝α + β ⎠⎝ λR ⎠
91 Since it is rational for a researcher to publish an anomaly when equation (16) holds, equation (16) offers two clear, empirically testable implications. First, the probability of an anomaly being published is increasing as the reputation (publication record) of the researcher who discovers it decreases. The left side of equation (16)
decreases as R, P, δO, λO, and α decrease and as λR and β increase. So low values of R,
P, δO, λO, and α, along with high values of λR and β, lead to a higher incentive for a finance professor to publish. Since R is a function of P, both R and P are low when a professor is not well published in terms of both quality and quantity. α is low and β is high (since β = 1 – α) when a professor places greater proportional utility on reputation
than on wealth. δO and λO are low when the marginal contribution of not publishing (π)
to wealth is low while λR is high when the marginal contribution of publishing (P) to 32 reputation is high. However, δO and λO are also on the right side of the inequality, so interpretation of their effect on the net incentive or disincentive to publish is less clear. The second clearly testable implication is that the profitability of an anomaly should be inversely related to the reputation (publication record) of the authors who publish it. This is so because a higher profitability (a high π) combined with a low reputation (lesser publication record) lead to the inequality in equation (16) holding.
Model Implications
The results are generally intuitively acceptable. A professor is most inclined to publish when the following hold (a) he has little reputation (R is low), meaning he is not well published (P is low), (b) he derives more proportional utility from reputation than from wealth (α > β), or (c) the marginal impact of publishing on reputation is relatively
high compared to the marginal impact of not publishing on wealth (δOλO < λR).
λ 32 Since the first derivative of reputation (R) with respect to publishing (P) is R and the first derivative of P δ Rλ wealth (W) with respect to not publishing (π) is O R . π
92 Conditions (b) and (c) are fairly straightforward and involve, in my opinion, little controversy. Condition (a), however, merits further discussion. The good news from condition (a) is that the model arrives at a condition that is prevalent in the real world (a finance professor having little reputation and few publications) under which it is perfectly rational for a utility maximizing professor to publish a market anomaly. Condition (a) also suggests, however, that a professor who is widely published and has a strong reputation will gain little marginal benefit from an additional publication containing a profitable strategy he has developed. He will, instead, derive more marginal utility from using the information in unpublished form to increase his wealth through the opportunity to act as a professional money manager, which opportunity is made available to him because of his well-established reputation and publication record. While condition (a) is intuitively acceptable, it is also academically troubling. The model predicts that the best and brightest minds in academia, defined by their record of publication, have the strongest incentive not to publish their work, especially if their personal utility is disproportionately derived from wealth (α > β in equation (1)). And the result seems to have some anecdotal support (see the list presented earlier of finance professors who have been or are employed as professional money managers). This may actually be bad for market efficiency as well. When a professor publishes his profitable strategy, a large number of investors presumably become aware of and trade on the strategy, which causes a relatively rapid dissipation in the profits of the strategy. Conversely, when a professor doesn’t publish the strategy, very few investors initially trade on the strategy, so that its profitability persists. Only over time do investors infer the unpublished strategy and trade on it. So the outcome of the model may be bad both for academia and for market efficiency. Fortunately, the theoretical model outcome offers a clear, testable empirical implication. The finance professors with the greatest incentive, and for whom it is most rational, to publish profitable anomalies to market efficiency or advances to asset pricing models are those with relatively lesser reputations. Since reputation in the model is a direct result of publications, the model predicts that anomalies are most likely to be published by authors with relatively few publications. Such professors will have lesser
93 reputations and, therefore, fewer or no opportunities to work as professional money managers. Hence, they derive more reputational utility from publishing the information than they would wealth utility from not publishing the information and using it in their own personal trading. Another observable variable that likely correlates positively with reputation is the years since a professor received a Ph.D. Professors who are newer to the field will likely have lesser reputations on average. Accordingly, the model predicts that anomalies to market efficiency are most likely to be published by professors with a relatively small record of publications and who are newer to the field. I empirically analyze this implication in Sections IV and V. The theoretical model also contains at least one other empirically testable point. The model suggests that the finance professors who are most likely to start their own hedge or mutual funds or to be employed as professional money managers are those with strong publishing records, since such professors will presumably have higher reputations and, therefore, more opportunities to work as professional money managers. I leave this point, however, for future work.
Analysis of Empirical Implications – Data and Methods
In this section I analyze the primary implication of the model outlined in Section II – that anomalies to market efficiency on average are expected to be published by finance professors with lesser reputations. Determining adequate proxies for reputation are critical to the success of the analysis. Since reputation in the theoretical model is driven by publications, I assume that the publication record of an author is one viable proxy for reputation. However, it is not clear what affects reputation the most. Is it the total publications or only the top publications? I include both non-top publications and top publications (to be defined later) in the analysis. Another viable proxy is the years since an author received his Ph.D. While it is conceivable that a newcomer may quickly establish a strong reputation, it is more likely that establishing a reputation takes time. Accordingly, a second proxy for reputation is the years since an author received his Ph.D.
94 The testing, then, requires data regarding the number of publications of finance professors in the study. Heck and Cooley (2005) use similar data to identify the most prolific authors in finance. They compile an extensive database of authorship data by reviewing all articles published in 72 finance journals since 1953. Heck has established an electronic database (econlibrary.com), which gives users access to his extensive data for an annual subscription fee. I rely on his database to obtain publication data for all finance professors in my study. To test the primary empirical implication of the model I identify the first authors who published the most well-known market anomalies. The list of well-known anomalies comes from work in the second essay of this dissertation and from a survey piece on the subject by Russell and Torbey (2002). Table 1 presents the anomalies used in the analysis that follows along with the authors identified as the first to publish the anomalies. A total of 33 anomalies are used included in the sample for this study. Deciding what qualifies as an anomaly requires some subjectivity. Generally, I rely on the survey articles mentioned above to identify candidates. I augment this list of candidates with my own reading of the literature, especially in years since the surveys were published. In order to enter the sample, an anomaly also has to contain a potentially implementable trading strategy. This generally precludes any anomalies from the sample that show abnormal returns over a relatively short event window, unless investors can easily foresee the event. In a few instances I allow two articles to represent a single anomaly. There are three reasons for this. First, it was difficult to determine which authors were the first to publish the anomaly. Second, there were subsequent authors who improved or highly modified the anomaly. Third, the first article to introduce the anomaly was published in an obscure source that might not have been widely read by investors. After generating the sample of anomalies, I then determine how many top (JB, JF, JFE, JFQA, and RFS) and non-top (all other journals in the Heck’s database) publications the authors had previous to the publication of their anomaly and how many years had passed between the year they received their Ph.D. and the year they published the anomaly. For each anomaly article, I next randomly select one other article from the same issue and journal in which the anomaly was published. This article is intended to
95 serve as a benchmark.33 Therefore, the only restriction is that the benchmark article cannot introduce or advance an anomaly to market efficiency. For this benchmark article, I similarly determine how many top and non-top publications the authors had previous to the publication of the article of interest and how many years had passed between the year they received their Ph.D. and the year of they published the article in the sample. I record an observation for each article-author combination. E.g., if there are three authors for a single article, this translates into three observations. I am left with a dataset containing an observation for every author-article combination. Each observation either represents an anomaly-related publication or a benchmark publication. I then employ a simple univariate probit model to assess the relationship between the probability of an article presenting an anomaly to market efficiency and the publication record of the authors previous to the anomaly publication and the number of years between their receiving their Ph.D. and publishing the article. The theory presented earlier in the paper also suggests that professors who derive a disproportionate amount of their utility from reputation (professors who have a high β in the Cobb-Douglas utility function in equation (1)) are more likely to publish anomalies. Since it is not possible to determine every author’s α and β from equation (1), I rely on the notion that professors with extremely high βs, ex-ante, in equation (1) should publish more on an ex-post basis than those who have lower βs. Accordingly, I include a binary variable and two interaction terms in the probit analysis. The binary variable takes the value of one if the author is not on any of Cooley and Heck’s (2005) top 50 most prolific authors lists and zero if the author is on any one of those top 50 lists (henceforth the binary variable is simply NTOP50). The two interaction terms in the regression are the interaction of NTOP50 with the previous top publications variable and with the previous non-top publications variable. The unconventional defining of the binary variable makes the estimates of the two interaction terms more intelligible. Assuming that authors with high βs in equation (1) are more likely to be on the prolific authors list than those with low βs, the interaction
33 The optimal benchmark would be anomalies that were not published. However, this is infeasible to obtain, so we satisfy ourselves with a benchmark dataset consisting of non-anomaly publications.
96 terms effectively eliminate the high β authors (the most prolific authors) from the analysis by forcing their previous publications to be zero and leaving only the previous publications data for professors with low βs (authors not on the prolific authors lists). As a further proxy for a professors’ β in the Cobb-Douglas utility function presented in equation (1), I also include two other control variables. The first control variable is the total non-top publications for each author subsequent to the publication of the article representing the observation. The second control variable is the total top publications for each author subsequent to the publication of the article representing the observation. The idea behind these control variables is that authors with relatively high βs in equation (1) are expected to continue publishing even after the anomaly publication, while those with lower βs are less concerned with continuing to publish once their reputation is established sufficiently. At least two other factors may help explain why finance professors publish anomalies. First, anomalies discovered jointly by two or more researchers may have a higher probability of being published. Even though one researcher may wish not to publish the anomaly, he may be unable to dissuade his colleague from publishing it. To control for this, I add a variable to the probit model that indicates the number of authors on the paper. Second, as demonstrated in the model implications, the profitability of the anomaly may influence the probability of its being published. Specifically, highly profitable anomalies may be more likely to be published by researchers with lesser reputations. I am unable to control for this relationship in the probit model because the non-anomaly observations have no equivalent variable. I do, however, perform further testing exploring this possibility. Including the two variables of interest, one binary variable, two interaction variables, and the control variables, the probit model takes the following functional form:
Ai = α + β1PNPi + β 2 PTPi + β3YEARSi + β 4 AUTH i + β5 NTOP50i + (24)
β 6 PNPi • NTOP50i + β 7 PTPi • NTOP50i +
β8 SNPi + β9 STPi + ε i
97 Ai takes the value of one if article/author combination i introduces a market anomaly and
zero if article/author combination i represents a benchmark observation. PNPi and PTPi are the number of non-top publications and top publications, respectively, the author had
previous to the article representing the observation. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the
observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of 1 if the author is not on any of the top 50 most
prolific authors lists presented by Cooley and Heck (2005). SNPi and STPi are the non- top publications and top publications, respectively, of author i subsequent to the publication representing the observations.
PNPi and PTPi and their corresponding interaction terms are critical in the analysis. I perform all tests using a number of different definitions for each of the variables. I begin with raw data – simply the aggregate number of non-top publications
and top publications prior to the publication representing the observation (hereafter PNPi
and PTPi, respectively). I then divide the variables by the number of years between the publication of the observation article and the year the author received his Ph.D. to create
a per year representation of the two variables (hereafter PNPYi and PTPYi, respectively). Lastly, I orthogonalize the previous non-top publications per year variable to the previous top publications per year variable. This is accomplished by using the residual from a regression of the previous non-top publications per year on previous top publications per
year to create a previous non-top publications per year variable (hereafter RESIDPNPYi) that is uncorrelated with the previous top publications per year. For robustness, I also replace NTOP50 with NPAL, which takes the value of 1 if the author is not on any of the prolific authors lists in Cooley and Heck (2005) and zero otherwise. Before outlining the priors for each of the coefficient estimates, recall that the model implication derived above implies that the incentive to publish an anomaly is increasing as reputation (R) and publications (P) decrease and as proportional utility derived from reputation (β) increases. Further, R is a function of P. But P is a variable that represents the joint effects of both quantity and quality of an author’s publications. The model, however, makes no attempt to disentangle the effects of quality and quantity of one’s publications. For this reason, it is useful to consider that the true driver of the
98 disincentive to publish an anomaly is reputation. Whatever increases reputation should decrease the incentive of a researcher to publish a profitable anomaly. It seems reasonable to assume that top publications have a greater effect on reputation than do non-top publications, so one might cautiously assume, ex-ante, that the variables related to top publications will show stronger significance than variables related to non-top publications. With this in mind and recalling that the model predicts anomalies are more likely to be published by authors with lesser reputations and fewer previous publications, I
expect the estimates for β1, β2, β3, β6, and β7 to be negative in order to support the model implication. Further, in absolute terms I expect the magnitude of the coefficients related to top publications to exceed the magnitude of the coefficient related to non-top
publications (|β2| > |β1| and |β7| > |β6|), although this is not necessary to support the
primary model implication. Since β4, β5, β8, and β9 are control variables, I make no predictions regarding their coefficient estimates. Clearly, the profitability of an anomaly will have an impact on the willingness of a researcher to publish it. The model implication outlined earlier argues that the profitability of a published anomaly should be inversely related to the reputation of its authors at the time of publication. If this is true, I might expect that only those researchers with lesser reputations or high βs in equation (1) would be observed to publish highly profitable anomalies.34 To test this hypothesis, I regress the profitability of each anomaly on the exact same variables in equation (24) plus four binary variables to control for the method of risk adjustment used by the original authors of the anomaly.
π i = α + β1PNPi + β 2 PTPi + β3YEARSi + β 4 AUTH i + β5 NPALi + (25)
β 6 PNPi • NTOP50i + β 7 PTPi • NTOP50i + β8 SNPi + β9 STPi +
β10 MATCHED + β11 APM + β12CAR + β13 LS + ε i
πi takes the value of the geometrically annualized abnormal profitability of the anomaly. Most anomaly papers present a measure of the abnormal profitability to the strategy over some sample period. Some present an average monthly measure, while
34 There may be a practical relationship at work here as well. The only anomaly papers authors with little or no reputation can publish are those that prove highly profitable, while authors with well established reputations can publish anomalies that are less profitable.
99 some present an average annual measure. I simply use the average geometrically annualized abnormal return reported in the anomaly paper. I also include four binary variables to control for the method used in each paper to adjust for risk. The risk adjustment methods are grouped into five categories: (1) no risk adjustment, (2) a matched-firm approach (MATCHED), (3) risk adjustment using an asset pricing model (APM), (4) risk adjustment using an event study methodology (CAR), and (5) risk-adjustment using a long-short strategy (LS). I include binaries for MATCHED, APM, CAR, and LS. The sample used to estimate equation (25) does not include all the anomalies included in the estimation of equation (24). The primary reason for an anomaly dropping out of the sample is if that anomaly is unable to provide a clearly inferable trading strategy. For instance, Hirshleifer and Shumway (2003) show that the performance of stock markets around the world is directly related to the amount of sunshine in the city where the stock exchange is located. There is no theoretical asset pricing justification for this relationship, thus qualifying it as an “anomaly.” However, there is also no easily inferable trading strategy that an investor could feasibly implement based on this strategy. Hence, this anomaly is included in the estimation of equation (24) but not in the estimation of equation (25). Out of the 33 anomalies used to estimate equation (24), 27 are used in estimating equation (25). Assuming that the profitability of an anomaly is affected by the same factors that drive the incentive to publish an anomaly, I expect the same signs on the coefficient estimates in equation (25) that I outlined for equation (24). The basic rationale is that professors with lesser reputations and with fewer publications are more willing to publish highly profitable anomalies in an attempt to establish their reputations, while well established professors with higher reputations are only willing to publish anomalies with lower relative profits that could not be practically profitable for them personally.35
35 Some have suggested that a potentially important determinant of whether a professor would publish an anomaly is whether he has earned tenure prior to the potential publication. While acknowledging the merits of this point, I believe it is reasonable to assume that whether a finance professor has earned tenure is strongly positively correlated with the number of publications he has, which is the primary variable of interest in the model. For this reason, I forego including tenure as a control variable.
100 Analysis of Empirical Implications – Results
Table 20 reports the mean and median values of the explanatory variables in equations (24) and (25), including all robust definitions of the explanatory variables. The table reports the difference in means and medians between the anomaly and matched subsamples for each variable and compares the actual signs of the differences to the signs of the differences predicted by the theoretical model outlined earlier in the paper. Of the 35 variables for which the differences between the matched and anomaly samples are reported in Table 20, the theoretical model of the paper provides predictions for 25. The differences in the mean and especially the differences in median values reported in Table 20 consistently support the primary model implication. The signs of 24 out of the 25 differences in means are consistent with the model predictions, two of which are significant at the 10% level or better. The sign of only one difference contradicts the model, and it is not significant. 15 out of 25 differences in medians are consistent with the model predictions, while only two of the 25 differences in medians contradict the model (eight of the differences in medians are zero). Out of the 15 differences in medians that agree with the model predictions, six are significant at the 10% level or better, while neither of the two contradictory median differences is significant. The six significant median differences are: (1) anomaly authors have a median of 0.52 previous total publications per year (PPY), while matched authors have a median of 0.88 previous total publications per year; (2) anomaly authors have a median of one previous non-top publication (PNP) compared to two for matched authors; (3) anomaly authors have a median of 0.2 previous non-top publications per year (PNPY) compared to 0.35 for matched authors; (4) a median of six years has elapsed between the time an anomaly author received his Ph.D. and the year he published the article representing the observation (YEARS) compared to eight years for matched authors; (5) anomaly authors not on any of Cooley and Heck’s (2005) top 50 most prolific authors lists have a median of one previous publication (PP x NTOP50) compared to two previous publications for matched authors meeting the same criterion; (6) and anomaly authors not on any of
101 Cooley and Heck’s (2005) top 50 most prolific authors lists have median previous total publications per year (PPY x NTOP50) of 0.29 compared to 0.67 for matched authors. The summary message of Table 20, which supports the theoretical model outlined earlier, is that anomaly papers are published by authors (a) who are newer to the field and (b) who have fewer previous publications (whether it be total publications, non-top publications, or top publications and regardless of whether the variables are measured on the aggregate or per year). Before estimating equation (24), I first estimate 25 iterations of a simple single- variable probit model. The dependent variable takes the value of one if the observation relates to an anomaly and zero otherwise. In each iteration, I use one of the 25 variables in Table 20, for which the theoretical model makes a prediction. Note first, however, that Table 20 reveals noticeable differences between means and medians suggesting the mean values are being influenced by extreme observations. Since regression analysis is sensitive to this problem, I estimate all regression models on three samples. The first sample (Full Sample) is the full sample. The second sample (Lower 95%) is truncated to exclude the top 5% of the observations based on the variable serving as the independent variable in the model. For example, if the independent variable is total previous publications, the Lower 95% sample excludes those observations in the top 5% of total previous publications. The third sample is truncated to exclude the top 10% of the observations based on the variable serving as the independent variable. Obviously, the latter two samples suffer less from the discrepancy between means and medians. The reason for truncating the sample only on the high end is that the explanatory variables are bounded by zero on the low side, so the extreme values affecting the mean are all on the high end. Some of have questioned the wisdom in using truncated samples as opposed to simply taking the natural log of the affected variables. Truncating further aids in controlling for professors who have high β in equation (1). For instance truncating at the 5% level based on previous total publications removes both anomaly and matched authors with previous publications in the top 5% of the sample. Presumably, these professors represent the professors with the highest βs in the sample. Therefore,
102 truncating thusly both helps mitigate the disparity between the means and medians and helps further control for high-β professors. Results from estimating the 25 single-variable probit models are reported in Table 21. Similar to Table 2, the signs of almost all the coefficients are negative as predicted by the theoretical model. Also not surprisingly, the significance increases as the extreme observations are excluded from the analysis. Only two of the 25 variables are significant using the full sample, while five of the 25 variables are significant using the sample truncated at the 90% level. The five significant variables are: previous total publications per year, previous non-top publications per year, previous top publications per year, the interaction of the previous non-top publications and NTOP50, and the interaction of the previous non-top publications per year and NPAL. Results from estimating equation (24) on the Full, Lower 95%, and Lower 90% samples are presented in Table 22. Results from specifications using the raw, per year, and orthogonalized definitions of the explanatory variables are presented in Panels A, B, and C, respectively. Panel D presents the correlation matrix of the independent variables using the raw versions of the variables. The collective results from Table 22 provide mild support for the primary model implication. Specifically, using the sample truncated at the 10% level, the previous top publications per year variable is negative and significant using both the per-year definitions (third column of Panel B) and the orthogonalized definitions (third column of Panel C) of the independent variables. The signs of the other coefficient estimates, however, do not consistently align with model expectations as they do in previous tables. Further, and not surprisingly, the parameter estimates of the variables of interest in estimations using the full sample, which are affected by the previously highlighted discrepancy between the means and medians, show no significance. The general lack of statistical significance and the inconsistency in the signs of the parameter estimates, however, may have a simple explanation. Relatively high correlation between some of the independent variables in the model, as shown in Panel D of Table 22, may be creating instability in the parameter estimates and making identification of statistical significance difficult.
103 There is an easily anticipated, strikingly consistent pattern revealed by the correlation matrices. Regardless of how it is defined, the previous non-top publications variable is strongly correlated with the interaction of the previous non-top publications variable and NTOP50. The correlations between these two variables are 0.82 (Table 22), 0.78 (unreported), and 0.93 (unreported) for the aggregate, per year, and orthogonalized versions of the variables, respectively. Similarly, the previous top publications variable is moderately correlated with the interaction of the previous top publications variable and NTPO50. The correlations between these two variables are 0.55 (Table 22), 0.67 (unreported), and 0.67 (unreported) for the aggregate, per year, and orthogonalized versions of the variables, respectively. Also, the YEARS variable is moderately correlated with the previous non-top publications and previous top publications variables. Correlations range from 0.32 (unreported) to 0.63 (Table 22) for these bi-variate pairs. While these strong bi-variate correlations are not unexpected, they pose the threat of multicollinearity, with its attendant parameter estimate variance inflation, which causes instability in parameter estimates and masks significance. It seems prudent to re- estimate equation (24) excluding highly correlated variables. Accordingly, I re-estimate equation (24) while including only those variables that are not highly correlated. I.e., I estimate it while including previous non-top publications and previous top publications but excluding the years variable and the interaction variables. I also estimate it while including the interaction terms but excluding the years variable and previous non-top publications and previous top publications. For brevity I report only the results from these re-estimations using the per year and orthogonalized definitions of the independent variables on the Lower 95% and Lower 90% samples.36 The Lower 95% and Lower 90% samples are generated by truncating based on the explanatory variables of interest in the model. I.e., if the model includes only previous non-top publications per year and previous top publications per year, along with control variables, the samples are truncated based on the previous non-top publications per year and the previous top publications per year. The results using the per
36 Complete results are available upon request. The unreported results are qualitatively similar but with lesser statistical significance.
104 year definitions of the independent variables are presented in Panel A of Table 23. Panel B reports the results using the orthogonalized definition of the variables. The primary supporting result of Table 23 is that the coefficients of variables pertaining to prior top publications per year (whether by itself or interacted on NTOP50) are reliably negative and statistically significant. Consistent with the primary implication of the model, the analysis suggests that anomaly papers are more likely to be written by authors with relatively fewer prior top publications per year. While none of the other explanatory variables of interest significantly support the primary implication of the model, it is noteworthy that none of them significantly contradict the model. It is worth noting that two control variables are reliably significant in the analysis. NTOP50 and the subsequent top publications per year are positive and significant. These are control variables, and therefore have little bearing on the model itself. However, their magnitude and significance beg investigation. I leave this for future work but offer a few qualitative comments. These variables are intended to control for the proportional utility that authors place on reputation (β in equation (1)). Admittedly, finding an adequate proxy for β in equation (1) is a difficult task. The manner in which I defined NTOP50 suggests that if these two variables proxy for a common underlying factor, the signs should be opposite. The congruency in the signs suggests that they may not proxy for a common underlying factor. The positive and significant coefficient estimate for NTOP50 indicates that anomaly papers are more likely be written by authors who are not on any of the top 50 most prolific authors lists. This would be consistent with the notion that anomalies are written by authors with few previous publications and who don’t publish prolifically after publishing an anomaly. However, the positive and significant coefficient estimate for the subsequent top publications per year variable somewhat contradicts this notion. It suggests that authors who publish anomalies subsequently publish top journal articles at a relatively high pace. The contradiction, however, may be reconciled by the considering the possibility that once an author publishes an anomaly, his reputational capital and aspirations evolve such that he will settle for nothing less than top journal publications. I.e., he may have
105 embraced any type of journal article prior to the anomaly paper as he fought to build reputation, but subsequent to the anomaly, he only cares to publish in top journals. Additionally, publishing a recognizable anomaly may generate name recognition that makes it easier for the author to publish in top journals in the future. Again, however, I leave this as a topic for future work. Similar to my analysis of equation (24), I begin my analysis of equation (25) by estimating 25 iterations of a simple regression model. The dependent variable is the annualized profitability of the anomaly as described earlier. In each iteration, I use one of the 25 variables in Table 20, for which the theoretical model makes a prediction, and I also include the four binary variables described earlier to control for the fact that the profitability of the anomalies in the study were obtained using differing risk-adjustment procedures. Results from the 25 iterations are presented in Table 24. The results from these simple regressions also support the primary implication of the theory outlined earlier. Specifically, the signs of 24 of the 25 variables are consistent with the predictions of the model. Of the 24 consistent signs, 10 are significant at the 10% level or better. Since previous testing revealed that multicollinearity is an issue with the initial specification of equation (25), I move directly to estimations of equation (25) that avoid the inclusion of highly correlated variables. I first perform these estimations on the full sample. Since the sample size for the testing involving equation (25) is about half that of the full sample, which is not large to begin with, I are hesitant to truncate the sample. Accordingly, instead of truncating the sample to mitigate the disparity between the means and medians, I simply perform the estimations on the sample in which the independent variables of interest, except the binary variables, have been redefined as the log transformation of the original variables. Although this log-transformed sample does not augment the attempts to control for high-β professors like the truncated samples do, it at least reduces the disparity between the means and medians while retaining as many observations as possible. Results from estimating equation (25) exclusive of highly correlated independent variables are reported in Table 25. For brevity, I report only the results using the per-year (Panel A) and orthogonalized (Panel B) definitions of the independent variables of
106 interest. The results from Table 25 are consistent with the theory but statistically unconvincing. The signs of the independent variables of interest are generally of the predicted sign. Specifically, the coefficient estimates for years, previous non-top publications, previous top publications, and the interaction variables are generally negative, which supports the predictions of the model. However, none of these coefficient estimates are significant. The only significant variables are the control variables: authors, the matched-firm risk adjustment binary variable, and the long-short risk-adjustment binary variable. Given the consistency in signs, however, perhaps the lack of significance is a function of the relatively few degrees of freedom. Since I am unable to increase the degrees of freedom through an expansion of the sample, I increase the degrees of freedom by decreasing the number of explanatory variables in the model. This can be easily accomplished and with compelling justification. All previous publications variables and the YEARS variable are intended to proxy for the same variable in the theoretical model outlined earlier. Specifically, the model asserts that reputation is driven by publications. Reputation and publications in the model are inextricably connected – they represent the single driving factor behind the incentive to publish an anomaly. Accordingly, it makes sense to reduce the previous publications variables and the YEARS variable to a single common factor. Similarly, the subsequent publications variables and NTOP50 are intended to proxy for the same underlying factor in the model – the proportional utility a professor derives from reputation in the Cobb-Douglas utility function. Accordingly, it makes sense to reduce the subsequent publications variables and NTOP50 to a single common factor. This is accomplished through the use of principal components analysis. I reduce the specifications reported in Table 25 using principal components analysis. In each specification in Table 25 I replace the independent variables of interest with the first principal component of those same independent variables of interest. I also replace the subsequent publications variables and NTOP50 with the first principal component of those variables. As an example, the first specification in Panel A of Table 7 includes the following variables: years, authors, NTOP50, previous non-top publications per year, previous top
107 publications per year, subsequent non-top publications per year, subsequent top publications per year, and the four risk-adjustment dummies. After performing the principal components analysis, this specification is altered such that the YEARS, previous non-top publications per year, and previous top publications per year are replaced by the first principal component of those three variables. Additionally, the subsequent non-top publications per year, subsequent top publications per year, and NTOP50 are replaced by the first principal component of those three variables. This effectively replaces six independent variables with two principal components. It buys additional degrees of freedom in the testing and firmly aligns intuitively with the theory outlined earlier since the variables combined to form the principal components are intended to proxy for common underlying factors. I present the factor loadings and eigen values of the first principal component of the relevant combinations of variables in Table 26. I also include the eigen value of the second principal component of each combination of variables. The table manifests some desirable properties. First, the eigen values of almost all of the first principal components are above one, while the eigen values of the second principal components are universally below one. Therefore, Table 26 clearly demonstrates that use of only the first principal component is justified.37 Second, the factor loadings of the previous publications variables and the years variable are all positive. This is a indication that these variables are, indeed, proxying for a common underlying factor (presumably reputation). Third, the factor loadings on the subsequent publications variables are positive, while the factor loadings of NTOP50 are negative. This is encouraging as I anticipated a negative relationship between subsequent publications and NTOP50 if they proxy for a common underlying factor (presumably β from equation (1)). I present the results from the regression estimations using principal components in Table 27. For brevity, I report results using the original non log-transformed sample.38 Panel A reports results using the raw definitions of the variables, Panel B reports results using the per year definitions of the variables, and Panel C reports results using the
37 The Kaiser-Guttman rule suggests that only factors with eigen values greater than one are valid for extraction (See Guttman (1953) and Kaiser and Rice (1974)). 38 Results using the log-transformed variables are qualitatively similar and available upon request.
108 orthogonalized definitions of the variables. Since the number of authors variable proves strongly significant in Table 25, I estimate specifications of the regression both including and excluding the authors variable. The results presented in Table 27 strongly support of the theory outlined earlier. The principal components related to the previous publications and YEARS variables are universally negative and significant in specifications excluding the authors variable. When the authors variable is included, the magnitude of these principal components are noticeably reduced, but they universally remain negative and generally retain their significance. Specifically, the principal components based on the previous publications interacted on NTOP50 remains negative and universally significant even in the presence of the authors variable. Although applying economic interpretation to principal components is tenuous, I can conclude that the common factor among the previous publications and years variables is negatively related to the profitability of the anomaly. If the common factor may be interpreted as reputation, I may conclude that a lesser reputation is associated with a higher profitability in the published anomaly. This is consistent with the prediction of the model that authors with lesser reputations have more incentive to publish anomalies. The summary conclusions from the empirical testing of the predictions of the theoretical model of the paper are as follows. The model predicts that the incentive to publish anomalies is inversely related to reputation, which is directly driven by publications. Accordingly, anomalies should be published by authors with fewer publications and who are newer to the field. The comparison of anomaly and matched authors support the predictions of the model. The difference in means and especially medians between anomaly authors and matched authors reveals that anomaly authors have been in the field for a much shorter period of time than matched authors. Further, anomaly authors have fewer previous publications than do matched authors. The single-variable probit analyses also support the predictions of the model. Anomaly papers are more likely to be written by authors with fewer publications and who have been in the field for a shorter period of time. The probit model including multiple independent variables (except for those that are highly correlated) identify the previous top publications per year as the most significant
109 explanatory variable in distinguishing anomaly papers from matched papers. Specifically, anomaly papers are more likely to be written by authors with fewer top publications per year, which corroborates the notion that anomalies are written by authors with lesser reputations. I also hypothesize that the variables that are inversely related to the incentive to publish an anomaly should have the same relationship with the profitability of a published anomaly: more profitable anomalies should be published by authors with fewer publications and who are newer to the field. Analyses regressing the profitability of published anomalies on these variables support the theory. The regressions, including a single explanatory variable of interest, along with binary variables to control for the method of risk adjustment used to identify the anomaly, produce coefficient estimates that are consistently negative and significant for the previous publications variables and for the years since obtaining a Ph.D. The regressions including various combinations of the explanatory variables result in coefficients whose signs are consistent with the theory but whose significance is less convincing. When replacing the explanatory variables with their first shared principal component and control variables with their first shared principal component, the first principal component from the explanatory variables of interest, which presumably proxies for reputation is reliably negative and significant. This is evidence that the profitability of an anomaly is inversely related to the common underlying factor between the previous non-top publications, previous top publications, and years since obtaining a Ph.D. of the author(s) who published the anomaly. If this common underlying factor can be interpreted as reputation, these results support the theory outlined above.
Conclusion and Discussion
Why do finance professor publish anomalies? On the surface, it seems like irrational behavior. But this is too disturbing a notion – to believe that finance professors are systematically engaging in irrational behavior when they publish anomalies – for us to embrace. Employing a simple utility function driven by wealth and reputation and constrained by a few seemingly realistic assumptions I demonstrate that publishing an
110 anomaly can actually be quite rational. The theory predicts that finance professors with the greatest incentive, and for whom it is most rational, to publish an anomaly are those (a) with little reputation (b) with few publications, and (c) who derive a disproportionately high amount of utility from reputation compared to wealth. I empirically test the primary implication of the theory through the use of probit and regression analyses. The results suggest that authors who publish anomalies have fewer publications, especially fewer top publications per year, than non-anomaly authors publishing in the same journal. Additionally, authors who publish anomalies have been in the field for a shorter period of time than their non-anomaly counterparts. The profitability of an anomaly is inversely related to the number of publications that an author has at the time of the publication of the anomaly and the number of years the author has been in the field. Moreover, the profitability of an anomaly is strongly inversely related to the first principal component of the previous publications of the author and the number of years the author has been in the field. If this principal component may be interpreted as reputation, I can conclude that authors with lesser reputations are much more likely to publish highly profitable anomalies, a conclusion that is consistent with the theory outlined in the paper. Why do finance professors publish anomalies? According to the theory outlined here, which seems empirically supported, they publish anomalies because they are fighting to build reputation and by publishing the anomaly they gain more utility through the joint effect on their reputation and wealth than they would by keeping the anomaly proprietary and trading on it for their own personal gain. In other words, they publish anomalies because they are rational utility maximizers. It should be somewhat relieving to discover that the group of researchers who widely employ assumptions of rationality in their work are themselves rational on this point. Unfortunately, this implies that some of the brightest minds in our field, some of the most widely published authors in finance, have the highest incentive not to publish any market anomalies they may find. So while finance professors appear to be rational utility maximizers, I should not expect to see the kingpins of our field publishing highly profitable anomalies, unless their utility functions are highly skewed toward reputation.
111 The potential for silence from our top authors on this particular topic is discouraging but not shocking.
112 Table 19 Anomalies and Authors
This table presents the anomalies used in the empirical testing of the paper. It also presents the authors who first introduced the anomaly and the year in which the paper was published.
# Anomaly Author 1 Author 2 Author 3 Author 4 Year 1 52-week high George Hwang 2004 2 Changes in analyst recommendations Jegadeesh Kim Krische Lee 2004 3 Changes to analyst target prices Brav Lehavy 2003 4 Christmas Day Effect Lakonishok Smidt 1984 5 Dispersion in estimates Diether Malloy Scherbina 2002 6 Dividend Yield Campbell Shiller 1988 7 Dividend Yield Fama French 1988 8 Dividends Drift Healy Palepu 1988 9 Ind Momentum Moskowitz Grinblatt 1999 10 January Effect Rozeff Kinney 1976 11 January Effect w/ Small Firm Effect Keim 1983 12 January Effect w/ Small Firm Effect Reinganum 1983 13 Momentum Jegadeesh Titman 1993 14 Monday Effect Cross 1973 15 Monday Effect French 1980 16 New Year's Day Effect Roll 1983 17 Oct - March Seasonality Ogden 2003 18 PE Basu 1977 19 Post Earnings Announcement Drift Ball Brown 1968 20 Post IPO blues Loughran Ritter 1995 21 Post Listing blues Dharan Ikenberry 1995 22 Post Merger blues Asquith 1983 23 Post SEO blues Speiss Affleck-Graves 1995 24 Pre Holiday Effect Ariel 1990 25 Qualitative content Asquith Mikhail Au 2005 26 Repurchases Ikenberrry Lakonishok Vermaelen 1995 27 Reversal DeBondt Thaler 1985 28 S&P 500 Effect Harris Gurel 1986 29 S&P 500 Effect Shleifer 1986 30 Sentiment Baker Wurgler 2006 31 Size Effect Banz 1981 32 Size Effect Reinganum 1981 33 Sunshine Effect Hirshleifer Shumway 2003 34 Turn of the Month Effect (first 3 days) Lakonishok Smidt 1988 35 Turn of the Month Effect (last day) Ariel 1987 36 Value Line Effect Black 1973 37 Value Line Effect Stickel 1985 38 Value Premium Rosenberg Reid Lanstein 1985 39 Value Premium Stattman 1980 40 Stock Split Ikenberrry Ramnath 2002
113 Table 20 Differences in Means and Medians: Anomaly vs. Matched Authors
This table reports the mean and median values of all the possible explanatory variables in equations (24) and (25) for both the anomaly observations and for the matched observations. The differences in the mean values between the anomaly and matched samples are tested using a simple pair-wise t-test, while the differences in the median values are tested using a simple non-parametric median two- sample test. The variables reported are defined in the first column of the table.
N Mean Sign Median Sign Variable Anom. Matched Anom. Matched Dif Sig. Actual Expected Anom. Matched Dif. Sig. Actual Expected Previous Total Publications (PP) 68 71 7.59 7.75 -0.16 - - 2.00 4.00 -2.00 - - Previous Total Publications per Year (PPY) 60 54 0.71 0.91 -0.20 - - 0.52 0.88 -0.37 ** - - Previous Non-Top Publications (PNP) 68 71 3.84 4.44 -0.60 - - 1.00 2.00 -1.00 ** - - Previous Non-Top Publications per Year (PNPY) 60 54 0.34 0.49 -0.15 - - 0.20 0.35 -0.15 * - - Residual Previous Total Publications (RPP) 68 71 -0.45 0.43 -0.88 - - -1.91 -1.55 -0.37 - - Residual Previous Non-Top Publications per Year (RPNPY) 60 54 -0.06 0.07 -0.13 - - -0.25 -0.17 -0.09 - - Previous Top Publications (PTP) 68 71 3.75 3.31 0.44 + - 1.00 1.00 0.00 - Previous Top Publications per Year (PTPY) 60 54 0.37 0.42 -0.05 - - 0.26 0.33 -0.08 - - Subsequent Publications (SP) 68 71 13.09 12.58 0.51 + -/+ 4.00 7.00 -3.00 - -/+ Subsequent Publications per Year (SPY) 68 71 0.77 0.72 0.04 + -/+ 0.50 0.50 0.00 -/+ Subsequent Non-Top Publications (SNP) 68 71 8.04 9.45 -1.41 - -/+ 2.00 4.00 -2.00 ** - -/+ Subsequent Non-Top Publications per Year (SNPY) 68 71 0.45 0.53 -0.08 - -/+ 0.25 0.28 -0.03 - -/+ Subsequent Top Publications (STP) 68 71 5.04 3.13 1.92 * + -/+ 1.50 1.00 0.50 + -/+ Subsequent Top Publications per Year (STPY) 68 71 0.32 0.20 0.12 ** + -/+ 0.11 0.09 0.02 + -/+ Residual Subsequent Publications (RSP) 68 71 -1.51 1.45 -2.96 - -/+ -5.47 -4.47 -1.00 - -/+ Residual Subsequent Non-Top Publications per Year (RSNPY) 68 71 -0.07 0.07 -0.14 - -/+ -0.26 -0.14 -0.12 * - -/+ Years Between Obtaining PhD and Publishing Paper (Years) 60 54 8.20 8.26 -0.06 - - 6.00 8.00 -2.00 ** - - # of Authors on Paper (Auth) 68 71 2.06 2.27 -0.21 - -/+ 2.00 2.00 0.00 -/+ Not in Top 50 Authors List (NTOP50) 68 71 0.85 0.90 -0.05 - -/+ 1.00 1.00 0.00 -/+ Not in Any Top Authors List (NPAL) 68 71 0.62 0.73 -0.11 - -/+ 1.00 1.00 0.00 -/+
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
114 Table 20 Continued
This table reports the mean and median values of all the possible explanatory variables in equations (24) and (25) for both the anomaly observations and for the matched observations. The differences in the mean values between the anomaly and matched samples are tested using a simple pair-wise t-test, while the differences in the median values are tested using a simple non-parametric median two- sample test. The variables reported are defined in the first column of the table.
N Mean Sign Median Sign Variable Anom. Matched Anom. Matched Dif Sig. Actual Expected Anom. Matched Dif. Sig. Actual Expected PP x NTOP50 68 71 4.54 6.00 -1.46 - - 1.00 2.00 -1.00 * - - PP x NPAL 68 71 2.65 3.42 -0.78 - - 0.00 0.00 0.00 - PPY x NTOP50 60 54 0.47 0.70 -0.22 * - - 0.29 0.67 -0.38 ** - - PPY x NPAL 60 54 0.29 0.44 -0.15 - - 0.00 0.00 0.00 - PNP x NTOP50 68 71 2.41 3.66 -1.25 - - 0.00 1.00 -1.00 - - PNP x NPAL 68 71 1.71 2.41 -0.70 - - 0.00 0.00 0.00 - PNPY x NTOP50 60 54 0.24 0.38 -0.15 * - - 0.00 0.17 -0.17 - - PNPY x NPAL 60 54 0.17 0.28 -0.10 - - 0.00 0.00 0.00 - RPP x NTOP50 68 71 -0.57 0.46 -1.03 - - -1.13 -1.44 0.32 + - RPP x NPAL 68 71 -0.07 0.37 -0.44 - - 0.00 -0.08 0.08 + - RPNPY x NTOP50 60 54 -0.07 0.04 -0.11 - - -0.10 -0.05 -0.04 - - RPNPY x NPAL 60 54 -0.02 0.04 -0.07 - - 0.00 0.00 0.00 - PTP x NTOP50 68 71 2.13 2.34 -0.21 - - 0.00 1.00 -1.00 - - PTP x NPAL 68 71 0.94 1.01 -0.07 - - 0.00 0.00 0.00 - PTPY x NTOP50 60 54 0.23 0.31 -0.08 - - 0.02 0.10 -0.07 - - PTPY x NPAL 60 54 0.12 0.16 -0.04 - - 0.00 0.00 0.00 -
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
115 Table 21 Single-Variable Probit Analyses
This table reports the parameter estimates and significance levels from estimating 25 iterations of a single-variable probit model on three samples. The dependent variable takes the value of 1 if the observation relates to an anomaly and 0 otherwise. The probit analysis is estimated 25 times using a different explanatory variable in each iteration. The 25 explanatory variables are those in Table 2 for which the theoretical model of the paper offers predications regarding the signs. The 25 iterations are performed on three different datasets. The first sample (Full Sample) is the full sample. The second sample (Lower 95%) is truncated to exclude the top 5% of the observations based on the variable serving as the independent variable in the iteration. For example, if the independent variable is total previous publications, the Lower 95% sample excludes those observations in the top 5% of total previous publications. The third sample is truncated to exclude the top 10% of the observations based on the variable serving as the independent variable. All but two variables (NTOP50 and NPAL) are defined in the first column of the table. NTOP50 takes the value of one if the author is not on any of Cooley and Heck’s (2005) top 50 most prolific authors lists and zero otherwise. NPAL takes the value of one if the author is not on any of their prolific author lists and zero otherwise.
Expected Full Sample Lower 95% Lower 90% Variable Sign N Estimate Sig. N Estimate Sig. N Estimate Sig. Previous Total Publications (PP) - 139 -0.001 0.93 133 -0.014 0.35 126 -0.008 0.68 Previous Total Publications per Year (PPY) - 114 -0.218 0.16 109 -0.355 0.06 * 104 -0.507 0.02 ** Previous Non-Top Publications (PNP) - 139 -0.009 0.60 134 -0.001 0.98 127 -0.022 0.53 Previous Non-Top Publications per Year (PNPY) - 114 -0.372 0.12 109 -0.238 0.42 103 -0.754 0.04 ** Residual Previous Total Publications (RPP) - 139 -0.017 0.39 133 -0.008 0.80 127 0.015 0.72 Residual Previous Non-Top Publications per Year (RPNPY) - 114 -0.361 0.15 109 -0.155 0.63 103 -0.431 0.29 Previous Top Publications (PTP) - 139 0.011 0.60 133 -0.009 0.75 127 -0.019 0.59 Previous Top Publications per Year (PTPY) - 114 -0.181 0.50 109 -0.317 0.32 103 -0.676 0.08 * Years Between Obtaining PhD and Publishing Paper (Years) - 114 -0.001 0.96 109 -0.009 0.67 105 -0.022 0.35 PP x NTOP50 - 139 -0.015 0.28 134 -0.004 0.84 126 -0.025 0.35 PP x NPAL - 139 -0.013 0.47 133 -0.035 0.29 127 -0.031 0.46 PPY x NTOP50 - 114 -0.316 0.07 * 109 -0.461 0.04 ** 103 -0.331 0.20 PPY x NPAL - 114 -0.317 0.15 109 -0.373 0.18 103 -0.549 0.12 PNP x NTOP50 - 139 -0.024 0.23 133 -0.017 0.63 127 -0.031 0.53 PNP x NPAL - 139 0.017 0.44 133 -0.043 0.40 127 -0.152 0.05 ** PNPY x NTOP50 - 114 -0.435 0.09 * 109 -0.573 0.10 * 103 -0.981 0.04 ** PNPY x NPAL - 114 -0.377 0.19 109 -0.439 0.31 104 -0.560 0.33 RPP x NTOP50 - 139 -0.022 0.29 133 -0.001 0.98 126 -0.005 0.93 RPP x NPAL - 139 -0.012 0.60 133 -0.055 0.32 126 -0.005 0.95 RPNPY x NTOP50 - 114 -0.352 0.20 109 -0.331 0.37 103 -0.514 0.29 RPNPY x NPAL - 114 -0.278 0.37 109 -0.122 0.79 103 -0.127 0.83 PTP x NTOP50 - 139 -0.010 0.73 134 -0.007 0.85 131 0.031 0.48 PTP x NPAL - 193 -0.012 0.83 134 -0.006 0.92 129 -0.069 0.52 PTPY x NTOP50 - 114 -0.343 0.27 109 -0.670 0.11 103 -0.627 0.22 PTPY x NPAL - 114 -0.394 0.38 109 -1.041 0.12 103 -0.378 0.68
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
116 Table 22 Full Specification Probit Analyses
This table presents the results from estimating equation (24) using three versions of the independent variables: (1) raw versions, (2) per year versions, and (3) orthogonalized versions. The estimation is performed on three samples: (1) the full sample (Full Sample), (2) the sample truncated by eliminating the observations with previous non-top publications or previous top publications in the highest five percent of the sample (Lower 95%), and (3) the sample truncated by eliminating the observations with previous non-top publications or previous top publications in the highest ten percent of previous publications (Lower 90%). Ai takes the value of one if article/author combination i introduces a market anomaly and zero if article/author combination i represents a benchmark observation. PNPi and PTPi are the number of non-top publications and top publications, respectively, the author had previous to the article representing the observation. PNPYi and PTPYi are the per-year definitions of these two variables, respectively. RPNPYi is the residual from regressing non-top publications per year on top publications per year. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of one if the author is not on any of the top 50 most prolific authors lists presented by Cooley and Heck (2005). SNPi and STPi are the number of non-top publications and top publications, respectively, the author had subsequent to the article representing the observation. SNPYi and STPYi are the per year versions of these two variables, respectively. RSNPYi is the residual from regressing subsequent non-top publications per year on top publications per year. Panel A reports the results from estimations using the raw definitions of the variables, Panel B reports results using the per year definitions of the variables, Panel C reports results using the orthogonalized versions of the independent variables, and Panel D reports the correlation matrix of the independent variables using the raw definitions of the varibles.
A = α + β PNP + β PTP + β YEARS + β AUTH + β NTOP50 + i 1 i 2 i 3 i 4 i 5 i (24) β6 PNPi • NTOP50i + β7 PTPi • NTOP50i +
β8SNPi + β9 STP + ε ii
Panel A: Aggregate Values of Independent Variables Panel B: Per Year Values of Independent Variables Full Sample Lower 95% Lower 90% Full Sample Lower 95% Lower 90% Expected (N = 114) (N = 109) (N = 104) Expected (N = 114) (N = 109) (N = 104) Parameter Sign Estimate p Estimate p Estimate p Parameter Sign Estimate p Estimate p Estimate p Intercept -/+ -0.507 0.56 0.439 0.67 0.406 0.69 Intercept -/+ -0.010 0.99 -0.078 0.94 1.111 0.50 PNP - 0.032 0.67 -0.036 0.75 -0.037 0.66 PNPY - -0.397 0.63 0.780 0.54 2.699 0.25 PTP - -0.039 0.64 -0.090 0.53 -0.075 0.43 PTPY - -0.067 0.95 -1.516 0.30 -5.869 0.06 * Years - 0.034 0.27 0.018 0.63 0.014 0.71 Years - 0.014 0.48 0.021 0.46 0.038 0.29 Auth -/+ -0.090 0.61 -0.210 0.26 -0.173 0.34 Auth -/+ -0.119 0.47 -0.094 0.58 -0.047 0.80 NTOP50 -/+ 0.697 0.34 0.012 0.99 -0.045 0.96 NTOP50 -/+ 0.326 0.69 0.260 0.77 -1.051 0.51 PNP x NTOP50 - -0.057 0.48 0.095 0.45 0.089 0.37 PNP x NTOP50 - 0.293 0.74 -0.782 0.57 -3.264 0.18 PTP x NTOP50 - -0.003 0.97 0.003 0.98 0.015 0.87 PTP x NTOP50 - -0.387 0.71 0.688 0.64 4.786 0.13 SNP -/+ -0.013 0.24 -0.025 0.04 ** -0.022 0.06 * SNPY -/+ -0.242 0.27 -0.092 0.74 -0.010 0.97 STP -/+ 0.052 0.09 * 0.077 0.03 ** 0.070 0.04 ** STPY -/+ 0.937 0.07* 1.083 0.06* 1.086 0.09 *
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
117
Table 22 Continued
This table presents the results from estimating equation (24) using three versions of the independent variables: (1) raw versions, (2) per year versions, and (3) orthogonalized versions. The estimation is performed on three samples: (1) the full sample (Full Sample), (2) the sample truncated by eliminating the observations with previous non-top publications or previous top publications in the highest five percent of the sample (Lower 95%), and (3) the sample truncated by eliminating the observations with previous non-top publications or previous top publications in the highest ten percent of previous publications (Lower 90%). Ai takes the value of one if article/author combination i introduces a market anomaly and zero if article/author combination i represents a benchmark observation. PNPi and PTPi are the number of non-top publications and top publications, respectively, the author had previous to the article representing the observation. PNPYi and PTPYi are the per-year definitions of these two variables, respectively. RPNPYi is the residual from regressing non-top publications per year on top publications per year. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of one if the author is not on any of the top 50 most prolific authors lists presented by Cooley and Heck (2005). SNPi and STPi are the number of non-top publications and top publications, respectively, the author had subsequent to the article representing the observation. SNPYi and STPYi are the per year versions of these two variables, respectively. RSNPYi is the residual from regressing subsequent non-top publications per year on top publications per year. Panel A reports the results from estimations using the raw definitions of the variables, Panel B reports results using the per year definitions of the variables, Panel C reports results using the orthogonalized versions of the independent variables, and Panel D reports the correlation matrix of the independent variables using the raw definitions of the varibles.
A = α + β PNP + β PTP + β YEARS + β AUTH + β NTOP50 + i 1 i 2 i 3 i 4 i 5 i (24) β6 PNPi • NTOP50i + β7 PTPi • NTOP50i +
β8SNPi + β9 STP + ε ii
Panel C: Per Year Orthogonalized Values of Independent Variables Panel D: Correlation Matrix (Raw Definition of Independent Variables) Full Sample Lower 95% Lower 90% PNP x PTP x SNP STP Expected (N = 114) (N = 109) (N = 104) PNP PTP Years Auth NTOP50 NTOP50 NTOP50 SNP STP Parameter Sign Estimate p Estimate p Estimate p PNP 1.00 0.46 0.63 0.15 -0.27 0.82 0.18 0.27 0.13 Intercept -/+ -0.205 0.81 0.057 0.95 1.780 0.31 PTP 0.46 1.00 0.63 0.17 -0.53 0.09 0.55 -0.05 0.36 RPNPY - -0.397 0.63 0.836 0.51 2.773 0.24 Years 0.63 0.63 1.00 0.17 -0.18 0.50 0.46 -0.02 -0.02 PTPY - -0.218 0.80 -1.283 0.27 -4.902 0.06 * Auth 0.15 0.17 0.17 1.00 0.24 0.25 0.35 -0.25 -0.32 Years - 0.014 0.48 0.029 0.33 0.039 0.28 NTOP50 -0.27 -0.53 -0.18 0.24 1.00 0.19 0.24 -0.23 -0.76 Auth -/+ -0.119 0.47 -0.090 0.59 -0.019 0.92 PNP x NTOP50 0.82 0.09 0.50 0.25 0.19 1.00 0.32 0.23 -0.22 NTOP50 -/+ 0.403 0.63 0.007 0.99 -1.967 0.26 PTP x NTOP50 0.18 0.55 0.46 0.35 0.24 0.32 1.00 -0.20 -0.21 PNP x NTOP50 - 0.293 0.74 -1.131 0.42 -3.470 0.16 SNP 0.27 -0.05 -0.02 -0.25 -0.23 0.23 -0.20 1.00 0.34 PTP x NTOP50 - -0.276 0.76 0.401 0.73 3.558 0.16 STP 0.13 0.36 -0.02 -0.32 -0.76 -0.22 -0.21 0.34 1.00 RSNPY -/+ -0.242 0.27 0.000 1.00 -0.025 0.94 STPY -/+ 0.828 0.10 * 1.017 0.06* 1.109 0.07*
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
118 Table 23 Multicollinearity Mitigated Probit Analyses
This table presents the results from estimating equation (24), exclusive of strongly correlated variables, using the per year and orthogonalized definitions of the independent variables from two samples: (1) the sample truncated by eliminating the observations with explanatory variables of interest in the highest five percent of the sample (Lower 95%), and (3) the sample truncated by eliminating the observations with explanatory variables of interest in the highest ten percent of the sample (Lower 90%). The dependent variable takes the value of one if the article/author combination introduces a market anomaly and zero if the article/author combination represents a benchmark observation. PNPYi and PTPYi are the number of non-top publications and top publications per year, respectively, the author had previous to the article representing the observation. RPNPYi is the number of residual non-top publications per year (non-top publications per year in excess of non-top publications per year expected given the number of top publications per year) the author had previous to the article representing the observation. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of one if the author is not on any of the top 50 most prolific authors lists presented by Cooley and Heck (2005). SNPYi and STPYi are the number of non-top publications and top publications per year, respectively, the author had subsequent to the article representing the observation. RSNPYi is the number of residual non-top publications per year (non-top publications per year in excess of the non-top publications per year expected given the number of top publications per year) the author had subsequent to the article representing the observation.
Panel A: Per Year Definition of Independent Variables Lower 95% Lower 90% Parameter Expected Estimate Estimate Estimate Estimate Estimate Estimate Estimate Estimate Sign N = 105 N = 105 N = 101 N = 101 N = 94 N = 94 N = 89 N = 87 Intercept 0.052 -0.624 -0.077 -0.873 0.200 -0.787 -0.074 -1.163 Years - 0.017 0.021 0.021 0.007 Auth -/+ -0.076 -0.019 -0.111 -0.057 -0.095 -0.051 -0.084 -0.061 NTOP50 -/+ 0.209 0.793 0.311 0.994 * 0.229 1.083 0.292 1.335 ** PNPY - 0.114 0.092 -0.272 -0.353 PTPY - -0.721 -0.829 * -0.966 * -1.132 * PNPY x NTOP50 - -0.033 -0.073 -0.320 -0.453 PTPY x NTOP50 - -0.938 * -1.069 ** -1.293 * -1.372 * SNPY -/+ -0.123 -0.094 -0.115 -0.077 -0.104 -0.191 -0.029 -0.076 STPY -/+ 1.051 * 1.125 ** 1.069 * 1.154 ** 0.947 1.473 *** 1.069 * 1.579 ***
Panel B: Orthogonalized Definition of Independent Variables Lower 95% Lower 90% Parameter Expected Estimate Estimate Estimate Estimate Estimate Estimate Estimate Estimate Sign N = 104 N = 105 N = 100 N = 102 N = 93 N = 93 N = 87 N = 86 Intercept 0.052 -0.659 -0.111 -0.902 0.069 -0.879 -0.210 -1.241 Years - 0.022 0.021 0.021 0.011 Auth -/+ -0.079 -0.019 -0.114 -0.057 -0.066 -0.051 -0.058 -0.068 NTOP50 -/+ 0.171 0.785 0.275 0.975 * 0.154 1.067 * 0.253 1.317 ** RPNPY - -0.037 -0.112 -0.170 -0.424 PTPY - -0.646 -0.802 * -1.074 ** -1.295 ** PNP x NTOP50 - -0.033 -0.073 -0.147 -0.271 * Significant at 10% level TP x NTOP50 - -0.950 ** -1.097 ** -1.448 ** -1.606 ** ** Significant at 5% level RSNPY -/+ -0.067 -0.094 -0.050 -0.077 -0.158 -0.154 -0.040 -0.039 *** Significant at 1% level STPY -/+ 0.970 * 1.082 ** 0.993 * 1.119 ** 1.062 * 1.410 *** 1.072 * 1.567 ***
119
Table 24 Preliminary Regression Analyses
This table reports the parameter estimates and significance levels from estimating 25 iterations of a simple regression model on the sample including an observation for each author who published an anomaly in the sample. The dependent variable in all iterations is the annualized profitability of the anomaly as reported in the paper. The regression is estimated 25 times. Four binary variables are included in every iteration to control for the risk adjustment procedure used by the authors of the anomaly paper. Matched takes the value of one if the author(s) use a matched firm approach and zero otherwise. APM takes the value of one if the author(s) use an asset pricing model to risk adjust and zero otherwise. CAR takes the vale of one if the author(s) use an event study approach to risk adjust and zero otherwise. LS takes the value of one if he authors use a long-short strategy and zero otherwise. In addition to the four binary variables, I add one explanatory variable of interest in each iteration. The explanatory variables of interest are those in Table 2 for which the theoretical model of the paper offers predications regarding the signs.
Expected Ind. Var Risk-Adjustmet Dummies Independent Variable of Interest Sign Intercept of Interest Matched APM CAR LS N R2 Previous Total Publications (PP) - 0.086 *** -0.001 * 0.023 0.010 -0.041 0.029 52 0.17 Previous Total Publications per Year (PPY) - 0.090 *** -0.021 * 0.017 0.004 -0.036 0.034 48 0.17 Previous Non-Top Publications (PNP) - 0.086 *** -0.003 * 0.021 0.010 -0.039 0.029 52 0.17 Previous Non-Top Publications per Year (PNPY) - 0.086 *** -0.044 ** 0.019 0.009 -0.030 0.038 48 0.20 Residual Previous Total Publications (RPP) - 0.075 *** -0.003 0.015 0.010 -0.034 0.027 52 0.14 Residual Previous Non-Top Publications per Year (RPNPY) - 0.065 *** -0.055 ** 0.020 0.013 -0.024 0.040 48 0.20 Previous Top Publications (PTP) - 0.085 *** -0.002 0.023 0.009 -0.041 0.030 52 0.16 Previous Top Publications per Year (PTPY) - 0.087 *** -0.025 0.015 0.002 -0.036 0.030 48 0.13 Years Between Obtaining PhD and Publishing Paper (Years) - 0.090 *** -0.001 0.020 0.005 -0.033 0.029 48 0.14 PP x NTOP50 - 0.087 *** -0.002 * 0.015 0.003 -0.041 0.032 52 0.18 PP x NPAL - 0.078 *** 0.000 0.014 0.008 -0.035 0.027 52 0.12 PPY x NTOP50 - 0.087 *** -0.025 * 0.014 0.000 -0.032 0.037 48 0.16 PPY x NPAL - 0.080 *** -0.021 0.017 0.007 -0.026 0.033 48 0.12 PNP x NTOP50 - 0.084 *** -0.003 0.013 0.006 -0.037 0.030 52 0.15 PNP x NPAL - 0.078 *** -0.001 0.014 0.008 -0.034 0.027 52 0.12 PNPY x NTOP50 - 0.082 *** -0.046 * 0.016 0.005 -0.025 0.042 48 0.17 PNPY x NPAL - 0.080 *** -0.037 0.015 0.007 -0.025 0.034 48 0.12 RPP x NTOP50 - 0.076 *** -0.002 0.012 0.008 -0.035 0.027 52 0.13 RPP x NPAL - 0.077 *** -0.001 0.012 0.008 -0.035 0.027 52 0.12 RPNPY x NTOP50 - 0.066 *** -0.053 * 0.016 0.010 -0.022 0.040 48 0.17 RPNPY x NPAL - 0.074 *** -0.055 0.009 0.006 -0.028 0.032 48 0.14 PTP x NTOP50 - 0.087 *** -0.005 * 0.017 0.003 -0.043 0.032 52 0.18 PTP x NPAL - 0.078 *** -0.001 0.015 0.008 -0.035 0.027 52 0.12 PTPY x NTOP50 - 0.086 *** -0.030 0.012 -0.001 -0.034 0.031 48 0.13 PTPY x NPAL - 0.078 *** -0.016 0.016 0.006 -0.027 0.031 48 0.10
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
120 Table 25 Multicollinearity Mitigated Regression Analyses
This table presents the results from estimating equation (25), exclusive of highly correlated variables, using the per year and orthogonalized definitions of the independent variables from two samples: (1) the full sample and (2) the sample in which the dependent variables, with the exception of the binary variables, have been log transformed. The dependent variable is the annualized profitability of the anomaly as reported in the original journal article introducing the anomaly. PNPYi and PTPYi are the number of non-top publications and top publications per year, respectively, the author had previous to the article representing the observation. RPNPYi is the number of residual non-top publications per year (non-top publications per year in excess of non-top publications per year expected given the number of top publications per year) the author had previous to the article representing the observation. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of one if the author is not on any of the top 50 most prolific authors lists presented by Cooley and Heck (2005). SNPYi and STPYi are the number of non-top publications and top publications per year, respectively, the author had subsequent to the article representing the observation. RSNPYi is the number of residual non-top publications per year (non-top publications per year in excess of the non-top publications per year expected given the number of top publications per year) the author had subsequent to the article representing the observation. Four binary variables are included in every iteration to control for the risk adjustment procedure used by the authors of the anomaly paper. Matched takes the value of one if the author(s) use a matched firm approach and zero otherwise. APM takes the value of one if the author(s) use an asset pricing model to risk adjust and zero otherwise. CAR takes the vale of one if the author(s) use an event study approach to risk adjust and zero otherwise. LS takes the value of one if he authors use a long-short strategy and zero otherwise.
Panel A: Per Year Definition of Independent Variables Full Sample (Raw Variables) Full Sample (Log Transformed Variables) Parameter Expected Estimate Estimate Estimate Estimate Estimate Estimate Estimate Estimate Sign N = 44 N = 44 N = 48 N = 48 N = 44 N = 44 N = 48 N = 48 Intercept 0.205 *** 0.192 *** 0.199 *** 0.183 *** 0.189 *** 0.170 *** 0.193 *** 0.180 *** Years - -0.001 -0.001 -0.006 -0.003 Auth -/+ -0.051 *** -0.053 *** -0.051 *** -0.053 *** -0.050 ** -0.050 ** -0.050 *** -0.052 *** NTOP50 -/+ -0.042 -0.022 -0.043 -0.017 -0.025 -0.004 -0.036 -0.013 PNPY - -0.039 -0.043 -0.053 -0.064 PTPY - 0.014 0.011 0.028 0.019 PNPY x NTOP50 - -0.035 -0.039 -0.058 -0.061 PTPY x NTOP50 - -0.006 -0.008 -0.003 -0.003 SNPY -/+ -0.014 -0.013 -0.014 -0.012 -0.015 -0.012 -0.022 -0.020 STPY -/+ -0.052 -0.049 -0.051 -0.048 -0.063 -0.058 -0.068 -0.067 Matched -/+ 0.078 ** 0.075 ** 0.075 ** 0.072 ** 0.077 ** 0.075 ** 0.075 ** 0.073 ** APM -/+ 0.041 0.034 0.041 0.033 0.042 0.036 0.040 0.034 CAR -/+ 0.028 0.025 0.029 0.026 0.022 0.019 0.024 0.024 LS -/+ 0.124 *** 0.124 *** 0.125 *** 0.126 *** 0.111 ** 0.113 ** 0.120 *** 0.122 ***
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
121 Table 25 Continued
This table presents the results from estimating equation (25), exclusive of highly correlated variables, using the per year and orthogonalized definitions of the independent variables from two samples: (1) the full sample and (2) the sample in which the dependent variables, with the exception of the binary variables, have been log transformed. The dependent variable is the annualized profitability of the anomaly as reported in the original journal article introducing the anomaly. PNPYi and PTPYi are the number of non-top publications and top publications per year, respectively, the author had previous to the article representing the observation. RPNPYi is the number of residual non-top publications per year (non-top publications per year in excess of non-top publications per year expected given the number of top publications per year) the author had previous to the article representing the observation. YEARSi is the years between the year author i received his Ph.D. and the year he published the article associated with the observation. AUTHi is the number of authors who wrote the article representing the observation. NTOP50i takes the value of one if the author is not on any of the top 50 most prolific authors lists presented by Cooley and Heck (2005). SNPYi and STPYi are the number of non-top publications and top publications per year, respectively, the author had subsequent to the article representing the observation. RSNPYi is the number of residual non-top publications per year (non-top publications per year in excess of the non-top publications per year expected given the number of top publications per year) the author had subsequent to the article representing the observation. Four binary variables are included in every iteration to control for the risk adjustment procedure used by the authors of the anomaly paper. Matched takes the value of one if the author(s) use a matched firm approach and zero otherwise. APM takes the value of one if the author(s) use an asset pricing model to risk adjust and zero otherwise. CAR takes the vale of one if the author(s) use an event study approach to risk adjust and zero otherwise. LS takes the value of one if he authors use a long-short strategy and zero otherwise.
Panel B: Orthogonalized Variables Full Sample (Raw Variables) Full Sample (Log Transformed Variables) Parameter Expected Estimate Estimate Estimate Estimate Estimate Estimate Estimate Estimate Sign N = 44 N = 44 N = 48 N = 48 N = 44 N = 44 N = 48 N = 48 Intercept - 0.189 *** 0.187 *** 0.182 *** 0.178 *** 0.177 *** 0.167 *** 0.178 *** 0.175 *** Years -/+ -0.001 -0.001 -0.006 -0.003 Auth -/+ -0.051 *** -0.053 *** -0.051 *** -0.053 *** -0.050 ** -0.050 ** -0.050 *** -0.052 *** NTOP50 - -0.042 -0.031 -0.043 -0.028 -0.025 -0.013 -0.036 -0.023 RPNPY - -0.039 -0.043 -0.053 -0.064 PTPY - -0.001 -0.005 0.004 -0.009 PNP x NTOP50 - -0.035 -0.039 -0.058 -0.061 PTP x NTOP50 -/+ -0.020 -0.023 -0.029 -0.030 RSNPY -/+ -0.014 -0.013 -0.014 -0.012 -0.015 -0.012 -0.022 -0.020 STPY -/+ -0.059 * -0.055 * -0.057 * -0.054 * -0.070 -0.063 -0.078 -0.076 matched -/+ 0.078 ** 0.075 ** 0.075 ** 0.072 ** 0.077 ** 0.075 ** 0.075 ** 0.073 ** apm -/+ 0.041 0.034 0.041 0.033 0.042 0.036 0.040 0.034 car -/+ 0.028 0.025 0.029 0.026 0.022 0.019 0.024 0.024 ls -/+ 0.124 *** 0.124 *** 0.125 *** 0.126 *** 0.111 ** 0.113 ** 0.120 *** 0.122 ***
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
122 Table 26 Principal Components
All previous publication variables, along with the years since an author received his Ph.D., are intended to proxy for a single factor – reputation. Similarly, all subsequent publication variables, along with the NTOP50 variable are intended to proxy for a single control variable – the proportional utility an author derives from reputation. This table contains the factor loadings and eigen values of the first principal component of various combinations of the explanatory (previous publications variables and years since an author received his Ph.D.) variables and various combinations of the control variables (subsequent publications variables and NTOP50).
Per Year Prev Pubs Factor Raw Prev Pubs Factor Loadings Orth Prev Pubs Factor Loadings Loadings Interaction Control Interaction Control Interaction Control Variable Factor Factor Factor Factor Factor Factor Factor Factor Factor Previous Non-Top Publications (PNP) 0.886 Previous Non-Top Publications per Year (PNPY) 0.846 Residual Previous Non-Top Publications per Year (RPNPY) 0.553 Previous Top Publications (PTP) 0.873 Previous Top Publications per Year (PTPY) 0.793 0.636 Subsequent Non-Top Publications (SNP) 0.681 Subsequent Non-Top Publications per Year (SNPY) 0.605 Subsequent Top Publications (STP) 0.98 Subsequent Top Publications per Year (STPY) 0.938 0.864 Residual Subsequent Non-Top Publications per Year (RSNPY) 0.305 Years Between Obtaining PhD and Publishing Paper (Years) 0.68 0.669 0.442 0.307 0.547 0.576 Not in Top 50 Authors List (NTOP50) -0.82 -0.791 -0.858 PNP x NTOP50 0.791 PNPY x NTOP50 0.838 RPNPY x NTOP50 0.424 PTP x NTOP50 0.793 PTPY x NTOP50 0.671 0.35
Eigen Value of 1st Principle Component 2.009 1.702 2.099 1.541 1.247 1.872 1.01 0.634 1.577 Eigen Value of 2nd Principle Component 0.001 0.001 0.139 0.002 0.004 0.001 0.001 0.001 0.001
123 Table 27 Principal Components Regression Analyses
This table presents the results from estimating equation (25), replacing the previous publications and years variables with the first principal component obtained using those variables and replacing the subsequent publications and not in any of the top 50 authors lists variables with the first principal component obtained using those variables. The dependent variable is the annualized profitability of the anomaly as reported in the original journal article introducing the anomaly. Raw Previous Pubs Factor is the first principal component from the previous non-top publications, previous top publications, and years variables. Raw Previous Pubs Interaction Factor is the first principal component from the years variable and the previous non-top publications and previous top publications variables interacted on the not in any of the top 50 authors lists variable. Raw Control Factor is the first principal component from subsequent non-top publications, subsequent top publications, and the not in any of the top 50 authors binary variable. Per Year Previous Pubs Factor, Per Year Previous Pubs Interaction Factor, and Per Year Control Factor are similar to the Raw Previous Pubs Factor, Raw Previous Pubs Interaction Factor, and Raw Control Factor, respectively, except the publications are per year, instead of aggregate values. Orthogonalized Previous Pubs Factor, Orthogonalized Previous Pubs Interaction Factor, and Orthogonalized Control Factor are similar to the Raw Previous Pubs Factor, Raw Previous Pubs Interaction Factor, and Raw Control Factor, respectively, except the non-top publications and top publications have been orthogonalized. Matched takes the value of one if the author(s) use a matched firm approach and zero otherwise. APM takes the value of one if the author(s) use an asset pricing model to risk adjust and zero otherwise. CAR takes the vale of one if the author(s) use an event study approach to risk adjust and zero otherwise. LS takes the value of one if he authors use a long-short strategy and zero otherwise.
Panel A: Raw Definition of Variables (N = 48 for all specifications) Parameter Estimate Estimate Estimate Estimate Intercept 0.074 *** 0.080 *** 0.135 *** 0.138 *** Raw Previous Pubs Factor -0.019 * -0.009 Raw Previous Pubs Interaction Factor -0.021 ** -0.015 * Auth -0.046 *** -0.046 *** Raw Control Factor -0.005 0.009 0.004 0.012 Matched 0.027 0.014 0.067 ** 0.061 ** APM 0.008 -0.001 0.026 0.020 CAR -0.034 -0.039 0.002 -0.003 LS 0.036 0.032 0.104 *** 0.102 *** R2 = .1687 R2 = .1901 R2 = .3669 R2 = .3955
Panel B: Per Year Definition of Variables (N = 48 for all specifications) Parameter Estimate Estimate Estimate Estimate Intercept 0.077 *** 0.076 *** 0.139 *** 0.140 *** Per Year Previous Pubs Factor -0.020 ** -0.011 Per Year Previous Pubs Interaction Factor -0.024 ** -0.019 ** Auth -0.049 *** -0.050 *** Per Year Control Factor 0.000 -0.011 -0.010 -0.018 ** Matched 0.019 0.014 0.068 ** 0.067 ** APM 0.005 0.002 0.028 0.027 CAR -0.036 -0.027 0.008 0.013 LS 0.034 0.038 0.110 *** 0.116 *** R2 = .1855 R2 = .2028 R2 = .4004 R2 = .4419
Panel C: Orthogonalized Variables (N = 48 for all specfications) Parameter Estimate Estimate Estimate Estimate Intercept 0.077 *** 0.077 *** 0.137 *** 0.138 *** Orthogonalized Previous Pubs Factor -0.024 ** -0.014 Orthogonalized Previous Pubs Interaction Factor -0.027 ** -0.020 * Auth -0.047 *** -0.048 *** Orthogonalized Control Factor 0.000 -0.006 -0.008 -0.012 Matched 0.021 0.019 0.067 ** 0.068 ** APM 0.005 0.004 0.027 0.026 CAR -0.037 -0.033 0.003 0.005 LS 0.033 0.032 0.106 *** 0.107 *** R2 = .1907 R2 = .1968 R2 = .397 R2 = .4179
* Significant at 10% level ** Significant at 5% level *** Significant at 1% level
124 CHAPTER 5
CONCLUSION
The primary objective of my dissertation is to analyze market efficiency and market anomalies through an acknowledgement of the unique dual role that finance professors play as both researchers and market participants. The three questions fundamental to my analysis that formed the bases for each of the three essays are: (1) how efficient do finance professors believe US stock markets really are and does their opinion really drive their behavior, (2) which of all the valuation techniques, asset-pricing models, and market anomalies do finance professors believe are most useful when considering whether to buy or sell a stock, and (3) why would presumably rational individuals choose to publish a market anomaly they could exploit through their personal investing activities when publishing it will likely lead to its demise? The 642 useable responses to my comprehensive survey indicate that finance professors agree that US stock markets are weak form efficient but not strong form efficient. They largely disagree about semi-strong form efficiency of US stock markets, however, their behavior suggests they do, in fact, accept markets as semi-strong form efficiency; twice as many respondents passively invest than actively invest. The most surprising result related to the first question, however, is the robust discovery that a respondent’s opinion about the efficiency of US stock markets has little, if anything, to do with whether he actively or passively invests. Alternatively, the primary driver of a respondent’s decision to actively or passively invest is his confidence in his own abilities to beat the market with his investment dollars. This finding motivates the need for further work on the influence of overconfidence in investing in the spirit of Barber and Odean (2001). Regarding the second research question of the dissertation, I obtain several intriguing results. First, the traditional valuation techniques, such as the constant and variable-growth dividend valuation models, and the traditional asset-pricing models, such as CAPM and Fama and French’s (1993) three-factor model are all conspicuously unimportant to finance professors when they buy and sell stocks. This finding highlights
125 the need for us to reevaluate the techniques we use to adjust for risk in our research and the information we teach to undergraduate and graduate students. Second, what does seem to matter to finance professors are firm characteristics and momentum variables. Specifically, respondents indicate that a stock’s PE ratio, market capitalization, returns over the past six to 12 months, and 52-week high and low are the most important information in their analysis when they buy and sell stocks. This is a useful contribution to the ongoing debate about whether firm characteristics or asset- pricing model factors are best able to explain the cross section of stock returns. It also gives students and teachers insights into what the experts think are most important in the decision of what to buy and sell. Third, I discover that finance professors have less investing experience than one might think. The median respondent to my survey has purchased an individual stock only between 10 and 19 times in his life. 14.5% of respondents had never purchased an individual stock. Outside of simply buying and selling stocks, the median respondent had never bought an ETF, traded on the margin, short sold, purchased or written a call or put option contract, or entered into a futures contract. Regarding the third research question, I discover that finance professors who publish market anomalies are acting rationally when they do so. I find that anomalies are more likely to be published by professors with fewer publications and lesser reputations. Also, the profitability of an anomaly is inversely related to the author’s previous publications and reputation. Accordingly, finance professors with fewer publications and lesser reputations have strong rational incentive to publish market anomalies. Unfortunately, this implies that professors with many publications and greater reputations have little incentive to publish market anomalies, which intimates that there likely are anomalies that have been discovered but are not published since their discovers are exploiting them for personal gain. Viewed from this angle, the third essay provides indirect evidence of market inefficiency.
126 APPENDIX A The Survey
Below is the dedicated survey, which is sent to one half of the finance professors randomly selected.
This survey is being conducted as a part of a dissertation of a doctoral candidate in the Department of Finance at Florida State University. No identifying information of respondents to the survey will be used in the dissertation. Florida State University requires that all participants involved in studies affiliated with the university consent to participation in the study. Please indicate your consent by responding to the following statement:
By answering one or more of the following questions, I consent to participate in this study: Yes No
Which of the following best describes your current position as a faculty member? Assistant Associate Full Endowed Eminent Adjunct Lecturer Other Professor Professor Professor Chair Scholar
Which of the following best describes your general area of specialty? Economics Finance Law Real Estate Other
What graduate and/or professional degrees do you hold? Please select all that apply DBA JD MBA MA MS PhD Other - Business Related
What are your specific areas of specialty? Please select all that apply. (To select more than one hold the control button while clicking on all that apply)
Approximately how many articles have you authored or co-authored that have been published or accepted for publication at a peer-reviewed journal (scale is 0 – more than 50)?
Approximately how many of your articles published in peer-reviewed journals are consistent with the efficient market hypothesis (scale is 0 – more than 50)?
Approximately how many of your articles published in peer-reviewed journals refute the efficient market hypothesis (scale is 0 – more than 50)?
Approximately how many articles have you authored or co-authored that have been published or accepted for publication at any one of the following journals (scale is 0 – more than 25):
Journal of Financial Economics (JFE) Journal of Finance (JF) Review of Financial Studies (RFS) Journal of Business (JB) Journal of Financial and Quantitative Analysis (JFQA)
Approximately how many of your articles published in JFE, JF, RFS, JB, or JFQA are consistent with the efficient market hypothesis (scale is 0 – more than 25)?
Approximately how many of your articles published in JFE, JF, RFS, JB, or JFQA refute the efficient market hypothesis (scale is 0 – more than 25)?
Please indicate the degree to which your findings and conclusions support or refute the efficient markets hypothesis (7 – point scale from strongly supports market efficiency to strongly refutes market efficiency) - The collective body of all my published research generally
127 - My most recently published journal article - My most recent unpublished working paper(s)
Please indicate how strongly you agree or disagree with the following statements (7 – point scale from strongly agree to strongly disagree) - It is possible to predict future returns to US stocks using only past returns - It is possible to predict future returns to US stocks using only past returns and publicly available information - It is possible to predict future returns to US stocks using only past returns, publicly available information, and private information - Investment returns are solely a compensation for risk - Investment strategies exist that consistently beat average market returns without taking above-average risk - Given sufficient time and resources, I could implement an investing strategy that would consistently beat the market - When I invest, my goal is to beat the market
Excluding your primary residence, approximately what percentage of your total personal wealth is currently invested in the following types of investments (11 – point ordinal scale from 0 – 100%)? - Bonds or Other Fixed-Income (Interest Bearing) Investments - Individual US Stocks - Individual Foreign Stocks or ADRs - Actively-Managed (Load) Mutual Funds - Passively-Managed (No-Load) Mutual Funds (Including Index Funds) - Exchange Traded Funds - Real Estate (Including Investment Properties and REITs) - Hedge Funds - Derivatives (Futures and Options) - Commodities (Gold, Wheat, Energy, Etc.)
How frequently do you buy the following investment vehicles (scale = more than once a day, daily, weekly, monthly, yearly, less than once a year)? - Bonds or Other Fixed-Income (Interest Bearing) Investments - Individual US Stocks - Individual Foreign Stocks or ADRs - Actively-Managed (Load) Mutual Funds - Passively-Managed (No-Load) Mutual Funds (Including Index Funds) - Exchange Traded Funds - Real Estate (Including Investment Properties and REITs) - Hedge Funds - Derivatives (Futures and Options) - Commodities (Gold, Wheat, Energy, Etc.)
How frequently do you sell the following investment vehicles (scale = more than once a day, daily, weekly, monthly, yearly, less than once a year)? - Bonds or Other Fixed-Income (Interest Bearing) Investments - Individual US Stocks - Individual Foreign Stocks or ADRs - Actively-Managed (Load) Mutual Funds - Passively-Managed (No-Load) Mutual Funds (Including Index Funds) - Exchange Traded Funds - Real Estate (Including Investment Properties and REITs) - Hedge Funds - Derivatives (Futures and Options) - Commodities (Gold, Wheat, Energy, Etc.)
128 When you are considering buying or selling stock, how important in making your decision are the following stock valuation models? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - The Constant Growth Dividend Valuation Model - The Variable Growth Dividend Valuation Model - The PE Multiple Valuation Model - Other Multiples Valuation Models (such as Price-to-Cash Flow and Price-to-Sales)
When you are considering buying or selling stock, how important in making your decision are the following asset-pricing and return-explaining models? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - Capital Asset Pricing Model (CAPM) - Arbitrage Pricing Theory - Fama and French's 3 - Factor Model - Carhart's 4 - Factor Model
When you are considering buying or selling a stock, how important in making your decision is the stock's correlation with or loading on the following factors? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - The Market - Fama and French's Size Factor (SMB) - Fama and French's Value Factor (HML) - Carhart's Momentum Factor (UMD)
When you are considering buying or selling a stock, how important are the following financial ratios and firm characteristics in making your decision? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - Market Capitalization - Book-to-Market Ratio - Dividend Yield - Price-to-Earnings Ratio
When you are considering buying or selling a stock, how important in making your decision are past returns to each of the following over the period indicated? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - The stock over the past six months - The stock's industry over the past six months - The stock over the past year - The stock's industry over the past year - The stock over the past three to five years - The stock's industry over the past three to five years
When you are considering buying or selling a stock, how important are the following in making your decision? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - Analyst buy and sell recommendations - Analyst earnings estimates - Analyst target prices - Changes to analyst buy and sell recommendations - Changes to analyst earnings estimates - Changes to analyst target prices - The qualitative content of analyst reports - The dispersion in analyst earnings forecasts or target prices
When you are considering buying or selling a stock, how important are the following corporate events in making your decision? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - Stock Splits - Stock Repurchases
129 - Initial Public Offerings - Seasoned Equity Offerings - Listing Switches - Dividend Increases or Initiations - Mergers and Acquisitions
When you are considering buying or selling a stock, how important are the following in making your decision? [Scale is from 1 (Not Important at All) to 7 (Extremely Important)] - The stock's 52-week high - The stock's 52-week low - The stock's most recent earnings announcement compared to analyst expectations - The month of the year - The day of the week - Investors sentiment measures (e.g., consumer confidence)
Approximately how many times have you done the following over your investing lifetime (scale = never, 1 – 9, 10 – 19, 20 – 29, 30 – 39, 40 – 49, 50 or more times)? - purchased an individual stock - purchased an exchange traded fund (ETF) - purchased an individual stock or ETF on the margin - short sold an individual stock or ETF - purchased an individual call option contract - purchased an individual put option contract - written a call option contract - written a put option contract - entered into a futures contract
Which of the following age ranges describes you? Under 30 30 - 39 40 - 49 50 - 59 60 – 69 Over 69
Please indicate your gender. Female Male
Please indicate your marital status. Single - Never Married Single - Previously Married Married
Into which of the following salary ranges does your total college or university paid nine-month compensation fall? Please exclude contributions to retirement accounts, health care benefits, and other similar benefits and perquisites (scale is in $10,000 increments from $50,000 - $200,000).
Please indicate your total 9-month salary relative to your peers (colleagues of the same rank, in the same department, and with the same number of years at your or a similar university or college) (7 – point scale from far below average to far above average)
130 APPENDIX B The Premail
Distributed February 19, 2007
Professor [Last Name],
Thank you for taking the time to read this email.
My name is Colby Wright. I am a Ph.D. candidate in the Department of Finance at Florida State University. As a critical component of my dissertation, I am conducting a survey to assess the real-world investing experience and strategies of finance professors to get a general opinion of how you invest and your perception of the market.
In two days you will receive an email containing a link to the survey and inviting you to participate. Each respondent who completes the survey will be entered into a random drawing for $500, which will be awarded 16 days from the day you receive this email.
Beta testing suggests it takes between 10 and 18 minutes to complete the survey. Your responses will be STRICTLY CONFIDENTIAL. Absolutely no identifying information will be used in the study. Please take the time to complete the survey when you receive the invitation email in two days.
If you prefer, you may take the survey now by following this link: link to survey
Thank you.
Colby Wright Ph.D. Candidate Department of Finance Florida State University
131 APPENDIX C The Invitation Email
Distributed February 21, 2007
Professor [Last Name],
This email is a follow-up to an email I sent you two days ago. If you have already taken the survey, thank you and please disregard this message. If you have not yet completed the survey, please follow the link at the bottom of this email and complete the survey that will appear in your browser.
As indicated in my previous message, as a part of my dissertation (follow this link to see my profile: http://cob.fsu.edu/fin/display_phd_profiles.cfm?pID=29) I am conducting a survey to assess the real-world investing experience and strategies of finance professors to get a general opinion of how you invest and your perception of the market.
Each respondent who completes the survey will be entered into a random drawing for $500, which will be awarded two weeks from today. Your responses will be STRICTLY CONFIDENTIAL. Excluding outliers, the median and mean response times of the participants who completed the survey to this point are 10 and 11 minutes, respectively (the range is from 4 minutes to 34 minutes). I greatly appreciate your taking the time to complete the survey.
Follow this link to the Survey: link to survey
Thank you.
Colby Wright Ph.D. Candidate Department of Finance Florida State University [email protected]
132 APPENDIX D The Postmail
Distributed February 23, 2007
Professor [Last Name],
This is the third and final email (I promise not to clog your inbox after today) inviting you to participate in a survey being conducted as a part of my dissertation. If you have already completed the survey thank you and please disregard this message. If you have not completed the survey, will you please follow the link at the bottom of this email and the complete the survey that will appear in your browser. The survey will become inactive on Monday morning at 8:00 am EST preventing any further responses.
The response rate to the survey to this point has been very high. I've also received a number of emails offering comments, suggestions, and insights related to the survey. I sincerely thank each of you for your participation and for your emails.
Follow this link to the Survey: link to survey
Thank you.
Colby Wright Ph.D. Candidate Department of Finance Florida State University [email protected] My Profile: http://cob.fsu.edu/fin/display_phd_profiles.cfm?pID=29
133 APPENDIX E Investment Strategies Suggested by Academia
In Order to Maximize Returns, Strategy Suggests Investors…
STRATEGY # GO LONG IN GO SHORT IN AVOID DESCRIPTION Stocks with high historical Classic portfolio returns, low standard deviations, 1 and whose covariance with theory current portfolio is low or negative Dividend Valuation Stocks whose prices are below 2 the intrinsic value based on the Models dividend valuation model used Multiples Models Stocks whose prices are below 3 (P/E, P/CF, P/S, the intrinsic value based on the P/BV) multiple model used 4 CAPM Stocks with high betas Stocks that load heavily on APT relevant macro economic factors 5 such as credit and term spreads, (Macro-Economic) inflation, and industrial production, etc. Stocks that load heavily on the three Fama and French factors (market – risk-free rate, small
6 Fama and French stocks – big stocks, high book- to-market stocks – low book-to- market stocks) Stocks that load heavily on the 7 Carhart Fama and French factors and the Carhart factor (winners – losers) 8 Size Strategy Small stocks Large stocks Stocks with highest book-to- Stocks with lowest book-
9 Value Strategy market ratios to-market ratios Stocks with highest dividend Stocks with lowest
10 Dividend-Yield yields dividend yields Stocks with highest PE Stocks with lowest PE ratios 11 Price-Earnings ratios Post-Earnings Stocks that have recently Stocks that have recently 12 announced higher than expected announced lower than Announcement Drift earnings expected earnings Worst performing stocks Jegadeesh and Top performing stocks over past over past six to twelve 13 six to twelve months Titman Momentum months Top performing stocks Debont and Thaler Worst performing stocks over over past three to five 14 past three to five years Reversal years Moskowitz and Stocks in top performing Stocks in worst performing 15 industries over past six to twelve industries over past six to Grinblatt Momentum months twelve months George and Hwang Stocks that are closest to their Stocks that are farthest 16 52-Week High 52-week high from their 52-week high Dividend initiations Stocks that initiate or increase 17 or increases dividends In late December in small stocks 18 January Effect that have had poor performance during the year 19 Weekend Effect Selling Stocks on Monday October – March Hold long positions during Long positions during April 20 Seasonality October – March – September
134 Analyst Stocks with the most positive 21 recommendation change in analyst changes recommendations Analyst target price Stocks with the most positive 22 revisions target price revisions Qualitative content of Stocks receiving most positive 23 analyst reports comments in analyst reports Dispersion in analyst Stocks with high dispersion 24 opinion in analyst earnings forecasts Stocks that have recently split
25 Stock splits their stock Stocks that have recently
26 Stock repurchases announced repurchase programs Investing in IPO stocks if 27 IPOs you are not involved in initial allotment Investing in stocks carrying
28 SEOs out SEOs Investing in stocks that have recently switched their
29 Listing switches listing to the NYSE or AMEX Stocks of firms that have 30 Mergers recently made an acquisition that was funded by stock Stocks that are small, young, highly volatile, unprofitable, Baker and Wurgler non-dividend paying, 31 Sentiment extremely growth oriented, and distressed when sentiment measures are high
135 APPENDIX F Graphical Representation of Reputation as a Function of Publishing
Appendix A contains a graphical representation of the assumed relationship between reputation (R) and publishing (P) with three different values of the constant parameter λR (0.75, 1, 1.25). The graph depicts a positive but decreasing marginal relationship where the lower values of P add much more to R than do the th higher values of P. I.e., the first few high quality publications increase reputation much more than the 30 high quality publication does. λR influences the magnitude of the effect of a given level of P on R.
R(P) = ln(PλR ) (2) from Chapter 4
Reputation as a Function of Publishing
10 9 8 7
6 Lambda = 0.75 5 Lambda = 1 4 Lambda = 1.25 Reputation 3 2 1 0 1 101 201 301 401 501 601 701 801 901 Publishing
136 APPENDIX G Partial Derivatives of Utility Function – Chapter 4
Begin with the utility function: U(W, R) = Aα R +βα (11)
λO where A = [δ S + δ O ln(π )] (12) and R = ln(PλR ) (2)
I eventually need the first derivative of A with respect to π and the first derivative of R with respect to P. dA δ λ = O O (12.1) dπ π dR λ = R (2.1) dP P
I next calculate the partial derivative of utility with respect to P. Note R is a function of P, but A is not. Using the chain rule and calling upon equation (2.1) above the partial derivative is calculated thusly:
+βα ∂U α ⎡d(R )⎤⎡d(R)⎤ = A ⎢ ⎥⎢ ⎥ (11.1A) ∂P ⎣ dR ⎦⎣ dP ⎦
∂U α βα −+ 1 ⎡λR ⎤ = A []()α + β R ⎢ ⎥ (11.1B) ∂P ⎣ P ⎦
I next calculate the partial derivative of utility with respect to p. Note that A is f function of π, but R is not. Using the chain rule and calling upon equation (12.1) above the partial derivative is calculated thusly:
α ∂U ⎛ dA ⎞⎛ dA ⎞ +βα = ⎜ ⎟⎜ ⎟R (11.2A) ∂π ⎝ dA ⎠⎝ dπ ⎠ ∂U ⎛ δ λ ⎞ = ()αAα −1 ⎜ O O ⎟R +βα (11.2B) ∂π ⎝ π ⎠
137 REFERENCES
1. Ackerman, Carl, McEnally, Richard, and David Ravenscraft, 1999, “The Performance of Hedge Funds: Risk, Return, and Incentives,” Journal of Finance, 59, p. 833 – 873. 2. Adams, John C. and Ken B. Cyree, “Market Efficiency and Diversification, An Experiential Approach Using the Wall Street Journal’s Dartboard Portfolio,” Journal of Applied Finance, 14, p. 40 – 51. 3. Ariel, Robert, 1990, “High stock returns before holidays: Existence and evidence on possible causes,” Journal of Finance 45, p. 1611-1626. 4. Ariel, Robert, 1987, “A Monthly Effect in Stock Returns,” Journal of Financial Economics 18, p. 161- 174. 5. Asquith, Paul, 1983, “Merger Bids, Uncertainty, and Stockholder Returns,” Journal of Financial Economics, 11, 51 – 83. 6. Asquith, Paul; Mikhail, Michael B.; and Andrea S. Au, 2005, “Information Content of Equity Analyst Reports,” Journal of Financial Economics, 75, p. 245 – 282. 7. Baker, Malcolm and Jeffrey Wurgler, 2004, “Investor Sentiment and the Cross-Section of Stock Returns,” NBER Working Paper No. 10449. 8. Ball, R. and P. Brown, 1968, “An Empirical Evaluation of Accounting Income Numbers,” Journal of Accounting Research, 6, p. 159 – 178. 9. Banz, Rolf, 1981, “The Relationship Between Return and Market Value of Common Stocks,” Journal of Financial Economics, 9, p. 3 – 18. 10. Baumgartner, H. and Homberg, C. (1996), "Applications of structural equation modeling in marketing and consumer research: a review", International Journal of Research in Marketing, Vol. 13, pp. 139- 161. 11. Barber, Brad and Terrance Odean, 2001, “Boys Will Be Boys: Gender, Overconfidence, and Common Stock Investment,” Quarterly Journal of Economics, 116, p. 261 – 292. 12. Basu, S., 1977, “Investment Performance of common Stocks in Relationto Their Price-Earning Ratios: A Test of the Efficient Markets Hypothesis,” Journal of Finance, 32, p. 663 – 682. 13. Basu, S., 1983, “The Relationship Between Earnings’ Yield, Market Value, and Return for NYSE Common Stocks: Further Evidence,” Journal of Financial Economics, 12, p. 129 – 156. 14. Benartzi, Shlomo; Michaely, Roni; and Richard Thaler, 1997, “Do Changes in Dividends Signal the Future or the Past?,” Journal of Finance 52, p. 1007 – 1034. 15. Bernard, V. and J. Thomas, 1990, “Evidence that Stock Prices do not Fully Reflect the Implications of Current Earnings for Future Earnings,” Journal of Accounting and Economics, 13, p. 305 – 340. 16. Bhattacharya , Utpal and Galpin, Neal E., "Is Stock Picking Declining Around the World?" (November 2005), AFA 2007 Chicago Meetings Paper Available at SSRN: http://ssrn.com/abstract=849627 17. Black, Fischer, 1973, “Yes, Virginia, there is hope,” Financial Analysts Journal, 29, p. 10 – 14. 18. Boehme, Rodney and Sorin M Sorescu, 2002, “The Long-run Performance Following Dividend Initiations and Resumptions: Underreaction or Product of Chance?,” The Journal of Finance, 57, p. 871–900. 19. Bollen, K. A., 1986, “Sample Size And Bentler And Benett's Nonormed Fit Index,” Psychometrika, 51, p. 375-377. 20. Bosnjak, Michael and Tracy L. Tuten, 2003, “Prepaid and Promised Incentives in Web Surveys: An Experiment,” Social Science Computer Review, 21, p. 208 – 217. 21. Brau, James C. and Stanley E. Fawcett, 2006, “Initial Public Offerings: An Analysis of Theory and Practice,” Journal of Finance, 61, p. 399 – 436.
138 22. Brav, Alon, Geczy, Christopher, and Paul A. Gompers, 2000, “Is the abnormal return following equity issuances anomalous?,” Journal of Financial Economics, 56, p. 209 – 249. 23. Brav, Alon, Graham, John R., Harvey, Campbell R., and Roni Michaely, 2005, “Payout Policy in the 21st Century,” Journal of Financial Economics, 77, p. 483 – 527. 24. Brav, Alon and Reuven Lehavy, 2003, “An Empirical Analysis of Analysts’ Target Prices: Short-Term Infomativeness and Long-Term Dynamics,” Journal of Finance, 58, p. 1933 – 1967. 25. Campbell, John Y., 1987, “Stock Returns and the Term Structure,” Journal of Financial Economics, 18, p. 373 – 400. 26. Campbell, John Y. and Robert Shiller, 1988, “The Dividend Price Ratio and Expectations of Future Dividends and Discount Factors,” Review of Financial Studies, 1, 195 – 228. 27. Carhart, M. M., 1997, “On Persistence in Mutual Fund Performance,” Journal of Finance, 52, p. 57 – 82. 28. Chan, Wesley S., 2003, “Stock Price Reaction to News and No-News: drifts and reversal after headlines,” Journal of Financial Economics, 70, p. 223 – 260. 29. Chen, Nai-Fu; Roll, Richard, and Stephen Ross, 1986, “Economic Forces and the Stock Market,” Journal of Business, 59, p. 383 – 403. 30. Cobb, Charles W. and Paul H. Douglas, 1928, “A Theory of Production,” American Economic Review, 8, p. 139 – 165. 31. Cooley, Philip, 1994, “Survival Strategies for the Fledgling Finance Professor,” Financial Practice and Education, 4, p. 8 – 17. 32. Cooper, Michael J., Gutierrez, Roberto C. Jr., and Allaudeen Hameed, 2004, “Market States and Momentum,” Journal of Finance, 59, p. 1345 – 1365. 33. Crawford, Scott D., Couper, Mick P., and Mark J. Lamias, 2001, “Web Surveys: Perceptions of Burden,” Social Science Computer Review 19, p. 146 – 162. 34. Cross, Frank, 1973, “The Behavior of Stock Prices on Friday and Monday,” Financial Analysts Journal, 29, p. 67 – 69. 35. De Bondt, W. F. M., and R. H. Thaler, 1985, “Financial Decision-Making in Markets and Firms: A Behavioral Perspective,” in Finance, ed. by R. Jarrow, V. Maksimovic, and W. Ziemba. Elsevier/Horth Holland, Amsterdam, vol. 9 of Handbook in Operations Research and Management Science, chap. 13, pp. 385–410. 36. Dharan, B. G. and D. L. Ikenberry, 1995, “The Long-Run Negative Drift of Post-Listing Stock Returns,” Journal f Finance, 50, 1547 – 1574. 37. Diether, Karl B., Malloy, Christopher J., and Anna Scherbina, 2002, “Differences of Opinion and the Cross Section of Stock Returns,” Journal of Finance, 57, p. 2113 – 2141. 38. Dillman, D.A., 1978, Mail and Telephone Surveys: The Total Design Method, New York: Wiley. 39. Dimson, Ellen and Paul Marsh, 1987, “The Hoare Govett Smaller Companies Index for the U.K.,” Hoare Govett Limited. 40. Dimson, Ellen and Paul Marsh, 1999, “Murphy’s Law and Market Anomalies,” Journal of Portfolio Management, 25, p. 53 – 69. 41. Eckbo, B. Espen, Masulis, Ronald W., and Oyvind Norli, 2000, “Seasoned Public Offerings: Resolution of the New Issues Puzzle’,” Journal of Financial Economics, p. 251 – 291. 42. Fama, Eugene, 1998, “Market Efficiency, Long-Term Returns, and Behavioral Finance,” Journal of Financial Economics, 49, p. 283 – 306. 43. Fama, Eugene and Kenneth French, 1988, “Dividend Yields and Expected Stock Returns,” Journal of Financial Economics, 22, p. 3 – 25.
139 44. Fama, Eugene and Kenneth French, 1992, “The Cross-Section of Expected Stock Returns,” Journal of Finance, 47, p. 427 – 465. 45. Fama, Eugene and Kenneth French, 1993, “Common Risk Factors in the Returns on Stocks and Bonds,” Journal of Financial Economics, 33. p. 3 – 56. 46. Fama, Eugene and G. William Schwert, 1977, “Asset Returns and Inflation,” Journal of Financial Economics, 5, p. 115 – 146. 47. Fornell, Claes and David F. Larcker, 1981, “Evaluating Structural Equation Models with Unobservable Variables and Measurement Error,” Journal of Marketing Research, 63, p. 39 – 50. 48. French, Kenneth, 1980, “Stock Returns and the Weekend Effect,” Journal of Financial Economics, 8, p. 55 – 70. 49. Friedman, Milton, 1953, The Methodology of Positive Economics in: Essays in Positive Economics. University of Chicago Press. 50. George, Thomas J. and Chuan-Yang Hwang, 2004, “The 52-Week High and Momentum Investing,” Journal of Finance, 59, p. 2145 – 2176. 51. Gerbing, David W. and James C. Anderson, 1992, “Monte Carlo Evaluations of Goodness of Fit Indices for Structural Equation Models,” Sociological Methods and Research, 21, p. 132 – 160. 52. Graham, J.R., and C.R. Harvey, 2001, “The Theory and Practice of Corporate Finance: Evidence from the Field,” Journal of Financial Economics, 60, 187 – 243. 53. Graham, J.R., Harvey, C.R. and S. Rajgopal, 2005, “The Economic Implications of Corporate Financial Reporting, Journal of Accounting and Economics, forthcoming. 54. Grossman, Sanford J. and Joseph E. Stiglitz, 1980, “On the Impossibility of Informationally Efficient Markets,” American Economic Review, 70, p. 393 – 408. 55. Gruber, Martin J., 1996, “Another Puzzle: The Growth in Actively Managed Mutual Funds,” Journal of Finance, 51, p. 783 – 810. 56. Grundy, Bruce D. and J. Spencer Martin, 2001, “Understanding the Nature of the Risks and the Source of the Rewards to Momentum Investing,” Review of Financial Studies, 14, p. 29 – 78. 57. Guterbock, T.M., Meekins, B.J., Weaver, A.C., and J.C. Fries, 2000, “Web Versus Paper: A Mode Experiment in a Survey of University Computing,” paper presented at the annual meeting of the American Association for Public Opinion Research, Portland, OR. 58. Hadad, Mahmoud M. and Arnold L. Redman, 2005, “Ivory Tower Versus the Real World: Do I Practice What I Preach?” Financial Decisions, 17, p. 1 – 19. 59. Hair, Joseph F., Black, Bill, Babin, Barry, Anderson, Rolph E., and Ronald L. Tatham, 2006, Multivariate Data Analysis, Sixth Edition, Prentice Hall. 60. Hartikainen, Outi and Sami Torstila, 2004, “Job-Related Ethical Judgment in the Finance Profession,” Journal of Applied Finance, 14, p. 62 – 76. 61. Harris, Lawrence and Eitan Gurel, 1986, “Price and volume effects associated with changes in the S&P 500 list: New evidence for the existence of price pressures,” Journal of Finance 41, p. 815-829. 62. Hausman, Jerry, 2001, “Mismeasured Variables in Econometric Analysis: Problems from the Right and Problems from the Left,” Journal of Economic Perspectives, 15, p. 57 – 67. 63. Healy, Paul M. and Krishna G. Palepu, 1988, “Earnings Information Conveyed by Dividend Initiations and Omissions,” Journal of Financial Economics, 21, p. 149 – 175. 64. Heck, Jean and Philip Cooley, 2005, “Prolific Authors in the Finance Literature: A Half Century of Contributions,” Journal of Finance Literature, 1, p. 46 – 69. 65. Hirshleifer, David, 2001, “Investor Psychology and Asset Pricing,” Journal of Finance, 56, p. 1533 – 1597.
140 66. Hirshleifer, David and Tyler Shumway, 2003, “Good day sunshine: Stock returns and the weather,” Journal of Finance, 53, p. 1009. 67. Hu, L. and P.M. Bentler, 1999, “Cutoff Criteria For Fit Indexes In Covariance Structure Analysis: Conventional Criteria Versus New Alternatives,” Structural Equation Modeling, 6, p. 1–55. 68. Ikenberry, D; Lakonishok, J.; and T. Vermaelen, 1995, “Market Underreaction to Open Market Share Repurchases,” Journal of Financial Economics, 39, 181 – 208. 69. Ikeberry, D.; Rankine, G. and E. K. Stice, 1996, “What do Stock Splits Really Signal?,” Journal of Financial and Quantitative Analysis, 31, p. 357 – 375. 70. Jegadeesh, N; Kim, J.; Krische, S.; and C. M. C. Lee, 2004, “Analyzing the Analysts: When do Recommendations Add, Value?,” Journal of Finance, 59, p. 1083 – 1124. 71. Jegadeesh, Narasimhan and Sheridan Titman, 1993, “Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency,” Journal of Finance, 48, p. 65 – 91. 72. Jegadeesh, Narasimhan and Sheridan Titman, 2001, “Profitability of Momentum Strategies: An Evaluation of Alternative Explanations,” Journal of Finance, 56, p. 699 – 720. 73. Jensen, M. C., 1978, “Some Anomalous Evidence Regarding Market Efficiency,” Journal of Financial Economics, 6, p. 95 – 102. 74. Jorgensen, Randy D. and John R. Wingender, Jr., 2004, “A Survey on the Dissemination of Earnings Information by Large Firms,” Journal of Applied Finance, 14, p. 77 – 84. 75. Keim, D. B., 1983, “Size-Related Anomalies and Stock Return Seasonality: Further Empirical Evidence,” Journal of Financial Economics, 12, p. 13 – 32. 76. Keim, D. B. and R. F. Stambaugh, 1986, “Predicting Returns in the Stock and Bond Markets,” Journal of Financial Economics, 17, p. 357 – 390. 77. Krigman, Laurie, Shaw, Wayne H., and Kent L. Womack, 2001, “Why do Firms Switch Underwriters?” Journal of Financial Economics, 60, p. 245 – 284. 78. Kwak, N. and B.T. Radler, 2000, “Using the Web for Public Opinion Research: A Comparative Analysis Between Data Collected Via Mail and the Web,” presented at the annual meeting of the American Association for Public Opinion Research, Portland, OR. 79. Lakonishok, Josef and Seymour Smidt, 1984, “Volume and Turn-of-the-Year Behavior,” Journal of Financial Economics, 13, p. 435 – 425. 80. Lakonishok, Josef and BLANK Smidt, 1988, “Are seasonal anomalies real? A ninety-year perspective,” Review of Financial Studies 1, p. 403-425. 81. Lintner, John, 1965, “The Valuation of Risky Assets and the Selection of Risky Investments in Stock Portfolio And Capital Budgets,” Review of Economics and Statistics, 47, p. 13 – 37. 82. Lo, Andrew and Jiang Wang, 2000, “Trading Volume: Definitions, data analysis, and implications of portfolio theory,” Review of Financial Studies, 13, 257 – 300. 83. Loughran, T and J. Ritter, 1995, “The New Issues Puzzle,” Journal of Finance, 50, p. 23 – 52. 84. Malkiel Burton G., 1996, A Random Walk Down Wall Street, W.W. Norton and Company, New York, NY. 85. Marquering, Wessel, Nisser, John, and Toni Valla, 2006, “Disappearing Anomalies: a Dynamic Analysis of the Persistence of Anomalies,” Applied Financial Economics, 16, p. 291 – 302. 86. Marsh, Balla, and McDonald 1988 87. Medlin, C., Roy, S. & Ham Chai, T. (1999) “World Wide Web versus mail surveys: A comparison and report,” Paper presented at ANZMAC99 Conference: Marketing in the Third Millennium. November 28 – December 1, Sydney, Australia. www.singstat.gov.sg/conferences/ec/f112.pdf
141 88. Mitchell, Mark. L., and Erik Stafford, 2000. Managerial Decisions and Long-Term Stock Price Performance, Journal of Business 73, 287 – 329. 89. Moskowitz, Tobias and Mark Grinblatt, 1999, “Do Industries Explain Momentum?,” Journal of Finance, 54, p. 1249 – 1290. 90. Ogden, Joseph P., 2003, “The Calendar Structure of Risk and Expected Returns on Stocks and Bonds,” Journal of Financial Economics, 70, p. 29 – 67. 91. Pinegar, J.M. and L. Wilbricht, 1989, “What Managers Think of Capital Structure Theory: A Survey,” Financial Management, 18, p. 82 – 91. 92. Reinganum, Marc, 1981, “Misspecification of Capital Asset Pricing: Empirical Anomalies Based on Earnings Yields and Market Values,” Journal of Financial Economics, 9, p. 19 – 46. 93. Reinganum, Marc, 1983, “The Anomalous Stock Market Behavior of Small Firms in January: Empirical Tests for Tax-Loss Selling Effects,” Journal of Financial Economics, 12, p. 89 – 104. 94. Roll, Richard, 1983, “Vast Ist Das? The Turn of the Year Effect and the Return Premia of Small Firms,” Journal of Portfolio Management, 9, p. 18 – 28. 95. Roll, Richard, 1994, “What Every CEO Should Know About Scientific Progress in Financial Economics: What is Known and What Remains to be Resolved,” Financial Management, 23 p. 69 – 75. 96. Rosenberg, Barr; Reid, Kenneth; and Ronald Lanstein, 1985, “Persuasive Evidence of Market Inefficieny,” Journal of Portfolio Management, 11, p. 9 – 17. 97. Ross, Stephen. A., 1976, “The Arbitrage Theory of Capital Asset Pricing,” Journal of Economic Theory, 13, p. 341 – 360. 98. Rozeff, Michael S. and William R. Kinney, Jr., 1976, “Capital Market Seasonality: The case of stock returns,” Journal of Financial Economics, 3, p. 379 – 402. 99. Russell, Philip and Violet Torbey, 2002, “The Efficient Market Hypothesis on Trial: A Review,” Business Quest Journal, January 2002, 1-19. 100. Schaefer, David R. and Don A. Dillman, 1998, “Development of a Standard E-Mail Methodology: Results of an Experiment,” Public Opinion Quarterly, 62, p. 378 – 397. 101. Schwert, G. William, 2002, “Anomalies and Market Efficiency,” NBER Working Paper No. 9277, National Bureau of Economic Research. Also included as a chapter in the Handbook of the Economics of Finance, edited by George Constantinides, Milton Harris,a nd Rene M. Stulz. 102. Sharpe, W. F., 1963, “A Simplified Model for Portfolio Analysis,” Management Science, 9, p. 277 – 293. 103. Sharpe, W. F., 1964, “Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk,” Journal of Finance, 19, p. 425 – 442. 104. Shleifer, Andrei, 1986, “Do demand curves for stocks slope down,” Journal of Finance 41, p. 579-590. 105. Spiess, D. and J. Affleck-Graves, 1995, “Underperformance in Long-Run Stock Returns Following Seasoned Equity Offerings,” Journal of Financial Economics, 38, p. 243 – 267. 106. Stattman, Dennis, 1980, “Book Values and Stock Returns,” The Chicago MBA: A Journal of Selected Papers, 4, p. 25 – 45. 107. Stickel, Scott, 1985, “The effect of value line investment survey rank changes on common stock prices,” Journal of Financial Economics 14, p. 121-144. 108. Swidler, Steve and Elizabeth Goldreyer, 1998, “The Value of a Finance Journal Publication,” Journal of Finance, 53, p. 351 – 363. 109. Trahan, E. A. and L. J. Gitman, 1995, “Bridging the Theory-Practice Gap in Corporate Finance: A Survey of Chief Financial Officers,” Quarterly Review of Economics and Finance, 35, p. 73 – 87.
142 110. Tucker, L. R. and C. Lewis, 1973, “A Reliability Coefficient For Maximum Likelihood Factor Analysis,” Psychometrika, 38, p. 1–10. 111. Tversky, Amos and Daniel Kahnemann, 1991, “Loss Aversion in Riskless Choice: A Reference Dependent Model,” Quarterly Journal of Economics, 106, p. 1039 – 1061. 112. Van Selm, Martine and Nicholas W. Jankowski, 2006, “Conducting Online Surveys,” Quality & Quantity, 40, p. 435 – 456. 113. Welch, Ivo. “Views of Financial Economists on the Equity Premium and on Professional Controversies.” The Journal of Business 73–4, 501–537, October 2000. 114. Wermers, Russ, 2000, “Mutual Fund Performance: An Empirical Decomposition into Stock-Picking Talent, Style, Transaction Costs, and Expenses,” Journal of Finance, 55, p. 1655 - 1703. 115. Wright, K. B., 2005, “Researching Internet-based populations: Advantages and disadvantages of online survey research, online questionnaire authoring software packages, and web survey services,” Journal of Computer-Mediated Communication, 10(3), article 11. http://jcmc.indiana.edu/vol10/issue3/wright.html
143 BIOGRAPHICAL SKETCH
Colby Wright was born February 21, 1976 in Lewistown, MT to wonderful and supportive parents Richard Wright and Judith Emerson Wright. Upon graduation from Fergus High School in Lewistown, MT in 1994, he attended the University of Utah in Salt Lake City for one year before serving a life-changing two year mission to the Philippines as a representative of the Church of Jesus Christ of Latter Day Saints from October of 1995 to October of 1997. Shortly after returning from his mission, he met and fell in love with Misty May whom he married on August 11, 1998 in the Jordan River temple of the LDS Church. They now have four children, the joys of Colby’s life: Herbie, Teddy, Donna, and Dash. Colby graduated magna cum laude (top 5% university wide) from Brigham Young University in December 2001 with a B.S. in accountancy. After graduating from BYU, he worked in the audit division of Ernst & Young in their Tyson’s Corner, VA (a suburb of Washington, D.C.) office for a little over a year, during which time he passed the CPA exam and obtained his official CPA license from the state of Virginia, which he maintains to this day. Recognizing his head and heart were much more interested in finance than accounting, he made the difficult decision to move his growing family from Centreville, VA to Tallahassee, FL so that he could pursue a Ph.D. in finance at Florida State University. While at Florida State University, Colby has received numerous awards and distinctions. He was awarded a Florida State University Fellowship for the 2004 – 2005 academic year. He was also a recipient of the College of Business Teaching Fellowship for the 2003 – 2004 academic year. Colby has also been recognized for his outstanding performance as an instructor. He was one of 15 graduate students awarded the university-wide graduate teaching award in 2006. In that same year he received the College of Business doctoral teaching award. He also recently had a paper accepted for presentation at the 2007 Midwestern Finance Association Meetings in Minneapolis Minnesota.
144 He will, again, be moving his family this summer. He recently accepted a tenure- track assistant professor position at Central Michigan University and eagerly looks forward to a future of productive research, a long career teaching and interacting with students, and many, many nights playing UNO with his children and snuggling with his wife on the couch as they watch movies.
145