*Title Page/Author Identifier Page/Abstract

Accounting Anomalies and Fundamental Analysis: A Review of Recent Research Advances*

Scott Richardson Barclays Global Investors Scott.Richardson@barclaysglobal.com rem Tuna London Business School ituna@london.edu Peter Wysocki University of Miami School of Business Administration pwysocki@bus.miami.edu

September 2009 Comments welcomed.

Abstract:
This paper surveys recent research advances in the areas of accounting anomalies fundamental analysis. We use investor forecasting activity as an organizing framework for the three main parts of our survey. The first part of the survey highlights recent research advances. The second part presents findings from a questionnaire given to investment professionals and academics on the topics of fundamental analysis and anomalies research. The final part outlines several new empirical techniques for evaluating accounting anomalies and suggests directions for future research. JEL classification: G12; G14; M41 Key words: Accruals; Anomalies; Forecasting; Fundamental analysis; Market efficiency; Risk

*Manuscript

Accounting Anomalies and Fundamental Analysis: A Review of Recent Research Advances

September 2009

Abstract:
This paper surveys recent research advances in the areas of accounting anomalies fundamental analysis. We use investor forecasting activity as an organizing framework for the three main parts of our survey. The first part of the survey highlights recent research advances. The second part presents findings from a questionnaire given to investment professionals and academics on the topics of fundamental analysis and anomalies research. The final part outlines several new empirical techniques for evaluating accounting anomalies and suggests directions for future research. JEL classification: G12; G14; M41 Key words: Accruals; Anomalies; Forecasting; Fundamental analysis; Market efficiency; Risk

1. Introduction
Objective The editors of the Journal of Accounting and Economics gave us the assignment to review the literature on accounting anomalies and fundamental analysis. Given the existence of numerous excellent prior literature surveys of closelyrelated topics such as market anomalies, market efficiency, fundamental analysis and behavioral finance, we have constructed our literature survey to complement and fillin-the-gaps left by related literature surveys. These prior surveys include Barberis and Thaler (2003), Bauman (1996), Bernard (1989), Byrne and Brooks (2008), Damodaran (2005), Easton (2009a), Fama (1970), Fama (1991), Hirshleifer (2001), Keim and Ziemba (2000), Kothari (2001), Lee (2001), Schwert (2003), and Subrahmanyam (2007). To complement these literature surveys, we focus on

research studies that: (i) have publication or distribution dates after the year 1999, (ii) examine accounting-related anomalies and fundamental analysis geared toward forecasting future earnings, cash flows and security returns, and (iii) focus on empirical research methodologies. An underlying theme of our survey is that information contained in general purpose financial reports helps investors make better portfolio allocation decisions. To this end, an investor can use information in general purpose financial reports to forecast free cash flows for the reporting entity, estimate the risk of these cash flows, and ultimately make an assessment of the intrinsic value of the firm which will be compared to observable market prices. We view this forecasting activity as the fundamental organizing principle for research on accounting anomalies and

fundamental analysis.1 While we recognize the co-existence of other accounting properties and objectives, we view forecasting as a powerful organizing concept for reviewing the recent literature on accounting anomalies and fundamental analysis. As part of our review, we adopt a number of complementary approaches to identify, organize and capture recent advances in this literature. The first part of our review tabulates a list of the most highly-cited research studies on accounting anomalies and fundamental analysis published or distributed since the year 2000. We also organize and categorize these highly-cited studies by identifying their common and overlapping citations to earlier papers in the literature. The second part of our survey presents results from a questionnaire of investment professionals and accounting academics about their opinions on investment anomalies and fundamental analysis and how academic research has informed investment practice. In the final part of our review, we offer suggestions for future research and draw on recent conceptual advances from both investment practice and academic research to demonstrate a more-encompassing definition and treatment of risk and transaction costs in empirical tests of equity market anomalies. Specifically, we propose a benchmark empirical model and then apply it to a case study of the relation between accruals and future stock returns for a sample of U.S. firms.2 The primary objective of our review is to produce a valuable research reference not only for academics and graduate students, but also for investment professionals. In addition, the findings from our questionnaire of investment
1

We keep the discussion of accounting anomalies and fundamental analysis distinct from each other as this is how the literature has evolved. But we note that fundamental analysis could be characterized as subsuming the accounting anomaly literature (i.e., both have primary goals of forecasting earnings and returns).
2

We choose the accruals anomaly as our case study because it is the most frequently-cited accounting anomaly over the period of our literature review. See section 2 for an analysis of citations and impact of research studies published since the year 2000.

professionals and academics highlight the spillovers from academic research to professional practice because, relative to other academic accounting research topics, academic research on anomalies and fundamental analysis has very direct applications and intellectual spillovers to actual practice. Accounting anomalies and fundamental analysis also have direct intellectual connections to the efficient markets and behavioral finance literatures in financial economics. Given these linkages, we now briefly summarize the coverage of prior related literature surveys in accounting and finance. Coverage of previous literature surveys Literature reviews of the academic literature on efficient markets have origins going back to Fama (1970). Given that financial market anomalies and market efficiency are two sides of single intellectual debate, prior surveys attempt to capture the tensions in this debate and give insights about the extent to which markets are informationally efficient (see, for example Kothari, 2001 and Lee, 2001). Surveys that summarize the literature in the 1980s and 1990s include Keim and Ziemba (2000), Hirshleifer (2001), Barberis and Thaler (2003), and Schwert (2003). More recent surveys that focus on papers in the finance literature include Subrahmanyam (2007), and Byrne and Brooks (2008). These surveys cover issues related to market efficiency, technical, fundamental and event-driven anomalies, and the now maturing field of behavioral finance. Papers that review the literature on accounting-based anomalies and fundamental analysis include Baumans (1996) survey of the fundamental analysis literature up to the mid-1990s and Kotharis (2001) broad survey of capital markets research in accounting (with a related discussion by Lee, 2001). While exhaustive at the time, Kothari (2001) and Lee (2001) cover the literature only up to the year 2000. Recent surveys by Damodaran (2005) and Ohlson

(2009) provide insightful technical overviews of finance and accounting valuation models. Similarly, Easton (2009) provides a literature review of and applications of implied cost of capital methods which have strong foundations in fundamental analysis. Below we present summary statistics of the coverage and focus of prior related surveys to provide a perspective on the coverage (or lack thereof) of this broad literature. Bauman (Journal of Accounting Literature, 1996) provides a focused overview of fundamental analysis research in accounting. He covers 66 papers that were published between 1938 and 1997 and 40 of these papers were published in academic accounting journals (including 11 papers from the Journal of Accounting Research, 9 papers from The Accounting Review, and 4 papers from the Journal of Accounting and Economics). Bauman (1996) does not focus on research related to financial market anomalies. Hirshleifer (Journal of Finance, 2001) provides a survey of research on investor psychology and asset pricing. He broadly covers 543 papers published up to the year 2001. Many behavioral finance papers began to be published around this time and 110 of the papers covered in his survey were either published or distributed in the years 2000 and 2001. Understandably, the vast majority of the papers in this survey are drawn from finance, economics and psychology journals. Fewer than 10 papers in the survey are from accounting journals. Fundamental analysis and other accounting-related topics with possible behavioral foundations are not highlighted in this survey. Schwert (Handbook of the Economics of Finance, 2003) surveys the finance literature on anomalies and market efficiency. He covers 107 papers published in finance and economics journals between 1933-2003, including 23 papers that were

published or distributed between 2000 and 2003. No accounting papers are included in the survey. In the same handbook, Barberis and Thaler (2003) survey the behavioral finance literature. They cover 204 papers between 1933-2003, including 66 papers published between 2000 and 2004. They only mention one paper published in an accounting journal (Bernard and Thomas, 1989). Subrahmanyam (European Financial Management, 2007) provides a review and synthesis of the behavioral finance literature. He reviews 155 papers published between the years 1979 and 2007, with the majority of the papers published in the year 2000 or later. The vast majority of the surveyed papers come from finance journals and only one cited working paper was eventually published in an accounting journal. Finally, Byrne and Brooks (Research Foundation of CFA Institute Monograph, 2008) provide a practitioner-focused survey of the current state of the art theories and evidence in behavioral finance. They review 79 papers published between the years 1979 and 2008, with the majority of the papers published in the year 2000 or later. They include 33 papers published in the Journal of Finance and 7 papers published in either the Journal of Financial Economics or the Review of Financial Studies. Only 1 reviewed paper come from an accounting journal (Journal of Accounting and Economics). A quick scan of these survey papers reveals where and when the prior surveys captured innovations in the literature. While Kothari (2001) and Lee (2001) provide an excellent coverage of research on anomalies and fundamental analysis in the accounting literature up until the year 2001, no survey covers papers in the accounting literature after that year. Furthermore, recent finance surveys on anomalies focus almost exclusively on behavioral finance and do not cover accounting anomalies or

fundamental analysis. Therefore, one of the goals of our survey is to fill in some of the gaps of prior literature surveys and capture research innovations since the year 2000. What we dont cover Our survey focuses on empirical research on accounting anomalies and fundamental analysis. However, empirical research is (or should be) informed by theory, since interpretation of empirical analysis is impossible without theoretical guidance. As we stated above, we do not review in detail papers already covered in prior surveys (especially papers published prior to the year 2000). In addition, within the empirical capital markets area, there are concurrent Journal of Accounting and Economics survey papers that may overlap with some of the topics covered in our survey [see, for example, Beyer, Cohen, Lys and Walther (Corporate Information Environment, 2009), and Dechow, Ge and Schrand (Earnings Quality and Earnings Management, 2009). Accordingly, we do not discuss in detail research papers in these areas, although we do reference them. Summary of main observations Our first major observation is based on a citation analysis of recent published and working papers on accounting anomalies and fundamental analysis. This citation analysis lets the academic research market speak on which research papers on accounting anomalies and fundamental analysis have attracted the attention of other researchers and have had a meaningful impact on the subsequent literature. While many of the most highly-cited papers are from finance journals, there are some very influential papers from accounting journals that are broadly cited in both accounting and finance journals (see, for example, Xie, 2001, and Richardson, Sloan, Soliman and Tuna, 2005).

Our second major observation is based on a complementary citation analysis that helps us organize the literature on accounting anomalies and fundamental analysis. Specifically, we analyze papers written or published since the year 2000 to identify common references of prior published research studies. This approach allows us to identify common themes or clusters of research topics. Our analysis reveals four main clusters of overlapping citations to common sets of prior papers. We apply the following labels to the four clusters of research papers: Fundamental Analysis, Accruals Anomaly (including related investment anomalies), Underreaction to Accounting Information (including PEAD and other forms of momentum), and Pricing Multiples and Value Anomaly. These four main clusters largely span the literature. The Fundamental Analysis cluster cites a number of prior foundational papers including Abarbanell and Bushee (1997 and 1998) and Feltham and Ohlson (1995). The citation foundation of the Accruals Anomaly cluster is based on the numerous citations to Sloan (1996) as the underlying prior research study that binds together this research cluster. The Underreaction to Accounting Information cluster most often cites Bernard and Thomas (1989, 1990), Foster, Olsen and Shevlin (1984), and Jegadeesh and Titman (1993) as foundational papers. The Pricing Multiples and Value Anomalies cluster is bound together by references to the foundational papers of Basu (1977), Reinganum (1981), Ball (1992), and Fama and French (1993 and 1995). We then use our forecasting framework to categorize, evaluate and discuss some of the main research advances since the year 2000 in each of the four research clusters. Our framework attempts to provide some unifying structure to the burgeoning empirical literature on accounting anomalies. We highlight that many of the anomalies are not unique and, in many cases, the apparent excess returns to a new anomaly are subsumed by other existing anomalies (see, for example, Dechow,

Richardson and Slaon, 2008, who document that the general accruals anomaly subsumes the external financing anomaly). We also explore why and how the anomalies persist in competitive markets, the robustness of the anomalies, and whether the observed returns are due to risk or mispricing. Our third major observation arises from a questionnaire we distributed to investment professionals (based on a survey of a subset of CFA members) and to accounting academics who teach and undertake research related to financial analysis. The questionnaire attempts to capture the important opinions of the creators and users of research on accounting anomalies and fundamental analysis. The findings suggest that many of the conventions and techniques used in academic research differ from those in the investment community. For example, in contrast to most empirical academic studies that use either the CAPM or the Fama-French 3-factor model for risk calibration, most survey respondents used other types of models. On the other hand, practitioners appear to have a robust interest in and demand for new academic research on fundamental analysis and anomalies. Interestingly, most respondents claimed that earnings or cash flow momentum has proven to be a successful active investment strategy in recent years while accounting quality has received less attention. Respondents also tend to use a range of fundamental valuation and analysis techniques in their work (including earnings multiples, book value multiples, cash flow multiples, and discounted free cash flow models Interestingly, only a small fraction of respondents frequently used residual income (economic profit) models for valuation. The survey respondents also indicated that they get most of their research insights from practitioner journals such as CFA Magazine, Financial Analysts Journal, and Journal of Portfolio Management, rather than academic publications such as the

Journal of Financial Economics, Review of Financial Studies, Journal of Accounting and Economics, Contemporary Accounting Research, or The Accounting Review. Both the practitioners and academics who completed our opinion survey placed high importance to future academic research on: (i) empirical tests of investor behavior; (ii) empirical tests of asset pricing, risk and factor models; (iii) empirical research on forecasting firm and industry fundamentals; and (iv) empirical discovery and investigation or new anomalies or signals. Next, based on: (i) the prominence of the accruals anomaly in the recent literature, and (ii) practitioner interest in future innovations related to empirical tests of investor behavior and empirical tests of asset pricing, risk and factor models, we conduct our own empirical analyses to help advance some concepts and approaches to be considered and applied in future research studies. Specifically, we provide new insights on: (i) the time-series variation in the negative relation between accruals and future returns (specifically, the extent to which this relation has disappeared, which is consistent with market learning), and (ii) whether the relation is robust to a more comprehensive empirical treatment of risk and transaction costs. Our empirical analysis shows that the negative relation between accruals and future stock returns has greatly attenuated over time. In recent years one could conclude that the information in accruals is now fully priced by the market, which is consistent with the market learning explanation and inconsistent with the academic research that has suggested accruals are a priced risk factor. As discussed in section 5, the time-varying

association between accruals and future stock returns creates a natural setting where researchers can evaluate the changes in the macroeconomic environment that prevented / allowed this risk factor to generate a premium.

Finally, we provide suggestions for future research on accounting anomalies and fundamental analysis. Based on our citation analysis, literature review, practitioner/academic questionnaire, and empirical analyses, we identify five major areas of opportunity. First, there is a lack of research that utilizes contextual information such as industry, sector and macro-environmental data to forecast future earnings, cash flow, risk and value. Second, current research does not fully exploit the wealth of information contained in general purpose financial reports but is outside of the primary financial statements. With the advent of XBRL and improved textual extraction techniques, this information could be used to improve forecasts of free cash flows, risk and firm value. Third, there appear to be limitations to current forecasting techniques and opportunities to overcome these limitations. Fourth, we discuss the use of accounting information by external capital providers beyond common equity holders. With the increased development of credit markets in the last decade there is now a wealth of data available on credit related instruments that can be used to help make inferences about the usefulness of accounting information for a wider set of capital providers. Fifth, we note that many capital market participants are using the same information sources to forecast the future and this has lead to a very crowded space in the investment world. We note that future research into the (mis)pricing of accounting information should undertake a more rigorous analysis of risk and the impact of transaction costs on the implementability of a given investment idea in a crowded information space with many users applying the same information and techniques. Outline of the rest of the paper Section 2 uses citation analysis to identify high impact papers from the recent literature on anomalies and fundamental analysis and organize the literature into four
10

main research clusters. Section 3 summarizes the results of a questionnaire of investment professionals and accounting academics opinions on academic research related to fundamental analysis and equity market anomalies. Section 4 provides a synthesis of recent advances in each of research clusters identified above. Section 5 presents a benchmark model for evaluating accounting anomalies using a moreencompassing definition and treatment of risk and transaction costs (with a specific case study of the relation between accruals and future stock returns for a sample of U.S. firms). Building on findings in section 2-5, we then discuss our suggestions for future research in section 6. Finally, section 7 summarizes and concludes.

2. Citation and cluster analysis In this section we utilize well established techniques to help identify specific high-impact papers and key research areas related to accounting anomalies and fundamental analysis. We then group recent research papers into four clusters based on their common citations to prior studies in the literature to identify the key topics for our subsequent literature review.

2.1 Identifying important recent papers on anomalies and fundamental analysis Our survey focuses on research studies published or circulated since the year 2000 to complement the prior literature reviews by Kothari (2001) and Lee (2001). As a starting point, we let the market speak and use academic citation data to identify high impact research papers on anomalies and fundamental analysis. Using citation analysis to quantify research impact has solid foundations in the accounting literature. There exist a number of citation-based studies of the prior general accounting literature including McRae (1974), Brown and Gardner (1985a and 1985b), and
11

Brown and Huefner (1994).3 In general, academic citation analyses utilize the number of citations listed on the ISI Web of Science and the SSCI (Social Sciences Citation Index).4 However, this citation data can paint an incomplete and stale picture of important recent developments and innovations in an academic field. Moreover, with the advent of the internet and research sites such as the Social Sciences Research Network (www.ssrn.com) and Research Papers in Economics (www.repec.org), working papers are quickly and widely cited by other researchers working papers and published research papers. Therefore, to capture a broad and timely picture of recent papers on accounting anomalies and fundamental analysis literature, we apply the methodology of Keloharju (2008) and analyze citations using results returned by Google Scholar, a service that complements the citations generated by the core journals covered by ISI Web of Science with citations by other journals and, more importantly, by working papers. The citations on Google Scholar are timely and include references to and from both working papers and published papers. We collect the citation data using the general citation search function of Anne-Wil Harzings Publish or Perish program, downloadable at http://www.harzing.com/. This program uses on-line data from Google Scholar to generate a list of published and working papers cumulative number of citations to each paper. Given that the cumulative number of citations to a research study depends not only on impact, but also by the passage of time since its original circulation or publication, we follow Schwert (2007) to account for this age effect
3

There are also some of citation analyses of sub-fields of accounting research (see, for example, the citation analysis of the management accounting literature by Hesford, Lee, Van der Stede and Young (2007).
4

For example, Schwert (2007) uses ISI Web of Science citation data to rank papers published in the Journal of Financial Economics between 1974 and 2005 by the number of citations per year. Citations reported in ISI Web of Science are for published papers that receive citations from other published papers drawn from a set of widely-read academic journals.

12

and divide the cumulative number of citations by the number of years since original circulation or publication of a paper. We construct a list of the most highly-cited recent papers by first performing a keyword search on the ssrn.com e-library database to identify candidate working papers and published papers related that to financial market anomalies and fundamental analysis.5,6 We then scan the titles and abstracts of the candidate papers to determine if they:(i) were posted or published after the year 1999, and (ii) focus on or have implications for empirical tests of accounting anomalies and fundamental analysis. We then obtain citation counts for these papers from Google Scholar using the Publish or Perish program. We collect citations to both working paper versions and published versions of each paper and combine duplicate entries to the same article and correct erroneous title, year, and publication year information. 2.1.1 Citation impact results For the sake of brevity, the full list of the most highly-cited research papers on anomalies and fundamental generated by our search of Google Scholar can be obtained from the authors directly. At the top of the list, the ten papers with the highest average number of citations per year are: 1) Jegadeesh and Titman (Journal of Finance, 2001), Profitability of momentum strategies: an evaluation of alternative explanations. 2) Hong, Lim, and Stein (Journal of Finance, 2000), Bad news travels slowly: size, analyst coverage, and the profitability of momentum strategies.

The keyword search on SSRN included separate searches based on the following key words in the title or abstract of papers posted on SSRN: accounting anomaly, fundamental analysis, fundamental accounting, valuation fundamental, accounting inefficiency, market inefficiency, earnings drift, price multiple, book market equity, accruals anomaly, and accounting reaction. We also use the bibliographic references in these papers to identify other recent papers on accounting anomalies and fundamental analysis that were not captured by our initial keyword searches on SSRN.
6

The bibliographic references contained in each paper are also used to classify related research papers and topics. This analysis is discussed in the next sub-section.

13

3) Diether, Malloy and Scherbina (Journal of Finance, 2002), Differences of opinion and the cross section of stock return. 4) Zhang (Journal of Finance, 2005), The value premium. 5) Chan, Chan, Jegadeesh, and Lakonishok (Journal of Business, 2006), Earnings quality and stock returns. 6) Lewellen (Journal of Financial Economics, 2004), Predicting returns with financial ratios. 7) Zhang (Journal of Finance, 2006), Information uncertainty and stock returns. 8) Xie (Accounting Review, 2001), The mispricing of abnormal accruals. 9) Richardson, Sloan, Soliman, and Tuna (Journal of Accounting and Economics, 2005), Accrual reliability, earnings persistence and stock prices. 10) Vuolteenaho (Journal of Finance, 2002), What drives firm-level stock returns?

Of the 165 papers, there are 54 papers published in accounting journals. The 10 papers published in accounting journals with the highest average citations per year are: 1) Xie (Accounting Review, 2001), The mispricing of abnormal accruals. 2) Richardson, Sloan, Soliman, and Tuna (Journal of Accounting and Economics, 2005), Accrual reliability, earnings persistence and stock prices. 3) Hirshleifer and Teoh (Journal of Accounting and Economics, 2003), Limited attention, information disclosure, and financial reporting. 4) Khan (Journal of Accounting and Economics, 2008). Are accruals mispriced evidence from tests of an intertemporal capital asset pricing model. 5) Mashruwala, Rajgopal, and Shevlin (Journal of Accounting and Economics, 2006), Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs. 6) Fairfield, Whisenant, and Yohn (The Accounting Review, 2003), Accrued earnings and growth: implications for future profitability and market mispricing. 7) Beneish, and Vargus (The Accounting Review, 2002), Insider trading, earnings quality, and accrual mispricing. 8) Desai, Rajgopal, and Venkatachalam (The Accounting Review, 2004), Valueglamour and accruals mispricing: one anomaly or two? 9) Pincus, Rajgopal, and Venkatachalam (The Accounting Review, 2007), The accrual anomaly: international evidence. 10) Bartov, Radhakrishnan, and Krisy (The Accounting Review, 2007), Investor sophistication and patterns in stock returns after earnings announcements. 2.2 Organizing the literature: common citations to prior work In the previous sub-section we used citation analysis of both published papers and working papers to let the market for academic research reveal which research papers on accounting anomalies and fundamental analysis have attracted the attention of other researchers and therefore had an influenced on the subsequent literature. To
14

complement this citation analysis, we organize the literature by identifying clusters of research papers that have overlapping references of prior research studies. In order to identify clusters of papers and topics, we look for common citation patterns across research papers. We start with the sample of highly-cited papers in section 2.1 and then gather all citations from these papers to other research papers. Each unique cited research paper is given an identifying code. 7 After coding each cited paper, we perform a k-means cluster analysis of overlapping citations from papers in our main sample. We limit the number of possible clusters to less than six to create a tractable mapping of the literature. The cluster analysis reveals four main clusters of overlapping citations to common sets of prior papers. Upon examination of papers in the four main clusters, we assign the clusters the following labels: Fundamental Analysis, Accrual Anomaly, Underreaction to Accounting Information including PEAD, Pricing Multiples and Value Anomaly. These four main categories largely span the literature. In addition, the four clusters include subcategories of related studies such as investment anomalies (falling within the Accruals Anomaly cluster), return momentum (falling within the Underreaction to Accounting Information cluster), and information uncertainty (as it relates to Underreaction to Accounting Information).8

This coding process was partially automated and, as a result, was subject to some errors as some papers in our sample cite the working paper version of a study, while other papers include a more upto-date citation of the published version of the same study. In addition, there are also possible transcription errors by both authors of the papers and by us in tabulating references to create the citation database.
8

Again, for the sake of brevity, the full tabulation of papers within each cluster are available from the authors upon request.

15

3. Practitioners and academics opinions on anomalies/fundamental analysis In addition to our citation analysis of high impact researcher papers on accounting anomalies and fundamental analysis, we supplement this with views from the academic and practitioner communities. In this section, we highlight some of the key responses received from the academic and practitioner respondents to the questionnaire. Throughout the rest of our survey, we also attempt to weave the respondents insights into our review of the literature (section 4), and into our suggestions for research (section 6). Past and future demand for research on accounting anomalies and fundamental analysis potentially is partially influenced by what is happening in practice. Therefore, to assess the relevance of past research and help inform directions for future research, we surveyed investment professionals and academics to gain a better understanding of how they view the state of the art on the fundamental analysis and anomalies. Moreover, we wanted to document any differences in opinions on research between these two major constituents. Finally, we wished to assess the awareness, demand for, and use of academic research on accounting anomalies and fundamental analysis. 3.1 Practitioner questionnaire To survey the opinions of investment professionals, we worked in cooperation with the market research group of the CFA Institute to construct and administer a mini-survey of investment professionals. We focused on the broad topic of academic research on investment strategies, accounting anomalies and fundamental analysis. We constructed the survey questions in order to capture how investment professionals apply fundamental analysis and other quantitative techniques in their daily job activities and how academic research informs their practice. In addition, we included

16

questions about the sources and uses of research information (including academic research) for their daily job activities. The market research team from the CFA Institute provided suggestions on the format of the questions that would maximize the likelihood and usefulness of survey responses. In spite of our interest to obtain additional information about the demographics of the practitioner respondents, the CFA Institute market research team had concerns about collecting detailed demographic information from respondents. As a result, the CFA Institute survey did not capture detailed demographic information from the practitioner respondents. In addition, we had to work within the CFA resource constraint which likely limited the final response rate and affected the overall survey structure. Once the CFA Insitute survey was distributed, we used a similar format for the academic survey. The practitioner survey was administered and distributed by the CFA Institute via e-mail on January 26, 2009. A reminder e-mail was sent to non-respondents February 10, 2009 and the survey closed on February 12, 2009. The population from which the sample was drawn consisted of all active members of the CFA Institute, excluding those without a valid e-mail address and those that requested not to be sent e-mails or surveys. The sample was generated using a stratified random sampling technique; this produced a representative sample of 6,000 members to receive the survey based on key demographics (in this case, region and years holding the CFA charter). The distribution of the survey sample across these two areas is shown in the chart below. There were 201 usable responses were obtained, giving an overall response rate of 3.4%. 3.2 Academic questionnaire In order to benchmark and contrast the practitioners opinions, we sent the questionnaire described in section 3.1 to a set of academics who work and teach in the

17

field of financial analysis. The sample of academics was identified by randomly selecting: (i) 40 active researchers whose names appears in the academic references listed at the end of this paper, and (ii) 40 accounting academics who teach financial statement analysis (FSA) classes to MBA students. The sample of FSA teachers was identified from a Google search using the combined search terms: MBA, Financial Statement Analysis, and Syllabus.9 The e-mail questionnaire was sent out to the sample of academics in May and June of 2009. The cutoff for the academics responses was June 30, 2009. As of that date, 63 out of 80 (79%) of the academics in the sample responded to the survey questions. The number of academic respondents for each question is listed in Table 1. The high response rate likely resulted from the fact that the e-mailed survey directly identified the purpose of the survey (i.e., for the upcoming Journal of Accounting and Economics Conference) as well as the likely familiarity of the respondents with the names of the accounting academics who directly distributed the e-mail survey. 3.3 Analysis of outcomes of survey questions Table 1 provides a summary tabulation of the responses to each of the survey questions. The samples consist of (i) 201 practitioner responses to the questionnaire, and (ii) 63 academic responses to the questionnaire. The test of difference across the sample mean for each answer is calculated using a chi-square test of populations of unequal size and unequal variance. The p-values are adjusted using Cochran-Coxs approximation of the degrees of freedom for the unmatched samples.

Additional factors influencing the selection of the sample of FSA teachers includes: (a) the availability of the FSA teachers valid e-mail address as generated from the Google search criteria, and (b) the ranking of the FSA teachers website/web presence as generated by Google (we sequentially gathered e-mail addresses based on the appearance of web hits generated from the original Google search criteria).

18

While there are many consistent responses across the sample of practitioners and academics, we wish to highlight and analyze some of key differences in views across the two samples of respondents. Specifically, Question 1 of the survey asked Which risk model is most appropriate for risk calibration of an equity trading strategy? There is a large gap between the opinions of academics and practitioners. While 55% of academics recommend some variation of the Fama-French 3-factor model, only 29% of practitioners recommended this approach. The largest fraction of practitioners (35%) recommended the use of a CAPM model with industry and size adjustments, while only 7% of academics recommended this approach. This

observation suggests a striking difference between how academics and practitioners assess risk. We revisit this issue directly in section 5 and point to this issue in our suggested directions for future research in section 6. Another area of major difference of opinion arises in Question 4 of the survey which focuses on which techniques had been used and generated successful outcomes for equity trading strategies. In this area, there are large differences of opinion in the success of various strategies over the past decade. While 61% of practitioner respondents claimed that earnings or cash flow momentum was successful, only 22% of academic respondents believed that this type of strategy was successful. Similarly, 57% of practitioner respondents claimed that growth strategies were successful, while only 22% of academic respondents believed that growth strategies were successful, and 56% of respondents claimed that value strategies were successful. On the other hand, 70% of academic respondents believed that accounting quality was a successful strategy over the past decade which far exceeds the 41% of practitioner respondents who believe that this signal was successful over the same period. These differences in opinions point to possible differences in: (i) how expected returns and risks are
19

measured, (ii) how trade impact and transactions costs are quantified and accounted for in trading models, and (iii) how research data differ across academics and practitioners. We highlight these issues in section 5 and 6 of this paper and suggest ways to close the gap between academic and practitioners in the treatment of risk, trade impact, transactions costs, and data. 4. Overview of Key Research Papers Our organizing framework highlights how external investors use accounting information to forecast a firms future prospects including future earnings, cash flows, risk and returns. Overall, we view forecasting as the fundamental principle underlying academic research on accounting anomalies and fundamental analysis. Given the large number of published and working papers written since the year 2000, we also attempt to provide additional structure to the literature by classifying papers into related research clusters. As discussed in section 2 of this survey, our citation analysis generates four main clusters of research topics: Fundamental Analysis, Accruals Anomaly (including related investment anomalies), Underreaction to Accounting Information (with a particular emphasis on post-earnings announcement drift (PEAD)), and Pricing Multiples/Value Anomaly. We survey key studies in these main areas that have been circulated since the year 2000. For each area, we highlight various issues including risk versus mispricing, transactions costs, and limits to arbitrage that capture the essence of the debate in the literature.

4.1 Forecasting Framework The organizing framework for our survey is that investors forecast the level and risk of a firms free cash flows and then discount the free cash flows to estimate the value of claims to a firm. If the estimated value and the observed market value of

20

these claims diverge, then an investor must decide if current and forecasted future transactions costs and forecasted arbitrage risk point to a profitable arbitrage opportunity. Finance, valuation and financial statement analysis textbooks (see, for example, Penman, 2009, Easton et al., 2009) often use discounted free cash flow analysis as the basis for determining firm value: Total Firm Value0 = E0 [ t fcft /t ] (1)

where is the factor used to discount future total free cash flows (fcf) generated by the firm in periods t=1 . To derive this value, investors must forecast both future free cash flows and the risk of these cash flows.10 A future free cash flow (fcf) to the firm equals its operating profits not used to grow operating asset (see, for example, Penman and Zhang, 2006, and Easton et al., 2009).11 Therefore, as long as no components of operating income or net operating assets are booked directly to equity, fcf in period t can be is defined as: fcft = oit - noat (2)

where oit equals operating income and noat equals the change in net operating assets in period t. Alternatively, if the unknown variable of interest is operating income (oit), then equation (1) can be restated as: oit = fcft + noat (2)

10

Our forecasting framework focuses total cash flows generated by the firm (or enterprise) that are then available to all providers of capital (debt and equity). However, insights from our framework also flow though to analyzing equityholders claims.
11

Operating income available to the enterprise is also commonly referred to as Net Operating Profit After Tax (see, for example, Easton et al, 2009).

21

The next question is what determines operating income in period t? By definition, operating income is: oit = rnoat*noat-1 (3)

where rnoat represents the expected and unexpected flows generated by beginning of period net operating assets (noat-1). The recent accounting literature emphasizes that accounting and other sources of information help investors develop forecasts of the level (and risk) of the firms future free cash flows and operating income. Based on equation (2), it can be seen that both operating income and change in net operating assets are determinants of free cash flow. Furthermore, equation (3) highlights the role of initial level net operating assets in determining the level of operating income over a period. If one uses a simple 1-period forecasting model and the insights from equations (2) and (3), then next periods free cash flows or operating income are likely to be determined by this periods operating income (oit), change in net operating assets (noat), initial net operating assets (noat-1), and a Kx1 vector of other current period information (OTHERt): Et[fcft+1] = g{oit , noat, noat-1, OTHERt} Et[oit+1] = f{oit , noat , noat-1, OTHERt} (4) (4)

where f{} and g{} are (possibly non-linear) functions that help forecast future-period flows based on current-period accounting and non-accounting information.12 The set of non-accounting information can include information such as current market prices (Pt) and changes in current market prices (rt) of the firms securities, and the
12

Penman and Zhang (2006) also present a forecasting model for future operating income, but apply more restriction (less general) assumptions about the link between current and future accounting items.

22

accounting and non-accounting information of other firms (especially related firms such as the same industry as the primary firm of interest). Equations (4) and (4) can be further generalized based on the observations that operating assets and operating income can be: (i) decomposed into their constituent components, and (ii) these constituent components can provide additional forecasting power for future cash flows and operating income beyond the aggregated accounting numbers. Therefore, more generalized one-period-ahead prediction models of free cash flow and operating income can be expressed as: Et[oit+1] = F{OICt , NOACt, NOACt-1, OTHERt} Et[fcft+1] = G{OICt , NOACt, NOACt-1, OTHERt} (4-G) (4-G)

where OICt is a Mx1 vector of the constituent components of operating income (oit) and NOACt is a Nx1 vector of the constituent components of net operating assets (noat) such that oit = m=1,M oimt , noat = m=1,N noant , and noat-1 = m=1,N noant1.

Again, F{} and G{} are functions that help forecast future-period flows based on the

vectors of current-period accounting and non-accounting information. Equation (1) highlights that the value of the firm (and changes in this value) are derived from forecasts (and changes in forecasts) of future free cash flows and the risk of these cash flows. Therefore, the forecasting equations (4-G) and (4-G) suggest that accounting and non-accounting information in period t have the ability to predict one-period-ahead security returns (i.e., security returns due to risk or mispricing or possibly both). Therefore, our forecasting framework can be applied to security returns as: Et[rt+1] = H{OICt , NOACt, NOACt-1, OTHERt}
23

(5-G)

where H{} is a function that forecasts next-period security returns based on currentperiod accounting and non-accounting information. Again, this generalized framework allows for the non-accounting information set to include market information such as current market prices (Pt) and changes in current market prices (rt) of the firms securities. In addition, forecasts of future returns can capture both risk and mispricing. In the following sections, we use these general forecasting equations to present the forecasting concepts that underlie most recent research studies on fundamental analysis and accounting anomalies.

4.2. Fundamental Analysis Penman (2004) defines fundamental analysis as the analysis of information that focuses on valuation. Fundamental analysis is obviously of critical importance to investors as they care about how much to pay for an investment and for how much to sell it. As noted by Kothari (2001), an important motivation for fundamental analysis research and its use in practice is to identify mispriced securities relative to their intrinsic value for investment purposes. Hence, the majority of the fundamental analysis research in accounting seeks to come up with better forecasts of earnings or stock returns to assist the valuation or identification of mispriced securities. As a result, there is some overlap between research on fundamental analysis and accounting anomalies discussed later. The beauty of fundamental analysis is that it is of interest to the believers and non-believers of market efficiency, as fundamental analysis research can help us understand the determinants of value which assists in informed investment decisions and the valuation of non-publicly traded assets, for which market inefficiency is not a necessary condition.
24

In recent years, fundamental analysis research has generally focused on forecasting earnings, forecasting stock returns or estimating a firms cost of capital. Therefore, research studies based on fundamental analysis can be viewed through the lens of equations (3G), (3G) and (4G): Et[oit+1] = F{OICt , NOACt, NOACt-1, OTHERt}, and Et[fcft+1] = G{OICt , NOACt, NOACt-1, OTHERt}, and Et[rt+1] = H{OICt , NOACt, NOACt-1, OTHERt} Prior to 2000, there was a flurry of research that used accounting variables (and ratios of these variables) to predict future returns (see, for example, Ou and Penman, 1989, Lev and Thiagarajan, 1993, and Abarbanell and Bushee, 1997). In general, these studies either explicitly or implicitly took the ideas behind the above equations to develop prediction models of future returns. The direction of more recent on fundamentals-based return prediction has focused on context specific or refined sub-samples of firms where with higher likelihood of market imperfections which might increase the ultility of fundamental analysis. For example, Piotroski (2000) focuses on high B/M firms and demonstrates that, within the high B/M sample firms, those firms with the strongest fundamentals earn excess returns that are over 20% greater than firms with the weakest fundamentals. Similarly, Beneish, Lee and Tarpley (2001) use a two-stage approach towards financial statement analysis. In the first stage, they use market based signals to identify likely extreme performers. In the second stage, they use fundamental signals to differentiate between winners and losers among the firms identified as likely extreme performers in the first stage. These results suggest the possible benefits of carrying out fundamental analysis in specific sub-samples of firms whose securities are more likely to be mispriced.
25

4.2.1. Under what circumstances do stock prices deviate from fundamental value? Recent papers attempt to shed light on the type of stocks in which there will be a large wedge between fundamental value and market prices. For example, Baker and Wurgler (2006) define a number of investor sentiment proxies at the aggregate level. These include share turnover, the closed-end fund discount and first-day IPO returns. They find that stocks that are difficult to arbitrage (e.g., small, highly volatile ones) exhibit the maximum reversals in subsequent months when investor sentiment is high in a given period. Similarly, Zhang (2006) argues that stocks with greater information uncertainty (e.g., those which are small and have low analyst following) exhibit stronger statistical evidence of mispricing in terms of return predictability based on ex ante book-to-market ranking cross-sectional regressions. Nagel (2005) also provides evidence that the mispricing is greatest for stocks where institutional ownership is lowest; here institutional ownership is a proxy for the extent to which short-selling constraints bind (the assumption is that short-selling is cheaper for institutions). 4.2.2. Putting additional structure on forecasting activity to derive valuation models As discussed in Kothari (2001), the residual income model (Ohlson, 1995) has had a sizable impact on valuation approaches and the application of fundamental analysis in both academics and practice (see, also Claus and Thomas, 2001; Gebhardt, Swaminathan and Lee ,2001; Easton et al., 2002; Baginski and Wahlen, 2003). Ohlson (2005), Ohlson and Gao (2006), Ohlson (2009) and Easton (2009) provide excellent overviews of some of the technical and analytical advances in accountingbased valuation models over the past decade. Within our forecasting framework, the Ohlson (1995) model and its subsequent extensions use various simplifying assumptions to place additional
26

structure on the forecasting equations outlined in section 4.1. These structured forecasting equations are then used to create valuation models. All valuation models are theoretically the same and are merely transformations of the discounted free cash flow model (or a discounted dividend model) with varying assumptions and data requirements. However, the applicability and utility of a given valuation model depends on the plausibility of the assumptions underlying the model and the quality and availability of empirical data required by the model. Recent important advances in this area include both refinements of the valuation models and application of these models. A particularly-interesting example of a recent valuation refinement is the OJ model presented in Ohlson and JuettnerNauroth (2005). This model focuses on abnormal earnings growth (the OJ model) with no clean surplus accounting requirement that is generally found in previous models (such as the Ohlson, 1995). The OJ model differs from a traditional residual income model by specifying earnings per share as the fundamental forecasting benchmark. The proliferation of valuation models has spawned a growing debate about the superiority, applicability and empirical properties of various models (see, for example, Francis et al. 2000; Lundholm and OKeefe 2001; Penman 2001; Courteau et al. 2001; Richardson and Tinaikar 2004; Juettner-Nauroth and Skogsvik, 2005; Chen, Long and Shelly, 2008). These studies have compared the bias and accuracy of different valuation models. Not surprisingly, the various benchmarking studies conclude that different implementation techniques and the different underlying assumptions of various valuation models lead to different abilities of the models to predict future returns.

27

Existing studies generally argue that no single accounting-based valuation model has dominant empirical properties. However, the OJ model has some advantages over traditional the residual income models. Specifically, Ohlson (2005) analyzes a number of situations and concludes that truncation errors of terminal streams are smaller and less frequent under the OJ model compared to a traditional residual income model which relies on book equity as a performance benchmark. This implies that a finite-term OJ model will likely outperform a finite-term residual income model. Ohlson (2005) shows that capitalized earnings under the OJ model better capture the market value of equity than the book value of equity in a world of conservative accounting. Chen et al. (2005) also argue that the OJ model is better able to handle the dirty accounting systems observed in the real world because the OJ model does not reply on the clean surplus assumption which is fundamental to the residual income model. Ali et al. (2003) compare the ability of different valuation measures to predict future abnormal returns. They find that all of the valuation measures, including the OJ model, have the ability to predict future returns, and that the incremental contribution of the OJ model is significant in regressions of future returns on the value-price and B/M ratios. These findings suggest that the OJ model has some ability to predict future abnormal stock returns. 4.2.3. Determining Implied Cost of Capital Using Fundamentals Another approach investors use to forecast expected future returns is model and then estimate a firms cost of capital. Most empirical asset-pricing studies tend to rely on realized stock returns as a proxy for ex ante expected stock returns because expected stock returns are not directly observable. However, these estimates are

28

problematic because the estimates are imprecise (see, for example, Fama and French, 1997). To address some of the limitations of asset-pricing methods used to determine a firms cost of capital, recent accounting and finance studies (e.g., Claus and Thomas, 2001; Gebhardt, Lee, and Swaminathan, 2001; Pstor, Sinha, and Swaminathan, 2007, and Easton, 2009) propose an alternative approach to estimate expected returns: the implied or imputed equity cost of capital. The implied equity cost of capital of a company is the internal rate of return (IRR) that equates the companys stock price to the present value of all expected cash flows available to equity-holders. In other words, it is the discount rate that the market uses to discount the expected cash flows of the company This implied cost of capital approach relies heavily on forecasting a firms future cash flows. From a practical perspective, much of the work on implied equity cost of capital uses analysts forecasts of future earnings (rather than free cash flow to equity holders) as the key forecasting variable. The major advantage of the implied cost of capital approach to risk measurement is that it does not have to rely on noisy realized returns or on a specific asset pricing model other than that investors use a discounted future cash flows (dividends) to derive fundamental value. Therefore, the implied cost of capital approach applies standard fundamental valuation techniques and uses observed market prices and forecasts of earnings (cash flows) to derive the markets assessment of the equity risk (cost of capital) of a firm. For the firm as a whole, we can apply valuation equation (1) to a situation where investors observe total firm value in period t, forecast future FCFs, and then solve for r=-1: Total Firm Valuet = t FCFt /()t Reverse Engineer Equation 1

29

Given the simple and practical foundations of the implied cost of capital approach, it has been used in many recent studies related to empirical asset pricing (e.g., Chava and Purnanandam, 2007; Chen and Zhao, 2007; Pstor, Sinha, and Swaminathan, 2007) and also applied in other settings where cost of capital is an important market outcome (e.g., Francis, Khurana, and Pereira, 2005; Hail and Leuz, 2006). On the other hand, other recent studies suggest that the empirical outputs derived from the implied cost of capital approach are noisy, flawed or biased. Several studies attempt to correlate ex ante implied cost of capital of a firm with a companys observed ex post stock returns (e.g., Easton and Monahan, 2005; Guay, Kothari, and Shu, 2005). Overall, these studies find that the ex ante implied cost of capital has a low association with future realized returns and, therefore, the implied cost of capital estimates are poor measures of a firms expected equity returns. Easton and Monahan (2005) show that implied cost of capital estimates are negatively correlated with ex post observed stock returns and they suggest that the problem arises from the quality of analysts earnings forecasts used to calculate the implied cost of equity capital. There are other potential problems with implied cost of capital estimates that rely on analysts forecasts of future earnings. For example, analysts earnings forecasts may not capture the markets forecasts of future cash flow. While analysts earnings forecasts are widely followed, they also appear to be inherently biased. There is a long literature that suggests that analysts forecasts are biased at various horizons (see, for example, Richardson, Teoh and Wysocki, 2003 and Easton and Sommers, 2007). In general, analysts medium and long-horizon earnings forecasts tend to be too optimistic. Also, analysts tend to cover relatively few firms and available forecasts tend to be for near-term earnings such as earnings for the coming quarter or fiscal year. There are also apparent biases in which firms are covered by analysts, for
30

example, financially distressed firms are often not covered or are dropped by analysts (Deither, Malloy, and Scherbina, 2002). To address some of the limitations of the implied cost of capital approach, Hou, van Dijk, and Zhang (2009) outline a novel method to estimate a firms implied equity cost of capital. They build on Fama and French (2000, 2008), Hou and Robinson (2006), and Hou and van Dijk (2008) and apply a cross-sectional model to forecast the earnings of individual firms. In essence they apply fundamental analysis techniques to forecasts future income using a simplified version of model (3-G): Et[oit+1] = F{OICt , NOACt, NOACt-1, OTHERt} Their approach uses a cross-sectional model to capture across-firm variation in future profitability using publicly-available accounting (and other) information at the time of the forecast. Hou et al (2009) then use these earnings forecasts as inputs for a discounted residual income model to estimate implied cost of capital. An advantage of this forecasting methodology is that it does not rely upon analysts forecasts to generate cost of capital estimates. An interesting aspect of this approach is that is has foundations in the fundamental analysis literature and it fits well with our view that the common principle underlying this literature is forecasting. Following Easton and Monahan (2005), Hou et al. (2009) assess the reliability of their model-based implied cost of capital estimates by testing their correlation with future observed stock returns. Hou et al. (2009) show that the cost of capital estimates are significantly positively correlated with future stock returns. They also show that the greater reliability of their forecasting-model-based estimates of implied cost of capital arises from the improved earnings forecasts generated by their cross-sectional model. Therefore, there appears to be promise in using this type of forecasting
31

methodology for future research on implied cost of capital and other fundamentalsbased research. 4.3. Accruals Anomaly The accruals anomaly originally documented by Sloan (1996) suggests that firms with high (low) reported accruals in a fiscal period tend to have abnormally low (high) stock returns in subsequent periods. Accruals are non-cash accounting items which are added to operating cash flows to generate a firms current reported accounting income. Sloans original paper hypothesizes that investors naively fixate on bottom line income and they do not appear to understand that: (i) earnings is composed of both operating cash flows and (non-cash) accruals, and (ii) the cash flow and accrual components of earnings have different abilities to predict future earnings. In particular, innovations to accruals tend to reverse in future periods and investors do not appear to understand this time-series property when they develop their forecasts of future earnings and cash flows and therefore set current stock prices. Sloan (1996) defines accruals using changes in balance sheet items and measures total accruals (ACC) as changes in non-cash working capital minus depreciation expense scaled by average total assets: ACC (CA CASH) (CL STD TP) DEP (6)

where CA is the change in current assets (Compustat annual item 4), CASH is the change in cash or cash equivalents (Compustat annual item 1), CL is the change in current liabilities (Compustat annual item 5), STD is the change in debt included in current liabilities (Compustat annual item 34), TP is the change in income taxes

32

payable (Compustat annual item 71), and DEP is depreciation and amortization expense (Compustat annual item 14). Sloans main finding that firms with high (low) reported accruals tend to have abnormally low (high) future stock returns has inspired a vast body of follow-up work that attempts to understand the underlying causes of this anomaly. Sloans original paper hypothesizes that nave investor fixation on bottom line earnings and that investors do not understand the differential persistence of the cash flow and accrual components of earnings. In particular, innovations to accruals tend to reverse in future periods and investors do not appear to understand this time-series property when they develop their forecasts of future earnings and cash flows. Using our forecasting framework, we restate the accruals anomaly as follows: Investors attempt to forecast a firms operating performance using current reported earnings and changes in net operating assets to generate these forecasts. However, Within the original Sloan (1996) framework, equations (3-G) and (4-G) are used in reduced form where oit cfot and accrualst are the only components of OICt and NOACt used in the forecasting exercise. Therefore, equations (3-G) and (4-G) are reduced to: Et[oit+1] = F{cfot , accrualst } Et[rt+1] = H{cfot , accrualst } where cfot captures operating cash flows in period t and accruals captures a specific subset of change in net operating assets in period t. The evidence presented in Sloan (1996) suggests that, investors do not properly weight the components of oit (namely, cfot and accrualst) in generating their forecasts. Much of the follow-up work on Sloan

33

(1996) essentially captures refinements to the forecasting of future earnings, cash flows, risk and returns. 4.3.1. Possible (non-risk) explanations for the accruals anomaly The first category of papers examines the reasons for the accrual anomaly. Sloans original paper hypothesizes that nave investor fixation on bottom line earnings and that investors do not understand the differential persistence of the cash flow and accrual components of current earnings in helping forecast future earnings and cash flows. Recent examples of papers that directly evaluate whether this

hypothesis is empirically supported include Ali, Hwang, and Trombley (2000), Zach (2005), Kothari, Lutskina, and Nikolaev (2006), and Hirshleifer, Hou, Teoh, and Zhang (2004). The first three of this set of papers do not find support for the investor fixation hypothesis. More specifically, Ali, Hwang, and Trombley (2000) find that abnormal returns are not lower for that are followed by sophisticated investors who might better understand the properties of accruals (such firms include the largest firms, those with high analyst following, and those with high institutional ownership). They also document that the association between accruals and future stock returns is not a function of transaction costs, transaction volume, or stock price. Consequently, they conclude that the nave investor fixation hypothesis cannot explain the accrual anomaly. Employing different sets of analyses, Zach (2005) concludes similarly. Finding no evidence of accrual reversals or overreaction, he argues that investor fixation could not be the reason for the accrual anomaly. Kothari, Lutskina, and Nikolaev (2006) find that the agency theory of overvalued equity, not investors fixation on accruals, explains the accrual anomaly. They state that overvalued firms have incentives to remain overvalued, whereas undervalued firms have no incentives to prolong their undervaluation, which results in an asymmetric relation between the
34

accruals and past and future returns. Kothari et al. interpret analysts greater degree of optimism and the distortion of insider trading and investment financing in high accrual firms to be consistent with the agency theory of overvalued equity. Though the findings in the aforementioned papers do not support Sloans investor fixation hypothesis, Hirshleifer, Hou, Teoh, and Zhang (2004) document that, limited attention of investors who focus on accounting profitability without taking into consideration the other factors in forecasting future cash profitability, could explain the mispricing of net operating assets scaled by total assets, which is consistent with the investor fixation hypothesis. A second explanation for the accrual anomaly is offered by Xie (2001), which finds that the anomaly is attributable to the mispricing of discretionary accruals as a consequence of overestimating the persistence of the discretionary accruals. Within our forecasting framework, Xie (2001) further subdivides the components of the NOACt vector into finer components that have differential predictive ability for future earnings. Chan, Chan, Jegadeesh and Lakonishok (2006) essentially replicate Sloan (1996) and find that firms with earnings increases accompanied by high accruals have lower future stock returns. Based on a variety of additional analyses, they conclude that most of the evidence is consistent with accruals capturing the earnings management activities of the management, consistent with the findings in Xie (2001). Richardson, Sloan, Soliman and Tuna (2005) bring a different

perspective to the debate and argue that investors do not understand the lower persistence of less reliable accruals, which leads to incorrect investor forecasts of future earnings and cash flows and to their mispricing of current accounting realizations. Within our forecasting framework, Richardson et al (2005) use an extended decomposition of the noat to identify components (ie, a NOACt vector) that
35

exhibit high versus low reliability in predicting future operating income. After ranking the components of noat according to their reliability, they find that the magnitude of the accrual anomaly is greater for the less reliable accruals. This finding is also consistent with that of Xie (2001) as discretionary accruals are expected to be less reliable and therefore less persistent. Yet more papers examine other potential reasons for the lower persistence of accruals. On the one hand, Beneish and Vargus (2002) find that the lower persistence of accruals is consistent with earnings management. They document that the lower persistence of income increasing accruals in the presence of insider selling is partially attributable to earnings management. On the other hand, Fairfield, Whisenant, and Yohn (2003) argue that the lower persistence of accruals maybe due to the effect of growth on profitability as they document that accruals covary more with invested capital, the denominator used in the computation of profitability, than cash flow does. Again, within our framework they apply a version of equations (3-G) and (4-G) to capture the broader effect of change in net operating assets as a possible driver of the accruals anomaly. Bringing another perspective to the earnings persistence debate, Dechow and Ge (2006), show that earnings persistence is a function of both the sign and the magnitude of accruals. They find that accruals increase (decrease) the persistence of earnings compared to cash flows in high (low) accrual firms. Dechow and Ge

document that the lower persistence of earnings in low accrual firms is due to special items. Low accrual firms with special items have higher future returns than other low accrual firms consistent with investors not understanding that special items are transitory.

36

In summary, these studies provide useful examples of how the literature has evolved over the past decade. Overall, each paper has attempted to provide insights into the problems that investors have in using current accounting information to correctly forecast future earnings, cash flows and returns. The primary explanation for the negative relation between accruals and future stock returns (holding aside the issue of risk) is that capital market participants fail to correctly utilize accrual information in their forecasts of future earnings (and cash flows). 4.3.2. Is the accruals anomaly distinct from other anomalies? The answer to this question is not conclusive, but the majority of the evidence supports the view that accruals anomaly is distinct and is incremental to other previously-documented anomalies. There is a large list of papers that study this question with an aim to document whether the accruals anomaly is subsumed by other anomalies. Collins and Hribar (2000) document that the accrual anomaly is distinct from the post-earnings announcement drift anomaly documented by Bernard and Thomas (1989). While both anomalies appear to be distinct in generating future stock returns, they both have the basis of incorrect investor responses to current accounting information to generate forecasts of future earnings, cash flows and returns. Barth and Hutton (2004) show that the predictive ability of accruals for future returns is not subsumed by the predictive ability of analysts forecast revisions. In a similar line of research, Cheng and Thomas (2006) document that the accrual anomaly is distinct from the value-glamour anomaly. However, Fairfield, Whisenant, and Yohn (2003) argue that accruals anomaly is a special case of a more general growth anomaly (see also section 4.3). They find that both accruals and growth in long-term net operating assets (components of net operating assets) have similar negative associations with future return on assets, and that the market seems to overvalue them similarly.
37

Related, Zhangs (2007) findings also suggest that the accrual anomaly is attributable to investment / growth information contained in accruals measured as the covariation between accruals and employee growth, rather than investors misunderstanding of the implications of accruals for earnings persistence. He finds that for industries and firms where accruals and employee growth covary more, accruals have a stronger power in predicting future returns. Challenging the conclusion that growth is the sole explanation for the accruals anomaly, Richardson, Sloan, Soliman, and Tuna (2006) find that the temporary accounting distortions also play an important role in explaining the lower persistence of accruals in addition to growth-based explanations. Again, this debate revolves around the issue of which factors influence or distort how investors generate their forecasts of future earnings, cash flows and returns. This debate leads to our next subsection which deals with the relation between the accruals anomaly and the growth/investment anomaly, which is identified as a sub-category of research related to accounting accruals. 4.3.3. Relation between accruals anomaly and investment anomaly Over the past decade, there have been numerous studies investigating the association between a firms corporate asset investment and disinvestment actions and future stock returns. The findings suggest that corporate events associated with the expansion of a firms scale and its assets (i.e., acquisitions, public equity offerings, public debt offerings, and bank loan initiations) tend to be followed by periods of abnormally low long-run stock returns. On the other hand, corporate events associated with decreases in the scale of the firm and asset contraction (i.e., spinoffs, share repurchases, debt prepayments, and other payouts) tend to be followed by periods of

38

abnormally high long-run stock returns.13 In this section, we discuss the links between this literature and the accruals literature. In addition to these long-run event studies, other work documents a negative relationship between various forms of corporate investment and the cross-section of returns. For example, an increase in accruals, capital investment, and sales growth rates, and external financing tends to be negatively correlated with subsequent stock returns. Recent studies include Fairfield, Whisenant, and Yohn (2003), Richardson and Sloan (2003), Titman, Wei and Xie (2004), and Hirshleifer, Hou, Teoh, and Zhang (2004). More recent research (see, for example, Richardson, Sloan, Soliman, and Tuna, 2006, and Cooper, Gulen and Schill, 2008) presents additional evidence on the debate over whether growth is fairly priced in the cross-section of future stock returns by introducing and fine-tuning measures of firm growth. In addition, these studies attempt to understand the underlying sources of firm-level growth effects. The refined measures of firm growth are motivated by the observation that prior studies on the effects of growth on returns use components of a firms total investment or financing activities, and often ignore the larger picture of potential total asset growth effects of comprehensive firm investment and disinvestment. Cooper et al (2008) use a general measure of firm asset growth, the year-onyear percentage change in total assets and a panel of U.S. stock returns. They document a negative correlation between firm asset growth and subsequent firm

13

References include acquisitions (Asquith, 1983; Agrawal Jaffe, and Mandelker, 1992; Loughran and Vijh, 1997, Rau and Vermaelen, 1998), public equity offerings (Ibbotson, 1975; Loughran and Ritter, 1995), public debt offerings (Spiess and Affleck-Graves, 1999); bank loan initiations (Billet, Flannery, and Garfinkel, 2006), spinoffs (Cusatis, Miles, and Woolridge, 1993; McConnell and Ovtchinnikov, 2004), share repurchases (Lakonishok and Vermaelen, 1990; Ikenberry, Lakonishok, and Vermaelen, 1995), debt prepayments (Affleck-Graves and Miller, 2003), and dividend initiations (Michaely, Thaler, and Womack, 1995).

39

abnormal returns. They find that asset growth remains significant in explaining future stock returns that include book-to-market ratios, firm capitalization, short- and longhorizon lagged returns, and other growth measures (including growth in sales from Lakonishok, Shleifer, and Vishny (1994), growth in capital investment from Titman, Wei and Xie (2004), accruals from Sloan (1996), and a cumulative accruals measure (net operating assets) from Hirshleifer, Hou, Teoh, and Zhang (2004).14 In a related line of research that focuses on the association between investments and future stock returns, Titman, Wei, and Xie (2004) find that companies that increase their investments the most have significantly lower returns than their benchmark over the next five years. Anderson and Garcia-Feijoo (2006) document similar findings, namely the future returns are significantly lower for firms that have accelerated their investments. Titman et al. (2004) interpret this finding as consistent with investors underreaction to increased investments for empire building purposes, after documenting that the abnormal returns are concentrated around earnings announcements. Dechow, Richardson, and Sloan (2008) document that the accruals anomaly subsumes the external financing anomaly and offer an alternative interpretation to the investment anomalies literature. Dechow et al. (2008) document that the accruals anomaly subsumes the external financing anomaly and they find that it is the use of the external financing proceeds that predicts future returns, rather than raising or distributing financing as suggested by the earlier studies and reviewed by Ritter (2003). Dechow et al. (2008) also find that firms with high accruals have lower

14

Richardson, Sloan, and Tuna (2006) show that the cumulative accrual measure is simply an algebraic transformation of the change in net operating assets documented in Richardson et al. (2005). Thus, the claim that the cumulative accruals measure captures past changes in net operating assets is misleading.

40

earnings persistence and lower future stock returns even if they use internally generated funds instead of raising external finances, unlike firms that expend the externally raised cash or swap equity for debt, which are not as likely to be mispriced. In summary, the accrual anomaly literature has evolved over time to make clear explicit links with other anomaly papers, most noticeably the financing and investing anomalies. Recent research suggests that the accrual anomaly actually subsumes these related anomalies. Overall, it appears the investors have difficulty interpreting the performance of firms that have significant changes in net operating assets (i.e. accruals) and then using this information to generate forecasts of future earnings, cash flows and returns. 4.3.4. Why is the accruals anomaly not arbitraged away? There are three potential answers to this question: (i) because there are limits to arbitrage and large transaction costs that prevent this, (ii) because accruals is a risk factor, and (iii) market has not learnt yet and investors continue mispricing the accruals. Although these answers are discussed in the context of accruals anomaly, they also apply to other accounting anomalies. 4.3.4.1 Limits to arbitrage and transaction costs: Mashruwala, Rajgopal, and Shevlin (2006) find that the accruals anomaly is concentrated in firms with high idiosyncratic return volatility, low price, and low volume, suggesting that transaction costs may provide an obstacle for investors to trade away the accrual anomaly. Similarly, Lev and Nissim (2006) find that the extreme accrual firms are small, risky, and have low profitability and hence do not attract the attention of large institutional investors. Although some active institutional

41

investors trade based on the accrual anomaly, the magnitude of their trades are too small to arbitrage away the anomaly. As for individual investors, given the

characteristics of these firms, transaction costs are too high. Consistent with the findings in Lev and Nissim (2006), Ali, Chen, Yao and Yu (2008) find that even the mutual funds that have the largest exposure to low accrual stocks have limited exposure, suggesting that few actively managed funds trade on the accrual anomaly. The ones that do are smaller, less diversified, have higher fund flow volatility and higher fund return volatility and earn 2.83% abnormal returns annually. This

evidence suggest that that accrual anomaly is not arbitraged away because it is too costly for individual and institutional investors to trade based on it. In fact, Collins, Gong, and Hribar (2003) find that firms with high institutional ownership and that exceed a minimum level of holdings by active institutional investors have accruals that are less mispriced. 4.3.4.2 Are observed future stock returns due to risk or mispricing of information in accruals? This question is at the crux of the debate about any stock market anomaly, and the accruals anomaly is no exception. As expected, the academic literature is quite divided on this issue. On the one hand, using an intertemporal CAPM based fourfactor model, Khan (2008) finds that the cross-sectional variation in the stock returns of extreme accrual portfolios can be largely explained by risk. On the other hand, Hirshleifer, Hou, and Teoh (2006) conclude that their findings are more consistent with a behavioral explanation than a risk-based explanation for the accruals anomaly, as their analysis documents that it is the accrual characteristic, not the accrual factor loading, that predicts future stock returns. In section 5 of this review paper, we revisit

42

the risk vs. mispricing question, and propose an innovative way to tackle it using a more comprehensive treatment of risk. 4.3.5. Robustness and generalizability of the accruals anomaly Although a vast majority of the papers that examine the accruals anomaly find that the original findings in Sloan (1996) are robust in different samples, several recent papers more directly test the implications of potential methodological concerns, sample characteristics, or the choice of benchmark returns on the accruals anomaly. For example, Kraft, Leone, and Wasley (2007) find that the accruals anomaly is not robust to the addition of omitted variables to the Mishkin test and that such omitted variables lead to incorrect inferences about the pricing of accruals. Zach (2005) finds that excluding observations related to mergers and acquisitions, divestitures, NASDAQ-listed firms, and the use of size and book-to-market adjusted returns instead of just size-adjusted returns reduce the abnormal returns to the accrual anomaly, but the majority of the returns are robust. A few papers examine whether accruals anomaly is globally generalizable. The findings from these studies indicate that accruals anomaly is somewhat mixed. LaFond (2005), and Pincus, Rajgopal, and Venkatachalam (2007), document that the accruals anomaly exists outside the U.S. Pincus et al. (2007) find that it is more likely to occur in countries where accrual accounting is used more, when there is lower shareholder concentration, lower shareholder protection, and if the legal system is of common-law origin. In contrast, Leippold and Lohre (2008) document that the

global results are sensitive to methodological choices. Finally, Hirshleifer, Hou, and Teoh (2008) examine whether the accruals anomaly extends to aggregate returns and find that aggregate accruals are positively
43

associated with future market returns. However, consistent with aggregate accruals containing information about changes in discount rates (or firms managing earnings due to aggregate undervaluation), Hirshleifer et al. find a negative contemporaneous association between changes in aggregate accruals and market returns. Overall, the body of literature that follows Sloan finds that the accrual anomaly is robust in various samples, and that it is mainly attributable to investors inability to incorporate the implications of discretion in accruals for the persistence of earnings in their forecasts of future earnings. 4.4. Underreaction to accounting information, in particular the post-earnings announcement drift (PEAD) anomaly The past 10 years has seen expanded research on the first documented major accounting-based market anomaly known as post-earnings announcement drift (PEAD) (see, for example, Bernard and Thomas 1989). The main feature of the PEAD anomaly is that investors appear to underreact to earnings news and a firms stock price drifts in the direction of the earnings news after an earnings announcement. Using our forecasting framework, we the PEAD anomaly as follows: Investors attempt to forecast a firms future free cash flows using innovations to current reported earnings to generate these forecasts. Et[fcft+1]= f(oit) The anomalous returns are generated because investors do not quickly impound in prices the information contained in oit. Recent studies attempt to explain why the anomaly occurs and why it persists. In addition, the literature has expanded to consider underreaction to other corporate information and the relation to momentum

44

in stock returns. We focus our review in this section on the PEAD literature, but readers should note that our discussion applies to other research studying the underreaction to other sources of accounting information (e.g. option expense disclosures, pension footnotes, etc.). We revisit this topic in section 6. Again, our organizing framework emphasizes the forecasting activities of equity investors. 4.2.1. Underreaction to earnings information: What are the reasons for PEAD? Trading by unsophisticated investors has been proposed as a reason for the post-earnings announcement drift (PEAD). These unsophisticated investors do not appear to realize the implications of current earnings for forecasts of future earnings and stock returns. For example, Bartov, Radhakrishnan, and Krisy (2000) find that institutional ownership is negatively associated with the magnitude of the abnormal returns after earnings announcements and that transaction cost proxies do not have any explanatory power incremental to institutional ownership. Bartov et al. conclude that individual investors trading activities drive the anomaly. Consistent with Bartov et al., Battalio and Mendenhall (2005) find that investors executing small trades seem to respond to a less sophisticated signal that does not fully impound the implication of current earnings changes for future earnings, suggesting that small investors underlie the PEAD. Similarly, Shantikumar (2004) also finds that small traders are more likely to underreact to earnings surprises relative to larger traders, although larger traders also under-react. This evidence suggests that small (and likely less informed) traders are a driver of the PEAD phenomenon. On the other hand, Hirshleifer, Myers, Myers, and Teoh (2008) present contrary findings that the returns to the PEAD strategy cannot be explained by the trading activity of the individual investors.

45

Related to investor un-sophistication, limited investor attention on the implication of current earnings for forecasts of future earnings is also suggested as an explanation to PEAD by Hirshleifer and Teoh (2006). Consistent with a broader definition of limited investor attention, DellaVigna and Pollet (2006) find that earnings announcements that take place on Fridays have more drift than earnings announcements that occur on other weekdays. Related, Liang (2003) finds that

investors overconfidence about their private information and the reliability of earnings result in the underreaction to current earnings innovations and a slow revision of investors forecasts of future performance, which and in turn leads to PEAD. An example of the possible factors that distort investors forecasts is highlighted in Chordia and Shivakumar (2005). They document evidence that inflation illusion hypothesis, i.e. that investors do not incorporate the effect of inflation in their forecasts of future earnings growth rates, provides a third explanation for the PEAD. They find that the sensitivity of earnings growth to inflation

monotonically increases across earnings surprise portfolios, and controlling for the predictive ability of inflation reduces the predictive ability of earnings surprises for future returns. A fourth explanation for the PEAD is transaction costs and limits to arbitrage. Ng, Rusticus, and Verdi (2008) find that transaction costs provide an impediment for informed investors forecasts of future performance to get quickly impounded into price and that returns to the PEAD strategy are significantly smaller when one controls for transaction costs. Mendenhall (2004) documents a cross-sectional

variation in post-earnings announcement drift based on arbitrage risk, and argues that his findings are consistent with an underreaction explanation to PEAD. In a similar
46

vein, Reed (2007) finds that firms that are more short-sale constrained have less of their long-run price reactions to earnings announcements occur on the announcement date. In section 5, we revisit the issue of transaction costs in our own empirical analysis. A fifth explanation documents cross-sectional variation in the strength of the PEAD anomaly based on the relative transparency of earnings-related information. An example is Kimbrough (2005), who finds that firms that initiate conference calls also experience less underreaction to their earnings surprises, resulting in a smaller PEAD. A similar example to this explanation in the accruals anomaly context is Levi (2008). Unlike the papers above, which suggest investors irrational underreaction causes PEAD, Francis, Lafond, Olsson, and Schipper (2007) (FLOS hereafter) argue that rational responses to information uncertainty by Bayesian investors resulting in the PEAD. Relying on the rational structural uncertainty theory by Brav and Heaton (2002), and using Dechow and Dichev (2002)s earnings quality measure as a proxy for information uncertainty, FLOS document empirical evidence that rational learning could also explain underreaction to earnings news. They find a lower sensitivity of announcement period returns to unexpected earnings of firms that have higher information uncertainty. FLOS also find that high information uncertainty firms have larger abnormal returns than low information uncertainty firms, and that the information uncertainty is higher in the extreme unexpected earnings portfolios relative to the non-extreme ones. However, as both Brav and Heaton (2002) and FLOS state, it is difficult to empirically identify whether irrational behavioral or rational learning theories explain the underreaction to news as both theories have similar predictions. Regardless of the explanation, there is a slow adjustment by
47

investors to their forecasts of future performance based on the news in current earnings. 4.4.2. Why does underreaction to accounting information persist? Insufficient trading by sophisticated investors and limits to arbitrage are the two empirically supported explanations. Related to investor sophistication, several papers study whether a subset of institutional investors trade based on and profit from the PEAD. For instance, Ke and Ramalingegowda (2005) find that transient

institutional investors generate large returns from trading on the PEAD and that their trading increases the speed with which prices impound the earnings information. However, they also document that these investors trade less in companies with higher transaction costs, which could explain why PEAD persists. Ali, Chen, Yao, and Yu (2007) present similar findings showing that actively managed U.S. equity mutual funds use PEAD as a trading strategy, and that those that trade on this strategy outperform those that do not. Consistent with the findings in Ke and

Ramalingegowda (2005), Ali et al. also find that there may be limits to arbitrage the PEAD, as funds that trade based on it have greater return and fund flow volatility and are less diversified, which may contribute to the persistence of PEAD. 4.4.3 Alternative measures of PEAD Some of the literature that follows Bernard and Thomas (1989) propose ways to fine-tune the anomaly. This set of papers generally use an earnings surprise measure different from Bernard and Thomas (1989) original seasonal random-walk based earnings surprise, or capitalize on some properties of earnings that are known to be associated with stock prices. For instance, Doyle, Lundholm, and Soliman (2006) and Livnat and Mendenhall (2006) both document that returns to the post-earnings
48

announcement drift strategy are larger if the earnings surprise is computed based on analysts expectations. In our forecasting framework, Et[fcft+1]= f(oit, Othert) where Othert information captures analysts earnings expectations. Brandt, Kishore, Santa-Clara, and Venkatachalam (2008) show that the return drift is stronger using the stock price reaction around earnings announcements as proxy for earnings surprise. Narayanamoorthy (2006) finds that investors

underestimate the implications of accounting conservatism for future earnings and documents that the returns to PEAD strategy are greater if accounting conservatism is incorporated into the analysis. Relying on the findings that cash flows have bigger impact on stock prices than accruals, Shivakumar (2006) decomposes earnings surprises into unexpected accruals and unexpected cash flows, and documents that the cash flow component is a better predictor of future stock returns and that a trading strategy that uses the two components separately outperforms the one that is based on the total earnings surprise. 4.4.4. Robustness and generalizability of the PEAD anomaly Using a sample of UK firms, Liu, Strong and Xu (2003) find that postearnings announcement drift exists outside of the US as well. Griffin, Kelly, and Nardari (2007) document that returns to the PEAD strategy in emerging markets are similar to implementing the strategy in developed markets. 4.5. Pricing Multiples and Value Effect Much of the literature on active investment strategies has argued that value stocks with high values of various price-based ratios including book-to-market (B/M),
49

earnings-to-price (E/P), or cash flow-to-price (C/P) outperform glamour stocks with low values of these ratios. Within our forecasting and valuation framework, pricing multiples anomalies essentially compare observed market prices with a single observable accounting attribute (such as current earnings) that proxy for future forecasted free cash flows and possibly the risk of these cash flows. An investor must then determine whether the single fundamental variable captures the essence of a complete valuation of the firm (i.e., all future expected free cash flows and the risk of these cash flows) and whether the observed market price is consistent with such a valuation. In an influential paper, Lakonishok, Shleifer and Vishny (1994) attribute the superior (inferior) performance of value (glamour) stocks to errors in expectations on the part of investors about forecasts of future growth prospects of these firms. Lakonishok et al (1994) posit and find that value (growth) stocks are underpriced (overpriced) because investors appear to extrapolate poor past growth rates into their forecasts of the future and hence, investors are pessimistic about the future prospects of these companies. Therefore, as actual growth rates tend to mean-revert in the future, investors are negatively (positively) surprised by the performance of companies that were categorized as glamour (value) companies. Dechow and Sloan (1997) document that it is not the extrapolation of past growth rates that causes the glamour effect, rather it investors nave reliance on analysts long-term growth forecasts. The primary debate over the 15 years about the book-to-market (value) effect is whether it is generated from risk or mispricing. Going back to Fama and French (1992) and Lakonishok, Shleifer and Vishny (1994), the literature consistently shows that the book-to-market ratio of a firm is strongly positively correlated with future
50

stock performance. The Fama and French (1992) risk explanation argues that high B/M stocks earn excess returns compared to most firms because of their greater risk, as many high B/M firms are in financial distress. More recent research by Vassalou and Xing (2004) appears to confirm that book-to-market risk essentially proxies for default risk in high B/M firms. On the other hand, Griffin and Lemmon (2002) show that firms with high distress risk exhibit the largest return reversals around earnings announcements, inconsistent with a risk based explanation. This explanation is less satisfying for low B/M firms, as there are few ex-ante reasons to believe that these firms, largely growth firms, are less risky than the entire population of firms. Ali et al (2003) who show that the book-to-market effect is greater for stocks with higher idiosyncratic return volatility, higher transaction costs and lower investor sophistication which supports the a mispricing story. Furthermore, the evidence in Penman, Richardson, and Tuna (2006) conflicts the risk explanation for the book-tomarket effect as they find that leverage, which falls out as a component of book-tomarket in Penman et al.s decomposition, commands a negative risk premium. 4.5.1 Relation between value and accruals anomalies Desai, Rajgopal and Venkatachalam (2005) posit that the value and accruals anomalies are related because both anomalies represent overreactions to past accounting data. In the value-glamour anomaly, investors extrapolate past growth in sales, earnings and cash flows, and realize subsequently (mostly at the time of future earnings announcements) that such growth is not sustainable because growth rates mean-revert. In the case of the accruals anomaly, investors extrapolate past accruals into the future and are surprised when earnings announced subsequently are lower or higher due to reversals in accruals. Thus, both anomalies relate to errors in expectations about future earnings. Furthermore, certain proxies for the value51

glamour effect and accruals are closely linked. For example, sales growth, one of the proxies for value-glamour, is positively correlated with accruals. 4.5.2 Value effect and financial statement analysis Financial statement analysis attempts to identify ex-post winners and losers on the basis of information in the financial statements that are not completely or perfectly impounded in prices. Ex-ante, it is unclear whether financial statement analysis will be effective for low B/M firms, even if they are mispriced, for the following reasons. First, low B/M firms tend to be growth stocks that attract the attention of sophisticated market intermediaries such as analysts and institutional investors. Second, such firms are likely to have many sources of disclosure other than financial statements. Third, the rapid growth in many low B/M firms potentially makes current fundamentals less important than other non-financial measures. Counterbalancing this is the fact that many of these stocks may be overvalued in departure from their fundamentals because of the hype or excitement surrounding their recent strong stock market performance. Further, while traditional fundamental analysis may have limited applicability for growth firms, other information from the financial statements can be potentially useful. Researchers have shown that the stock market tends to naively extrapolate current and forecasted fundamentals of growth stocks (e.g. Dechow and Sloan (1997)), or ignore the implications of conservative accounting for future earnings (e.g. Penman and Zhang (2002)). Recent research examines whether applying financial statement analysis can help investors earn excess returns on a broad sample of growth, or low book-to-market (B/M) firms. An example of this line of research is Mohanram (2005). Mohanram uses financial statement information to create signals relating to nave extrapolation and conservatism and augments traditional fundamental analysis

52

of earnings and cash flow profitability. He tests the ability of these growth oriented fundamentals to identify winners and losers in terms of ex-post stock returns. The results indicate that financial statement analysis, appropriately tailored for growth firms, is successful in differentiating between ex-post winners and losers. A strategy of buying firms with the strongest growth fundamentals and selling or shorting the weakest firms earns very significant abnormal returns. The results are robust across a variety of partitions including size, analyst following and exchange listing and also hold when recent IPO firms are excluded. Piotroski (2000) applies a similar approach on the high B/M group. Guay (2000), in his discussion of Piotroski (2000), notes the importance of clearly articulating the reason that the accounting variables are not efficiently priced by the market. A general criticism of financial statements analysis that aims to predict future stock returns is the large degrees of freedom available to the researcher. There is considerable risk of in-sample data fitting. This can be mitigated by having clear priors and emphasizing out-of-sample prediction.

5. Benchmark model for evaluating anomalies using a more-encompassing treatment of risk and transaction costs: Accruals case study In this section, we suggest several new empirical techniques for evaluating accounting anomalies derived from our framework based on forecasting activity. Our proposed techniques are motivated by two main observations: (i) our survey of academics and practitioners highlights important differences in how these two groups view forecasting activity and the treatment and forecasting of future expected risks and transactions costs, and (ii) there is very little evidence on whether academic research helps improve real-world forecasting efficiency which could lead to

53

increased market efficiency (i.e., do investors use academic research to learn how to forecast better?). Specifically, our survey of academics and practitioners suggests that: (i) the academic approach to assessing risk is based on an ex post analysis of returns, with very little attempt to forecast risk, (ii) a firms industry affiliation is completely missing from most academic research on expected risk, and (iii) the typical academic paper fails to capitalize on actual trade data to capture current and future forecasted transactions costs that affect the implementability of a trading strategy based on an anomaly. Therefore, we undertake a case study of the influential accruals anomaly to demonstrate a more comprehensive measurement of forecasted risk and transactions costs. Our aim is to articulate the importance of ex ante considerations of risk and trading costs in building a portfolio. Specifically, we are interested in examining whether time-varying, ex ante, considerations of risk and trading costs change the well known negative relation between accruals and future returns. While we have chosen to focus on accruals, the approach outlined in this section can be extended to any accounting or non-accounting signal claimed to have an anomalous relation with future stock returns. This section proceeds in distinct sub-sections. First, we describe our accrual variables and the data to be used in the following empirical analysis. Second, we describe the standard mean-variant approach to portfolio construction, with particular emphasis on using a fundamental multiple factor model to forecast risk at the portfolio level (e.g., Ross, 1976). Third, we compare and contrast this ex ante

approach to incorporating risk with the traditional ex post analysis of returns. Fourth,

54

we describe a standard model of transaction costs (direct and indirect) as used in practice to forecast transaction costs. Fifth, we document that the negative relation between measures of accruals and future returns is robust to a careful consideration of risk and transaction costs for the period 1973-2000, but it has attenuated substantially in the last decade. We attribute this affect to an adaptively rational/efficient market. Sixth, we provide evidence on the performance difference between a simple equal or value weighted zero-cost investment portfolio and a full characteristic portfolio resulting from optimally combining forecasts of expected returns, risk and trading costs (i.e., does all the additional effort help?). Finally, we conclude with some philosophical points on the risk versus mis-pricing debate. 5.1 Measures of accruals We construct our sample from U.S. firms during the 1973 to 2008 period and require sufficient data from Compustat to compute two measures of accruals: WC and NOA. These correspond to the traditional Sloan (1996) measure of accruals which is the change in non-cash working capital (WC) and the broader measure of accruals (NOA) first described in Richardson, Sloan, Soliman, and Tuna (2005) (RSST hereafter). Consistent with RSST (2005) we measure these items using

balance sheet data so we can extend our sample period back further in time. The precise definitions of these two accrual measures are as follows: NOA is defined as the change in non-cash assets less the change in non-debt liabilities. Noncash assets is calculated as total assets (Compustat annual data item 6, quarterly data item 44) less cash and short-term investments (Compustat annual data item 1, quarterly data item 36). Non-debt liabilities is calculated as total liabilities

(Compustat annual data item 181, quarterly data item 54) less total debt (Compustat

55

annual data item 9 plus Compustat annual data item 34, quarterly data item 51 plus quarterly data item 45). WC is defined as the change in non-cash current assets less the change in non-debt current liabilities. Non-cash current assets is calculated as total current assets (Compustat annual data item 4, quarterly data item 40) less cash and short-term investments (Compustat annual data item 1, quarterly data item 36). Non-debt current liabilities is calculated as total current liabilities (Compustat annual data item 5, quarterly data item 49) less short term debt (Compustat data item 34, quarterly data item 45). We require total assets and total liabilities to be non-negative. Other variables are set equal to zero if they are missing. We deflate each of these variables by average total assets, where assets are measured using Compustat data item 6. Consistent with previous research, we winsorize each deflated component of earnings at +1 and 1 in order to eliminate the influence of extreme outliers. To generate as realistic results as possible, we compute trailing twelve month values for NOA. We wait a full three months before assuming that this data is available to the market. For example, if we have a December 31 year end firm then the value for NOA in July of year t+1 will reflect the four quarters up to the end of March year t. Our stock return tests use data from the CRSP monthly files, inclusive of dividends and other distributions. We use all data available to us from 1973-2008, the limitation from 1973 is due to use of the common factor risk model described in section 5.2. To restrict our attention to a subset of securities that would typically be traded we limit each cross section to the largest 1000 securities based on market capitalization.

56

5.2 Forecasting risk What is risk? We discuss the forecasting of risk within the standard meanvariant framework (see e.g., Grinold and Kahn, 2000). The investor is concerned about comovement in the securities that are held in the portfolio. This is equivalent to stating that the investor faces a quadratic utility function and is interested in maximizing expected returns subject to minimizing comovement risk of this portfolio. We can represent this optimization problem as follows for a set of N assets: max Th hTXFXTh - hTh where is a (Nx1) vector of expected returns, h is the (Nx1) vector of portfolio holdings, X is a (NxK) matrix of exposures to common factors, F is a (KxK) matrix of variances and covariances of common factor returns, is a diagonal (NxN) matrix with the diagonal terms representing the variances of specific returns. is the aversion to risk. In this framework risk is F and . How do we measure risk? There is a long tradition in financial economics suggesting that assets with similar characteristics will behave in similar ways. Ross (1976) develops a linear asset pricing framework where security returns are related to the expected returns associated with a set of underlying systematic factors. These multiple common factor models are widely used to understand the risk associated with a given set of securities. The empirical challenge with this arbitrage pricing theory (APT) is that you need to specify the underlying systematic factors and a given securities exposure to them. Fortunately, there are many who have endeavored to do this and we can use their common factor models for our purposes here.

57

In the context of the standard optimization problem described above the investor needs to make a forecast of the risk they are likely to face from holding a given portfolio. To make things tractable, rather than estimate a complete NxN asset level variance-covariance matrix, the dimensionality of the problem is reduced by projecting total returns onto common factors, thereby decomposing total returns into a common factor component and a residual (or specific) return component. We allow common factor returns to covary in this framework, but not the specific returns. A standard way to specify the common factors is via characteristic exposures (alternative approaches include statistically motivated principal component

analysis).15 The approach is relatively straight forward: 1. Specify the set of common factors. This typically includes a set of

characteristics that are cross-sectionally standardized to allow direct comparison across factors (e.g., leverage, size, earnings yield, growth, value, momentum, and various macro sensitivities), and fixed effects capturing industry membership. 2. Run periodic (e.g., monthly) cross sectional regressions where total returns are projected onto the common factors specified in 1. 3. Extract the regression coefficients from 2. common factor returns. 4. Compute all of the pair-wise correlations and volatilities of the common factor returns. These are estimated over rolling windows (e.g., using the These are your estimated

15

Companies like BARRA and Axioma provide this service commercially.

58

last 60 months with a specified half life). This generates the necessary inputs to fill out the F matrix defined above. 5. The residuals from the regressions described in 2 are the specific returns. Using time series variation in the specific returns you compute the specific risk (e.g., a rolling 24 month standard deviation of the regression residuals). This generates the diagonal elements of the matrix defined above. This exercise is repeated every period (e.g., month) to obtain a set of exposures, X, and a forecast of risk (F and ). It is very important to note that this view on risk is based on recent realized stock level returns. It has a built-in timevarying component. Specifically, the correlations and volatilities are not static But note that they are

through time, i.e. they change as circumstances dictate.

changing only to the extent that realized equity returns respond to these changing state variables: this approach is not an explicit state dependent forecasting model. More sophisticated versions can be built that incorporate forward looking information (such as implied equity volatility or credit spreads) as well as explicit conditioning on latent macro state variables. For our analysis, we are use a plain vanilla version of the common factor risk model as is typically used in portfolio management. We use the BARRA USE3 risk model (only for the 1973-2008 period, due to data limitations on the common factor risk estimation). For a full description of the BARRA risk model estimation approach, as well as the included set of common factors please refer to BARRA (1998). How can we assess the quality of this approach to forecasting risk? Fortunately, there is a well established literature (see e.g., Connor 2000). A necessary

59

condition for including a common factor is that it helps explain cross-sectional variation in returns. But a sufficient condition for including a common factor is that it improves the ability of the set of common factors to forecast risk. A standard

measure for assessing the quality of common factor risk models is the bias test. This test evaluates the ability of the model to forecast the active risk for a portfolio over a pre-specified period. The relevant measure (bias statistic) is the standard deviation of the ratio of the realized active portfolio return to the forecast active portfolio risk. In the bias test the null hypothesis is that the active risk forecasts are unbiased estimates of the deviation of the active returns of the test portfolios. The expected value of the bias statistic is equal to one when the null hypothesis is true. If the bias statistic is greater (less) than one this indicates that the active risk has been under (over) estimated. Standard tests can be applied to assess whether the observed bias statistic deviates from one. The risk model we use (BARRA USE3) has been tested

extensively with this bias statistic. The final set of included common factors is decided based on the relative improvement in the bias statistic from the included variables. For the sake of brevity we do not go into the detail of constructing a portfolio risk forecast. The interested reader can pursue extensive material produced by vendors such as MSCI/BARRA, Axioma Inc., and Northfield Investment Services Inc. 5.3 Contrasting ex ante and ex post approaches to evaluate risk The approach described in section 5.2 is to forecast risk. These forecasts are used to build a portfolio that explicitly penalizes stocks that produce a riskier portfolio. It is useful to contrast this forecasting approach with the typical analysis of risk in the academic literature, which is very much an ex post assessment. This is not

60

to say that ex post assessment is irrelevant. Ex post analysis is a standard method to assess the quality of investment decisions. The standard ex post analyses are (i) cross-sectional characteristic regressions, and (ii) time series portfolio regressions. The cross-sectional characteristic regression seeks to explain variation in returns with nominated risk attributes. The approach is to include a set of candidate risk variables as independent variables as well as your nominated anomaly variable (all of the variables are measured at the start of the return measure which is the dependent variable). The relevant test is whether you find a statistically significant relation on your anomaly variable after including the candidate risk measures. The time series regression seeks to explain variation in a zero-cost portfolio formed on your anomaly variables with portfolio returns of known risk attributes, such as market return, size, momentum, value and liquidity factor mimicking portfolios. The relevant test is whether your zero-cost portfolio has a positive average return after controlling for comovement with the factor mimicking portfolios. The limitations of the prevailing ex post approaches are (i) that they are time invariant, and (ii) the existing approach (e.g., Fama-French) considers only a very narrow set of known risk attributes. The time series regression approach described above is time invariant by construction: you have very few degrees of freedom to capture time variation in risk. By forecasting correlations and volatilities of common factor returns every month you allow comovement risk to update continually, which is important as you allow common factors to impact risk differentially through time. Perhaps the more troubling aspect of the prevailing academic approaches to assessing risk is that they completely ignore the importance of industries and sectors. Industries explain a significant fraction of the cross-sectional variation in equity returns and
61

substantially more than the standard characteristics that are included in the typical academic risk models.16 It is also useful to note that a valid criticism of the forecasting approach outlined in section 5.2 is that it does not include the accrual measure. Specifically, if you are interested in whether a given attribute generates a return in excess of its risk you would include that attribute in the risk model estimation. The USE3 BARRA risk model described above contains 13 characteristics and 55 industry classifications. The issue at hand is whether adding an accrual measure as the 69th element in the risk model would improve the forecasting quality of that model. For that to be true, it would need to be the case that the return associated with the accrual common factor exhibited substantial positive comovement with other common factors included in the risk model for the optimal portfolio to tilt away from the accrual characteristic. This would be one additional variance and 68 covariances relative to a total of over 2,300 variance and covariance terms. For the ease of empirical analysis the reported results in section 5.5 are based on the standard BARRA risk model, but it is worth noting that adding a standard accrual measure does not materially improve the bias statistic. It would be desirable for future research to utilize extended common factor risk models like those provided by BARRA/MSCI, Northfield Information Services Inc., or Axioma Inc. These models allow for (i) time variation in the estimation of risk, and (ii) are flexible enough for the proposed characteristic to be added into the risk estimation process. This will allow stronger inferences to be made about anomalous stock return patterns.

16

Specifically for the period 1973-2008 the importance of industry membership (as captured by 55 industries) in explaining cross sectional variation in equity returns is 1.5 times relative to a set of 13 standardized characteristics. Thus, the majority of the variation in returns are missing from the extant academic analyses.

62

5.4 Forecasting transaction costs Transaction costs are obviously an important factor to help determine the implementability of an anomaly. The prevailing treatment of transaction costs in the academic anomaly literature is typically very cursory and limited to partitioning samples based on measures of perceived trading costs (e.g., market capitalization, stock price, volumes, etc.). There are some exceptions, notably the papers examining arbitrage risk and liquidity costs (e.g., Pontiff 2006, Mashruwala, Rajgopal and Shevlin, 2006, and Ng, Rusticus and Verdi 2008). A closer inspection of these papers, however, reveals a round trip transaction cost in excess of several percentage points. These numbers are unreasonable. As noted in section 4 the magnitude of returns to certain anomalies are too large to be readily explained by transaction costs. The typical round trip transaction cost (inclusive of market impact for a large capitalization security trading no more than 5 percent of average daily volume) is about 20-25 basis points. A complete treatment of the forecasting of transaction costs is beyond the scope of our paper. We refer interested readers to an extensive discussion of trading costs in Madhavan (2000) and Rakhlin and Sofianos (2006). For our purposes we wish to communicate the key features of a transaction cost forecast and how this impacts the construction of a characteristic portfolio. Trading costs are typically described in terms of direct and indirect costs. Direct costs include commissions, taxes, and crossing the spread. These tend to be readily observable and are easy to incorporate into a forecasting framework. Indirect costs are the market impact costs from your trading activity moving prices. As an order is executed the price typically moves against you: buying (selling) activity is associated with increasing (decreasing) prices. These moves are costly as they subtract from your expected return. The two
63

most relevant factors that are able to explain this impact are volatility and trading volume. Impact is an increasing (decreasing) function of volatility (volume). The precise functional form and sensitivity to these factors is an empirical question. To calibrate such models requires access to large amounts of order and trade execution data. Unfortunately such data is typically proprietary and not generally available. In our analysis we use a piece-wise linear forecast, where the expected transaction cost (i.e., the cost associated with changing portfolio weight, h, from ht to ht+1) is increasing in forecast volatility and decreasing in the square root of forecast trading volume (e.g., Rakhlin and Sofianos, 2006). How does a transaction cost forecast impact a characteristic portfolio? If stock A and stock B have the same expected return, the same set of exposures to common factors and the same level of idiosyncratic risk, you prefer the stock that has lower forecasted trading costs. Consistent with our treatment of risk in section 5.2, we are incorporating expectations about trading costs. Thus, our characteristic portfolio is aware of the existence of both direct and indirect transaction costs. We want to emphasize that our analysis of trading costs is not ex post. We do not look at the actual costs from trading. This requires incorporating proprietary data on order flow and executed trades, which is not publicly available. Our forecasts of market impact can be incorrect, primarily because forecast trading volume is not realized. This creates at least two types of costs: (i) greater impact from trading a greater fraction of daily volume than you had forecast, and (ii) opportunity costs from not capturing your expected return from unfilled orders.

64

5.5 Putting it all together to build a portfolio Our aim is to use the forecasts of risk and transaction costs described above to re-assess the strength of the negative relation between accruals and future stock returns. This will allow us to make stronger inferences about (i) whether the relation is attributable to risk, and (ii) whether it is actually implementable. Specifically, we use the forecasts of risk and transaction costs to solve a more general objective function of expected return maximization subject to a risk penalty and a trading penalty: max Tht htTXFXTht - htTht g(ht-ht-1) The solution to this objective function can be estimated numerically. We solve this objective function every month, over the 1973-2008 period. The typical hypothesis in the accrual anomaly literature is that expected returns are a linear function of the respective accrual measure. Given this assumed linearity, we

standardize the accrual measure (NOA) at each point in time to ensure that the scale of the variable is comparable with that of the risk forecasts. It is not informative to maximize an objective function when the scale of expected returns is an order of magnitude different from the expected volatility of those returns. Our forecasts of risk and transaction costs follow from sections 5.2 and 5.4 respectively. The solution to the optimization problem is subject to additional constraints. The relevant constraints that we examined are: (i) the target risk of the final portfolio is 10% (annualized), (ii) the sum of the positions in the final portfolio is zero (fullyinvested), and (iii) the portfolio weights can be positive and negative but no position can be greater than 5% of the over-all portfolio. There is no restriction on the

65

leverage of the final portfolio. We define leverage to be the average of the sum of the long and the short positions respectively. Across the time period examined the

leverage of the characteristic portfolio incorporating expectations on transaction costs is about 3.7 on average with an inter-quartile range of 3.3-4.2. It is important to appreciate the impact of the 10% target risk constraint. The optimization problem described above is solved numerically to identify a vector of portfolio weights (i.e., the long and short positions of each security in the investment universe). In expectation, the annualized volatility of the returns to this vector of portfolio weights is 10%. In other words, if (i) our hypothesis that accruals is a source of expected returns our risk forecast is correct, and (ii) our risk model is good, then we should see returns to this portfolio of about 10% per year with an ex post volatility of around 10%. For our empirical analysis the portfolio has a realized annual

volatility of 9.7% suggesting that our risk model is good. We present the results of our analyses graphically (Figure 1) and with standard statistical tests (tables 1 and 2). Figure 1 reports the log of the cumulative returns to the NOA characteristic portfolio. The return to the portfolio for a given month is computed as the sum of the product of the active weight and the total return for each position at the start of that month. We use total returns as this is a fully invested long/short portfolio (i.e., the sum of the long positions equals the sum of the short positions in dollar space so we are looking at a zero cost portfolio). This, of course, ignores cross-sectional and time series variation in rebate rates associated with achieving the desired short positions. We capture differences in rebate rates as part of the transaction cost forecast model. The blue (red) line in Figure 1 shows the

cumulative return of the NOA characteristic portfolio without (with) transaction costs. The key inferences we make from this graph is as follows: (i) the negative
66

relation between measures of accruals and future stock returns is robust to a comprehensive ex ante analysis of risk and transaction costs, and (ii) the negative relation has attenuated substantially in recent years. Table 2 reports the returns to the NOA characteristic portfolio through time. We split the full sample into four sub-groups based on decade as well as comparing the pre 2000 period to the post 2000 period. Panel A (B) reports the analysis for the portfolio which ignores (includes) expected transaction costs. Sharpe ratios (the ratio of realized returns to realized volatility) are also lower (note though that there are not well defined test statistics to compare these ratios). Panels C (D) report statistical tests where we document some weak evidence of (i) a temporal decline in the returns to the NOA characteristic portfolio (as evidenced by the marginally significant negative coefficient on the TIME variable), and (ii) a quickening in this decay in more recent years (as evidenced by the marginally significant negative coefficient on the TIME2 variable). There are many reasons as to why the relation between accruals and future returns has attenuated over time. The explanation we find most appealing is

adaptive market efficiency (e.g., Grossman and Stiglitz 1980), where capital market participants learn about the relevance of information for security prices and prices adjust accordingly. Additional empirical analysis could strengthen this claim

including (i) tracking time series variation in buy and sell driven volume at the time financial statement information is released to the market (increases in the directionality of volume changes with changes in NOA is consistent with the market responding more quickly to this information), (ii) tracking aggregate capital flows associated with investors who are known to exploit these types of anomalies (returns to these anomalies should decrease as such flows increase), and (iii) exploiting cross
67

country variation in the strength of the relation between measures of accruals and future returns and the covariates described above. Greene, Hand and Soliman (2009) also note the attenuation of the relation between a measure of current accruals (e.g., the change in non-cash working capital construct in Sloan, 1996) and future stock returns. They find this attenuation to be coincident with an increase in capital

allocation to that strategy, a finding that they argue is consistent with adaptive market efficiency. As noted above there are additional analyses to strengthen this assertion, and we leave this more comprehensive analysis to future research. We also run a standard ex post analysis of the returns to the NOA characteristic portfolio. Specifically, we regress monthly returns of the NOA

characteristic portfolio on monthly market returns and monthly returns of characteristic portfolios of B/P, Momentum and Size. Table 3 reports the results. Panel A (B) contains the regression results for the case with (without) transaction costs. The first row of each panel reports the realized beta to the NOA characteristic portfolio. The realized beta is insignificantly different from zero over the 1973-2008 time period (marginally negative in the with transaction costs case). More

interesting is the analysis in the second row. This suggests that standard risk attributes of momentum, value and size are only able to explain between 3-5% of the monthly returns to the NOA characteristic portfolio. A comprehensive, ex ante, treatment of risk greatly reduces the influence of standard attributes to explain returns. Panel C of Table 3 contains our final set of empirical analyses. This is the typical Fama-Macbeth characteristic regression. Rather than include all of the

candidate risk measures as independent variables (i.e., the 13 characteristics and 55 industry fixed effects) we subtract the total effect of these variables from total returns to derive an excess return. We then regress future realizations of this excess return
68

on to values of NOA today. We find a very strong relation between NOA and excess returns (t-statistics are above 4 even 12 months ahead). In summary, we find that (i) the negative relation between accruals and future returns is robust to a more comprehensive, ex ante, measurement of risk and transaction costs, and (ii) there is time series variation in the strength of this negative consistent with an adaptively efficient market (e.g., Grossman and Stiglitz, 1980 and Lo, 2004). 5.6 Is all this extra effort worth it? Up to this point we have shown that the negative relation between accruals and future returns is robust to incorporating forecasts of risk and transaction costs. But we have only asserted that it is better to use an explicit forecasting model of risk and transaction costs to construct a portfolio. While it is true that risk forecasts and transaction cost forecasts are used in practice (this is prima facie evidence in support of the assertion that they matter), we now turn to the data to corroborate the assertion that they lead to better investment decisions. What are we comparing? The typical academic anomaly paper will sort each cross-section into deciles based on the variable of interest (e.g., accruals) and take a long (short) position in the extreme deciles. The equal or value weighted difference of these extremes is termed as a zero-cost investment strategy. In contrast, the approach described in section 5.5 above takes the sorting variable (accruals) and combines this with risk and transaction costs to derive a complete set of portfolio weights. We term this the characteristic portfolio for our discussion here. So we are comparing the zero-cost investment strategy with the characteristic portfolio.

69

There are multiple ways that we can document differences in these two approaches. First, we document correlations between the standardized NOA

measure and the set of portfolio weights at each point in time. It is important to remember that the NOA measure is our forecast of expected returns in section 5.5, so this correlation documents the impact from risk and transaction cost considerations at the portfolio level. Panels A (B) of Figure 2 report these correlations for the case with (without) transaction costs. The most striking result here is the high correlation between the series through time. This suggests that (i) incorporating a high dimension common factor risk model does not change the ranking of securities across these two approaches (the Pearson correlation averages about 0.8 in the case without transaction costs), and (ii) incorporating a forecast of transaction costs has a more significant impact on the ranking of securities across these two approaches (the rank correlation averages about 0.7 in the case including transaction costs). The second observation is not surprising, as prior research has found that securities with extreme values of accruals tend to be smaller, more volatile and less liquid (e.g., Mashruwala, Rajgopal and Shevlin, 2006). Panels C (D) extend this analysis by showing the average relation between the standardized NOA measure and the portfolio weights. For each cross section we form 100 groups based on the standardized NOA measure. We then compute the average NOA measure for that group as well as the average portfolio weight for that group. We then plot the averages of these 100 buckets through time. There is a striking linear relationship between the averaged standardized NOA measure and the averaged portfolio weights, suggesting that the impact of the risk and transaction cost forecasting does not have a large impact on ranking of securities across these two approaches.

70

Second, we look at the return differences between the zero-cost investment strategy and the characteristic portfolio. The primary purpose of the risk model forecast is to build a portfolio. Simply comparing the rank ordering of expected returns and portfolio weights will give an incomplete picture of the impact of using risk and transaction forecasts. Table 4 shows the ability of the standard Fama-French factors (MKT, HML and SMB) to explain four sets of NOA portfolio returns: (panel A) the characteristic portfolio returns with full risk model, (panel B) characteristic portfolio returns with full risk model and transaction cost forecast, (panel C) equal weighted returns of a zero-cost investment portfolio that is long (short) the bottom (top) 10% of dNOA firms, and (panel D) value weighted returns of a zero-cost investment portfolio that is long (short) the bottom (top) 10% of dNOA firms. There is a very noticeable difference in the volatility profiles of these sets of returns. In all four cases: (i) the ability of the standard Fama-French factors to explain the zero-cost investment portfolios and the characteristic portfolios is similar (i.e., both have very low R2), and (ii) both the zero-cost investment portfolios and the characteristic portfolios have statistically significant intercepts suggesting an 'anomalous' return relation. It is important to note that the statistical significance is much weaker for the zero-cost investment strategy approach. Annualized Sharpe ratios from the 430

months of data support this inference: the Sharpe ratio for the characteristic portfolios approach is about 1.3 and that for the zero-cost investment strategy is about 0.6.17 This difference is not surprising as the risk forecast is calibrated to do precisely this: build a portfolio that not only captures your expected return but also minimize risk at the portfolio level. Perhaps more importantly, the point is that the academic hurdle of
17

One approach to testing the relative attractiveness of these two series of portfolio returns is to treat them as asset returns and run standard tests for identifying efficient portfolio weights (e.g., BrittenJones, 1999). The optimal weight on the characteristic portfolio (zero-cost investment strategy) is 90.5% (9.5%) and these weights are statistically different at the 1% level (F-stat of 45.5).

71

abnormal performance is higher than that in practice because academics do not optimize (much) on risk. In contrast, in practice, the optimization approach chooses low risk stocks from among those with similar alphas. Academics often ignore this last step except in some approaches like weighted least squares regression, where weighting is on the basis of volatility. On the other hand, practitioners must weigh the benefits of the additional complexity from including forecasts of risk and transaction costs against the costs of such effort.18 The primary purpose of our empirical analysis was to see empirically if an anomalous return result (e.g., accruals) could be refuted by incorporating forecasts of risk and trading costs. We can clearly show that the reported negative relation between measures of accruals and future stock returns is robust to a comprehensive treatment of risk and transaction costs. However, we note that while the impact of more comprehensive treatments of risk and transaction costs do not appear to change the rank ordering of securities very much, they are effective in increasing the Sharpe ratio at the portfolio level. 5.7 Discussion and caveats Our goal in this section was to present new empirical analysis showing the impact of a more comprehensive treatment of forecasting risk and transaction costs. A primary motivation for this analysis was to help provide new empirical evidence on the risk versus mis-pricing debate that permeates a significant fraction of the academic literature devoted to anomalies. Specifically, the association between a given attribute and future stock returns is always subject to the two competing
18

Obviously, investors utilities functions would incorporate these costs and benefits. Moreover, Grinold and Kahn (2000) show how utility functions are monotonically related to statistics such as the Sharpe ratio (or information ratio).

72

explanations of risk and mispricing. Under the risk explanation the relation between the attribute and future stock returns is explained by a lower (higher) risk premium for firms with that attribute. Under the mispricing explanation the relation between the attribute and future stock returns is caused by capital market participants failing to incorporate the impact of the attribute for future prices (typically through changes in expectations about future free cash flows for the types of attributes that we discuss in this survey). Absent an agreed upon asset pricing model, it is difficult to distinguish between the two explanations. The comprehensive risk forecasting described above, we hope, convinces the reader that the negative relation between measures of accruals and future stock returns is not attributable to risk. Of course, our empirical analysis can still be criticized as an incomplete assessment of risk. To help with this residual criticism, it is worth noting the

importance of non-price based tests to help confirm the hypotheses as to why a given attribute (e.g., accruals) is associated with future stock returns. Prior academic

research has sought to corroborate the relation accruals and future stock returns by looking at the behavior of sell side analysts (e.g., Bradshaw, Richardson and Sloan, 2001). Assuming that sell-side analysts are a reasonable proxy for the markets expectations, then documenting a systematic relation between sell-side forecast errors and accruals adds support to the mispricing explanation. This research design has since been used elsewhere in the literature to help confirm a mispricing based explanation for 'anomalous' patterns between accounting attributes and future stock returns. We encourage future research to make this a standard diagnostic test. For those who are still unconvinced about the risk/mispricing debate and argue that measures of accruals reflect differences in discount rates or expected returns (e.g., Wu, Zhang, and Zhang 2009), then we have a natural experiment ahead of us: as time
73

moves forward will the negative relation between accruals and future stock returns reappear? If it does not, then the claim that accruals is a risk factor is not valid. If it does, then the claim that it is a risk factor may be valid. We would need to see more work to understand what it is about the macro-environment that allowed this risk factor to generate a premium again. 6. Suggestions for future research In this section we bring together the streams of research discussed in the prior section, the feedback from the academic and practitioner questionnaire, and our own insights to make suggestions for future research on the use of accounting information in forecasting both future firm fundamentals and future returns. To make this section tractable, we have deliberately chosen to focus on the use of accounting information for security valuation purposes. To justify this focus it is important to keep in mind that one of the primary objectives of providing general purpose financial reports is to communicate relevant and reliable information about the reporting entity to users outside the firm who hope to make informed decisions about the allocation of their scarce resources. Accounting standard setters (e.g., the FASB and IASB) have also made it clear that a primary user of the general purpose financial reports is the common equity investor. Needless to say, there are many other users of financial reports (e.g., governments, regulators, employees, suppliers, customers, competitors, managements etc.). There is a long literature on the use of accounting information for contracting purposes (e.g., Watts and Zimmerman, 1986). While our suggestions for future

research do not primarily focus on these alternative users of accounting information, the typical decision of an investor is not that dissimilar from that of other external
74

users. An investor considers a wide array of information to continually rebalance his/her investment portfolio. To this end, the investor will use information in general purpose financial reports to forecast free cash flows for the reporting entity and ultimately make an assessment to the intrinsic value of the firm which then is compared to observable market prices. Other external users of general purpose

financial reports such as customers, suppliers, and competitors are also interested in forecasting future free cash flows and firm value to help with their own pricing and contracting decisions. We organize our suggestions for future research into six sub-sections. First, we discuss the dearth of research making use of contextual information such as industry, sector and macro-environmental data. We make suggestions on areas where industry specific information and more macro-economic level information could be used both unconditionally and conditionally to help forecast future firm fundamentals. Second, we discuss some the wealth of information that is available to researchers within general purpose financial reports but is outside of the primary financial statements. With the advent of XBRL, and improved textual extraction techniques, researchers are better equipped to utilize the rich set of information contained in financial reports outside of the primary financial statements to improve forecasts of free cash flows and firm value. Third, we discuss the potential use of accounting information for external capital providers beyond common equity holders. With the increased development of credit markets in the last decade, there is now an enormous amount of data available on credit related instruments that can be used to help make inferences about the usefulness of accounting information for a wider set of capital providers. Fourth, we observe that there are many equity market participants The systematic approach to equity

attempting to forecast future fundamentals.

75

investing has been described as a crowded strategy. The period of August 2007 is a good example of a crowded trade deleveraging period. This offers a unique research setting to (i) examine the extent to which the security pricing implications of forecasting insights are unique, and (ii) analyze daily (as opposed to monthly) return volatility to investment strategies. Fifth, we briefly discuss the choice of forecasting techniques that are used in the existing research and suggest some possibilities for expanding the suite of statistical tools that can be used to increase the usefulness of accounting information for forecasting purposes. We offer a few suggestions as to where the use of these statistical techniques should help improve the use of accounting information in a forecasting context. Sixth, we conclude with some

philosophical comments about the literature and its ability to inform the risk vs mispricing debate. 6.1 Using contextual information The majority of the surveyed literature examines the ability of accounting attributes in the primary financial statements to help predict measures of future firm profitability (e.g., earnings, return on net operating assets, etc.) and future stock returns. In the set of papers identified in our review there is very little use of other contextual information either unconditionally or conditionally. For example, within our forecasting framework, equation (5G) highlights both the vectors of accounting information and the vector of OTHER information in forecasting future returns: Et[rt+1] = H{OICt , NOACt, NOACt-1, OTHERt} The vector of OTHER information can include market information, information from a firms disclosures, accounting and non-accounting information from other related firms, and macroeconomic information. One example of a study
76

that uses other information is Amir and Lev (1996) which examines industry specific metrics to help forecast profitability in a technology intensive industry. The reader should bear in mind that the observed under-use of contextual information is a function of the journals that we emphasized earlier in this review. There is obviously an extensive set of practitioner oriented publications and sell side research that makes use of contextual information, as well as work outside of the more mainstream accounting and finance journals that attempts to incorporate context into the forecasting framework. We argue that incorporating such information would be

fruitful for the mainstream literature. What do we mean by contextual information? We have two types of such information in mind that help improve forecasts of future firm performance based on information contained in general purpose financial reports. First, there are numerous industry specific financial metrics that are captured in financial reports. Examples include same store sales metrics for retailers, load factors for airlines, capacity utilization for manufacturers, etc. Such metrics are routinely mentioned in sell side equity analyst reports and they are available as part of financial reports (typically as part of the MD&A). Academic research in accounting and finance typically has not made use of this information either explicitly or to help condition standard models of earnings persistence. Consistent with how we typically teach financial statement analysis, enriching our understanding of the relevant key success and risk factors for relatively homogenous firms (e.g., industries), and incorporating this improved understanding in our research, should result in superior forecasts of earnings and stock returns. Second, there is a wealth of macro-economic information that can be used as part of the exercise to build forecasts of future firm fundamentals. Examples include
77

GDP growth, unemployment, inflation expectations, interest rates, shape of the yield curve, commodity prices, currency movements, etc.19 Obviously there is a large set of potential macroeconomic information that can be chosen, and with it the associated risk of in-sample data fitting. A good example of how this macro-economic

information could be used to forecast industry level fundamentals is Goldman Sachs WaveFront product (Goldman Sachs, 2004). Goldmans start with a relatively broad list of macro-economic variables. Clearly, these macro variables are inter-dependent and a model can be generated to capture the inter-temporal correlation across them yielding total effects over some future period from a given change in one of the included macro variables. For each industry grouping, a subset of the macroeconomic variables is selected, and industry level forecasts of sales growth, margins and turnovers are generated. These industry level forecasts are explicitly conditioned on key macro variables and are then combined with firm specific accounting information to generate final forecasts of firm value. The diagram below captures the overall schema used in GS WaveFront. Suffice it to say that it is a complicated structure relative to the simple linear regression approaches we tend to see in a typical accounting or finance paper. However, if we truly want to build better forecasting models, then approaches such as these could be incorporated into the standard fundamental analysis to yield superior measures of firm value.

19

Recent research on financial analysts forecasting activities suggests that they do not correctly account for inflation in their forecasts (see, for example, Basu, Markov, and Shivakumar, 2009).

78

6.2 Using accounting information beyond what is contained in the primary financial statements General purpose financial reports contain a lot more information than just the three primary financial statements. The vast majority of the research examining accounting information in the context of security pricing emphasizes the primary financial statements. This is not an arbitrary research design choice. The primary financial statements provide an articulated view of the firms ability to generate free cash flows. While there is disagreement as to the mixed attributes combined within these financial statements (e.g., Penman, 2009), there is still consensus that these are the natural starting point for any forecasting exercise. We agree with that sentiment: the primary financial statements are a natural starting point for forecasting future fundamentals. But they can be extended in various ways. Our suggestions in this
79

sub-section are limited to the information contained in the general purpose financial reports. Again, our forecasting framework identifies both the vectors of accounting information and the vector of OTHER information used to predict future stock returns: Et[rt+1] = H{OICt , NOACt, NOACt-1, OTHERt}, where there are many opportunities to identify forecast relevant information outside of the vectors of traditional accounting variables. One obvious source is the footnote information that has been examined in the prior literature. Examples include: (i) the ongoing debate on the differential value relevance of information disclosed in the financial statements relative to that disclosed in the footnotes, and (ii) the more general value relevance study of pension related items, fair value disclosures for banks, etc. As discussed earlier in the paper this literature is part of the value relevance stream. These papers are mostly about documenting contemporaneous relations between footnote items from general purpose financial reports and stock returns. Our focus is on the ability to forecast future earnings and stock returns. Some examples from recent research that fit our earnings and stock return forecasting framework include the use of valuation assumptions related to the expensing of stock options and option exercise decisions (e.g., Bartov and Mohanram, 2004 and Bartov, Nissim and Mohanram 2007), and extracting measures of readability of the annual report and documenting that firms with easier to read financial reports have more persistent earnings (Li, 2008). Several recent papers also examine footnote disclosures of fair value level 1, 2 and 3 financial assets to condition measures of book value for the purposes of equity valuation (e.g., Goh, Ng and Yong 2009).

80

There is a lot of information contained in the general purpose financial reports and it is becoming increasingly easy for users to access this information set through the tagging of tables and text that accompany the primary financial statements. The development and US adoption of eXtensible Business Reporting Language (XBRL) and the increasing sophistication of natural language processing machinery (e.g., Manning and Schutze, 1999) means that users now have substantially more information in machine readable form to conduct large scale archival analyses for the usefulness of that information for forecasting purposes. The set of information

contained in financial reports is too detailed to list, but we expect to see research efforts utilizing this information to be worthwhile. 6.3 Using information from (and for) credit markets The growth in credit markets globally in recent years has produced abundant market data, which can be used to test a large number of hypotheses. For example, trading in the secondary loan market has increased by orders of magnitude over the last decade, and credit default swaps (CDS) are now the premier instrument to trade credit risk. CDS contracts have the added benefit of tracking credit risk of differing maturities. Accounting information is potentially useful to credit investors, who are primarily interested in protecting their principal and ensuring the ability of the firm to generate cash flows sufficient to service their interest obligations. For example, utilizing information in financial statements about a firms current obligations to debtholders, researchers can identify how current financing choices will impact free cash flow available to external capital providers in future periods. Combined with

structural model estimates of credit risk, it is possible to build forecasts of expected

81

free cash flow shortfalls, which in turn can be used to estimate the term structure of corporate default probabilities. Accounting information can be used on a stand alone basis (e.g., Altman Z-Score (1968) and Ohlson O-Score (1980) type discriminant models) or in conjunction with structural models (e.g., Moodys/KMV Expected Default Frequency, EDF) to build forecasts of expected default. These types of forecasting models actually fit very nicely within our general forecasting framework: Et[defaultt+1] = H{OICt , NOACt, NOACt-1, OTHERt} where a credit investor wishes to predict the likelihood of negative outcomes (default, need for renegotiation, etc.) in the next period. These default forecasts can then in turn be used to assess the relative quality of prices in credit markets. The potential for extending the use of accounting information for forecasting aspects of future firm performance and the related pricing of securities is significant. Perhaps one of the most exciting aspects of the rich credit market data is that researchers can track prices of multiple securities for a given firm, and each of these securities differs in its return profile and free cash flow participation. This richness in capital structure creates unique opportunities to combine information across segmented capital markets (the marginal investor is likely to be different across these markets) for the purposes of forecasting, as well as emphasize different aspects of the accounting information that is relevant across the capital structure. Financial

economic theory suggests that there exist certain agency costs associated with the existence of equity and debt holders. These theories and implications for future firm profitability (e.g., debt over-hang, over-investment, asset substitution) are testable with this rich market data that is only recently available.

82

6.4 Lessons from the crowded market of August 2007 One of the key challenges in making suggestions for future research is that it is rather difficult to extend a relatively mature field. The case of identifying accounting variables that help predict future earnings and stock returns is no exception. Markets are competitive and incentives are such that inefficiencies should be quickly spotted and exploited. On average this is true. Indeed, investors are increasingly similar in the set of accounting information on which they base their decisions. The market experience in August 2007 is a great natural experiment to highlight this issue. The summer of 2007 was the start of what became a market period of unprecedented volatility, with the first week of August 2007 being a particularly acute example of this. As discussed in Lo and Khandani (2007), a large multi-strategy investment vehicle allegedly had to deleverage. The more liquid securities in this investment portfolio were equity instruments and a large equity portfolio was dumped into the market. This caused a significant feedback loop in the market. Systematic (or

quant) investors discovered that they were holding correlated portfolios and additional investors were forced into closing out positions due to their leveraged portfolios and the need to meet margin calls. This natural experiment is a great opportunity to assess the relative uniqueness of a given investment strategy. Figure 3 shows the daily returns for a characteristic portfolio that is formed on accruals (using the change in net operating measure). This portfolio is formed with forecasts of both expected returns and risk consistent with the analysis in section 5. The return volatility is clearly very large intra-month. What does it all mean? For the period of August 7-13, 2007 there was enormous volatility for generic (or well known) characteristic portfolios (e.g., B/P or analyst revisions). This suggests that

83

these information attributes are widely used within a set of quant investors. Future research could make use of this setting to help document a novel investment insight. An additional aspect of the August 2007 experience is the striking difference volatility profile of the characteristic portfolios within the month. If the researcher focuses on the returns for the month of August 2007 the volatility in the second week of August is completely masked. The extant research only examines the relation between accounting information and future stock returns at a monthly frequency (and sometimes only annually). While this may seem reasonable as the investment insight from accounting information is typically longer in nature, it does ignore the short term volatility from crowded positions. This is clearly important for the broader decision as to the usefulness of accounting information for forecasting earnings and returns, especially if one objective of the researcher is to have impact on practice. 6.5 Using alternative archival research techniques The most common statistical technique used in the surveyed archival research papers is cross-sectional and/or panel linear regression. The general linear model seems reasonable as a first approximation to the underlying data generating process. But there are a variety of other approaches that could be used to help improve our ability to forecast future fundamentals and stock returns. In this sub-section we make a few suggestions for future research along with some recent examples. There is a substantial literature examining structural relationships within social networks. The field of social network analysis has its roots in sociology, social psychology and anthropology with the development of graph theory and sociograms to capture the relationships between members of a network (e.g., Moreno, 1934). This is a particularly rich research area and one that has only recently received
84

attention in the accounting and finance literatures (see, for example, the analyses in Larcker, Richardson, Seary and Tuna, 2008, on the impact of director links on executive compensation, and Kuhnen, 2008, on the impact of mutual fund linkages and mutual fund performance). The techniques developed in social network analysis can be applied to any setting where there are relations between agents in the network and there is reason to believe that these links impact the decision making of the agents in the network. For forecasting future firm profitability the structural linkages

between firms implied by these networks could improve those forecasts. Cohen and Frazzinis (2008) analysis of the links between customers and suppliers from supplemental financial statement disclosures is an excellent example of how to use information about structural linkages between firms to generate improved forecasts of future firm performance. A related body of literature is the intra-industry information transfer research which documented spillover effects from one firm onto another (e.g., Baginski 1987). One aspect of these networks that is often over-looked in previous accounting and finance research is the density of the network. As an example, Figure 4 gives a graphical representation of the links between directors of public companies listed on the Australian Stock Exchange. Figure 5 shows the linkages between Japanese firms in 2008 based on cross-holdings between companies: this is clearly a very dense network. We show these two graphs as (i) examples of the types of links that exist between firms (e.g., over-lapping directors and cross-equity holdings), and (ii) to help visualize the richness of measures that can be extracted. There are many other ways in which firms can be linked and there are many measures that can be extracted from such graphs. It is not only the links between immediate firms that matter for the

85

purposes of capturing spill over affects, but also the overall inter-connectedness of the network. The interested reader would benefit from referring to Scott (2000). The majority of empirical research on accounting anomalies and fundamental analysis assumes that the underlying data generating process is the same for all firms. This need not be true. For example, the persistence of abnormal earnings may not vary solely based on industry membership: it may vary on other difficult to observe characteristics. The traditional research paper would suggest candidate conditioning variables and then progressively add interaction affects to test for their significance. Alternative approaches to adding progressively more interaction variables include (i) Bayesian hierarchical linear modeling, where the strength of an attribute on future earnings or stock returns is directly a function of firm characteristics and (ii) clusterwise regression and recursive partitioning, where the relation between an attribute and future earnings ad/or returns is allowed to be non-linear and have complicated interaction affects. Research in areas other than fundamental analysis has made some use of such techniques (see e.g., Bushee and Goodman, 2007, Larcker and Richardson, 2004, and Larcker, Richardson and Tuna, 2007). Valid criticisms of these approaches are that they are less theoretically motivated and are at risk of in-sample data fitting. However, this latter concern can be mitigated through the use of out-ofsample cross validation techniques, which researchers can then use to build superior forecasting models. 6.6 General Discussion In this sub-section we offer a few philosophical comments about the entire literature we review. The debate about market efficiency, which under-pins a large fraction of the literature on the use of accounting information for security pricing, has

86

moderated substantially over the years.

Grossman and Stiglitz (1980) make the

pertinent observation that security prices need to become efficient and this can only happen if capital market participants actively trade on useful information driving security prices toward a true price. In this sense the capital market is adaptive in its efficiency (see e.g., Lo, 2004 on market efficiency from an evolutionary perspective). The analysis that we provide in the previous section speaks to this issue directly: markets adapt over time and what was once mis-priced is no longer so. This field of research is coming full circle: by documenting robust relations between a given accounting attribute and future stock returns the research process helps improve market efficiency in this evolutionary view. A lot of prior research effort has attempted to explain away anomalous return patterns by showing that the association between a given attribute and future stock returns is concentrated in a subset of firms. These sub-sets tend to be smaller, less liquid securities where transaction costs and/or idiosyncratic risk are greater. This is a very useful and informative literature, but there is a risk of misinterpreting the results. Market (in)efficiency is clearly related to these partitioning variables. When

securities are more liquid and information is more readily available, we would expect security prices to be more efficient. Thus, finding that a relation between an attribute and future stock returns is concentrated in a sub-set of less liquid securities is not prima facie evidence in support of market efficiency. It may well still be possible to successfully implement an investment idea in this subset of securities, and, in Grossman-Stiglitz terms, receive an adequate compensation for your efforts. With the wealth of information available to investors today, and the computational power available for empirical archival analysis, there is a risk of information load which could exacerbate frictions in the market and impede the price
87

discovery process. There has been considerable research on information over-load and the impact that this w/could have on the price discovery process (see e.g., Daniel, Hirshleifer and Teoh, 2002, and Libby, Bloomfield and Nelson, 2002 for surveys of this literature). Going forward we expect the quantity of information, as well as the ease with which researchers can access it (i.e., machine readability), to increase. Clearly, our ability to conduct archival analysis has increased through time (both in terms of research design but more importantly in terms of computing power to be able to manipulate large datasets). This creates a significant inefficiency bias in research that uses long time series of data. The further one goes back in time to examine associations between an attribute and future stock returns the greater this risk. In some sense it is relatively easy to find an anomaly in the 1960s, 1970s and even in the 1980s. It is much harder to do so in the last ten years. Part of this is attributable to advances in computing ability, but part is also surely attributable to the increase in capital invested on the basis of many of the insights discussed in this review. For a more detailed discussion of this effect we refer the reader to Green, Hand and Soliman (2009). The joint impact of the increased information (i.e., the democratic access), and an increasing ability of investors to process and manipulate this information, on the ability of investors to forecast earnings and stock returns is surely a productive future research field. A common thread that we weaved throughout our survey is the ability of accounting information to help forecast future earnings and future stock returns. This is a necessary condition for the usefulness of the research effort (i.e., are we able to improve our forecasts?). It is a separate, and arguably, more interesting question, as

88

to whether security prices reflect that information on a timely basis. For example, the question of whether an attribute is associated with future stock returns is the focus of much of the literature that we have surveyed. We would like to re-emphasize that this understates the potential usefulness of this field of research. Given the multiple other users of general purpose financial reports (e.g., customers, suppliers, competitors, management etc.) that make financial decisions, analysis to improve forecasting models of future earnings are invaluable regardless of the answer to the risk vs mispricing debate. 7. Conclusion The goal of our survey is to highlight recent advances in the research areas of accounting anomalies and fundamental analysis. A key theme to our survey is that information contained in general purpose financial reports help facilitate better decision making about the allocation of scarce economic resources from an investors perspective. So, we deliberately focus our attention on research whose primary aim is to forecast earnings and stock returns. Given the existence of prior related surveys, our main goal is to complement and fill-in-the-gaps from prior literature reviews. Moreover, our review strategy attempts to both let the market speak on which research topics and studies are deemed to be of greatest interest to academics and to highlight papers that fit within our forecasting framework. We adopt a number of complementary approaches to organize and capture recent advances in this literature. The first part of our review tabulates a list of the most highly-cited research studies on accounting anomalies and fundamental analysis published or distributed since the year 2000. We also organize and categorize these highly-cited studies by identifying their common and overlapping citations to earlier papers in the literature. The second part of our survey examines practitioner and
89

academic views on this literature.

In the final part of our review, we offer

suggestions for future research and draw on the recent conceptual advances from both investment practice and academic research to demonstrate a more-encompassing definition and treatment of risk and transaction costs in empirical tests of equity market anomalies. Our first major observation based on the citation analysis indicates that the most of the highly cited papers are published in finance journals (see, for example, Jegadeesh and Titman, 2001, Hong, Lim and Stein, 2000). On the other hand, there is a large number of accounting papers on anomalies and fundamental analysis that have received high citation counts (see, for example, Xie, 2001, and Richardson, Sloan, Soliman and Tuna, 2005). Based on this set of high impact papers, we use common citation analysis of prior research and identify four clusters of research papers. Accordingly, we organize our literature review into Fundamental Analysis, Accruals Anomaly (including related investment anomalies), Underreaction to Accounting Information (with a particular emphasis on post-earnings announcement drift (PEAD)), and Pricing Multiples/Value Anomaly sections. Given that research on anomalies and fundamental analysis has important and immediate applications and intellectual spillovers to actual practice, we survey investment professionals and leading academics about their views on academic research in this area. The questions were also designed to gather useful information about future demand for specific academic research on anomalies and fundamental analysis. Both the practitioner and academic respondents placed high importance on future academic research on: (i) empirical tests of investor behavior; (ii) empirical tests of asset pricing, risk and factor models; (iii) empirical research on forecasting

90

firm and industry fundamentals; and (iv) empirical discovery and investigation or new anomalies or signals. Based on (i) the prominence of the accruals anomaly in the recent literature, and (ii) practitioner interest in future innovations related to empirical tests of investor behavior and empirical tests of asset pricing, risk and factor models, we conduct our own empirical analyses to help advance some concepts and approaches to be considered and applied in future research studies. In particular, we provide new insights on (i) the time series variation in the negative relation between accruals and future returns (specifically, the extent to which this relation has disappeared consistent with market learning), and (ii) whether the relation is robust to a comprehensive empirical treatment of risk and transaction costs. Our analysis shows that, while the negative relation between accruals and future stock returns is robust to the comprehensive treatment of risk and transaction costs, the relation has greatly attenuated over time. In recent years one could conclude that the information in accruals is now fully priced by the market. Finally, we offer a variety of suggestion for future research on the use of general purpose financial reports for forecasting earnings and stock returns. This is an area that is at the core of the accounting profession. We are hopeful that future research will embrace the suggestions offered in this paper and capitalize on the wealth of accounting information that is available for forecasting future earnings and stock returns. We would like to remind researchers of the opportunities to be gained from improved forecasts of future earnings and cash flows. It is still the case that perfect foresight of future earnings and cash flows would generate a very profitable investment strategy. So, while it may be the case that forecasting has become an

91

increasingly competitive activity, the potential rewards from this activity are still substantial.

92

References Abarbanell, J., Bushee, B., 1997. Fundamental analysis, future earnings, and stock prices. Journal of Accounting Research 35, 124. Abarbanell, J., Bushee, B., 1998. Abnormal returns to a fundamental analysis strategy. The Accounting Review 73, 1945. Affleck-Graves, J., and R. Miller, 2003. The information content of calls of debt: Evidence from long-run stock returns. Journal of Financial Research 26, 421-447. Agrawal, A., J. Jaffe and G. Mandelker, 1992. The post-merger performance of acquiring firms: A re-examination of an anomaly. Journal of Finance 47, 16051621. Ali, A., X. Chen, T. Yao, and T. Yu (2008). Do mutual funds profit from the accruals anomaly? Journal of Accounting Research 46, 1-26. Ali, A., X. Chen, T. Yao, and T. Yu. 2007. Riding the post earnings announcement drift: Mutual fund trades, market frictions, and market efficiency. Working Paper. Ali, Ashiq, Lee-Seok Hwang, and Mark A. Trombley, 2003. Arbitrage risk and the book-to-market anomaly. Journal of Financial Economics 69, 355-373. Ali, D. A., L. Hwang, and M. Trombley, 2000. Accruals and future stock returns: tests of the nave investor hypothesis. Journal of Accounting, Auditing & Finance 15, 161-181. Ali, A., L. Hwang, and M. Trombley, 2003. Residual-income-based valuation predicts future stock returns: Evidence on mispricing vs. risk explanations. Accounting Review 78, 377-396. Altman, E. 1968. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance 23, 589-609. Amir, E., and B. Lev, 1996. Value-relevance of nonfinancial information: The wireless communications industry." Journal of Accounting and Economics, 3-30. Anderson, C., and L. Garcia-Feijoo, 2006, Empirical evidence on capital investment, growth options, and security returns, Journal of Finance 61, 171-194. Asquith, P., 1983. Merger bids, uncertainty, and stockholder returns. Journal of Financial Economics, 11, 51-83. Baginski, S. 1987. Intraindustry information transfers associated with management forecasts of earnings. Journal of Accounting Research, 25, 196-216. Baginski, S., and J. Wahlen, 2003. Residual income risk, intrinsic values, and share prices. The Accounting Review 78.

93

Baker, M., and J. Wurgler 2006, Investor sentiment and the cross-section of stock returns, Journal of Finance. Ball, R., 1992. The earningsprice anomaly. Journal of Accounting and Economics 15, 319345. Barberis, N., and R. Thaler, 2003. A survey of behavioral finance. Chapter 15 in Handbook of the Economics of Finance, Edited by G.M. Constantinides, M. Harris and R. Stulz. Elsevier Science B.V., Netherlands. Barth, M. E., and A. P. Hutton 2004. Analyst earnings forecast revisions and the pricing of accruals. Review of Accounting Studies 9, 59-96. BARRA Inc., 1998. United States Equity: Version 3 (USE3). Risk Model Handbook. Bartov, E., and P. Mohanram. 2004. Private information, earnings manipulations and executive stock-option exercises. The Accounting Review, 79, 889-920. Bartov, E., D. Nissim, and P, Mohanram 2007. Managerial discretion and the economic determinants of the disclosed volatility parameter for valuing ESOs, Review of Accounting Studies 12, 155-179. Bartov, E., S. Radhakrishnan, and I. Krisy, 2000. Investor sophistication and patterns in stock returns after earnings announcements. The Accounting Review 75, 43-63. Basu, S., 1977. The investment performance of common stocks in relation to their priceearnings ratios: a test of the efficient markets hypothesis. Journal of Finance 32, 663682. Basu, S., S. Markov, and S. Shivakumar, 2009. Do financial analysts suffer from inflation illusion? Review of Accounting Studies, Forthcoming. Bauman, M., 1996, A review of fundamental analysis research in accounting. Journal of Accounting Literature 15, 1-33. Beneish, M., and M. Vargus 2002. Insider trading, earnings quality, and accrual mispricing. Accounting Review 77, 755-791. Beneish, M., Lee and Tarpley, 2001. Contextual financial statement analysis through the Prediction of Extreme Returns Review of Accounting Studies 6. Bernard, V., 1989. Capital markets research in accounting during the 1980s: A critical review. The state of accounting research as we enter the 1990s: Illinois PhD Jubilee 1939-1989, edited by T.J. Frecka. Urbana-Champaign, IL: University of Illinois. Bernard, V., Thomas, J., 1989. Post-earnings-announcement drift: delayed price response or risk premium? Journal of Accounting Research 27, 148. Bernard, V., Thomas, J., 1990. Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics 13, 305340.

94

Beyer, A., D. Cohen, and T. Lys, 2009. Corporate information environment. Journal of Accounting and Economics, this issue Billet, Matthew, Mark Flannery, and Jon Garfinkel, 2006, Are bank loans special? Evidence on the post-announcement performance of bank borrowers, Journal of Financial and Quantitative Analysis 41, 733-752. Bradshaw, M., S. Richardson, R. Sloan, 2001. Do analysts and auditors use information in accruals? Journal of Accounting Research. Brandt, Kishore, Santa-Clara, and Venkatachalam, 2008. Earnings announcements are full of surprises, working paper. Brav, A. and Heaton 2002. Competing theories of financial anomalies. Review of Financial Studies; 15, 575-606. Britten-Jones, M., 1999. The sampling error in estmates of mean-variance efficient portfolio weights. Journal of Finance, 54, 655-671. Brown, L., and Gardner, J., 1985a. Applying citation analysis to evaluate the research contributions of accounting faculty and doctoral programs. The Accounting Review 60 (2): 262-277. Brown, L., and Gardner, J., 1985b. Using citation analysis to assess the impact of journals and articles on contemporary accounting research (CAR). Journal of Accounting Research 23 (1): 84-109. Brown, L., and Heufner, R., 1994. The familiarity with and perceived quality of accounting journals: views of senior accounting faculty in leading U.S. MBA programs. Contemporary Accounting Research 11, 223-250. Bushee, B., and T. Goodman, 2007. Which institutional investors trade based on private information about earnings and returns? Journal of Accounting Research 45, 1-31. Byrne, A. and M. Brooks, 2008. Behavioral finance: Theories and evidence. Research Foundation of CFA Institute Monograph Series: 1-26. Chan, K., L. Chan, N. Jegadeesh, and J. Lakonishok, 2006. Earnings quality and stock returns. Journal of Business 79, 1041-1082. Chava, Sudheer and Purnanandam, Amiyatosh K., 2007. Is default risk negatively related to stock returns? Working paper. Chen, L, and S. Zhao, 2008. What drives stock price movement? Working paper. Chen, F., Jorgensen, B. and Y. Yong Keun, 2005. Implied cost of equity capital in earnings-based valuation: International evidence. Accounting and Business Research 34, 323-344. Cheng, C. and W. Thomas, 2006. Evidence of the abnormal accrual anomaly incremental to operating cash flows. The Accounting Review 81, 1151-1168.
95

Chordia, T. and L. Shivakumar. 2009. Inflation illusion and post-earningsannouncement drift. Journal of Accounting Research, forthcoming. Claus, J. and J. Thomas, 2001. Equity premia as low as three percent? Evidence from analysts' earnings forecasts for domestic and international stock markets. Journal of Finance 56, 1629-1666. Cochrane, J.H., 2005. Asset pricing. Princeton: Princeton University Press. Cohen, L., and A. Frazzini, 2008. Economic links and predictable returns. Journal of Finance 63, 1977 - 2011. Collins, D., and P. Hribar, 2000. Earnings-based and accrual-based market anomalies: one effect or two? Journal of Accounting & Economics 29, 101-123. Collins, D. G. Gong, and P. Hribar, 2003. Investor sophistication and the mispricing of accruals. Review of Accounting Studies, 8, 251-276. Conner, 2000. Cooper, M, H. Gulen, and M. Schill, 2008. Asset growth and the cross-section of stock returns, Journal of Finance, forthcoming. Core, J., W. Guay, and R. Verdi., 2008. Is accruals quality a priced risk factor? Journal of Accounting and Economics 46, 2-22. Cusatis, M, and Woolridge, 1993. Restructuring through spinoffs: The stock market evidence. Journal of Financial Economics, 293-311. Courteau, L., J. Kao, and G. Richardson, 2001. Equity valuation employing the ideal versus ad hoc terminal value expressions. Contemporary Accounting Research. Damodaran, A., 2005. Valuation approaches and metrics: A survey of the theory and evidence. Foundations and Trends in Finance Voume. 1 (8), 693-784. Daniel, K., D. Hirshleifer, and S. Teoh. 2002. Investor psychology in capital markets: evidence and policy implications. Journal of Monetary Economics, 49, 139-209. Dechow, P., and I. Dichev, 2002. The quality of accruals and earnings: The role of accrual estimation errors. The Accounting Review 35-59 Dechow, P., and W. Ge, 2006. The persistence of earnings and cash flows and the role of special items: implications for the accrual anomaly. Review of Accounting Studies 11, 253-296. Dechow, P., S. Richardson, and R. Sloan 2008, The persistence and pricing of the cash component of earnings, Journal of Accounting Research. Dechow, P, W. Ge, and C. Schrand, 2009. Earnings quality and earnings management. Journal of Accounting and Economics, this issue.

96

Dechow, P. and R. Sloan, 1997. Returns to contrarian investment strategies: Tests of nave expectations hypotheses, Journal of Financial Economics 43, 3-27. DellaVigna, S., and Pollet, J., 2006. Investor inattention, firm reaction, and friday earnings announcements. Working paper. Desai, H. M., S. Rajgopal, and M. Venkatachalam, 2004. Value-glamour and accruals mispricing: one anomaly or two? The Accounting Review 79, 355-385. Diether, K., C. Malloy, and A. Scherbina Anna, 2002. Differences of opinion and the cross section of stock return, Journal of Finance 57, 2114-2141. Doyle, J., R. Lundholm, and M. Soliman. 2006. The extreme future stock returns following I/B/E/S earnings surprises. Journal of Accounting Research 44, 849887. Easton, P., 2009(a). Estimating the cost of capital implied by market prices and accounting data. Foundations and Trends in Finance Vol. 2. No. 4: 241-364. Easton, P., 2009(b). Discussion of accounting data and value: The basic results. Contemporary Accounting Research 26, 261-272. Easton, P., M. McAnally, P. Fairfield, and X. Zhang. 2009. Financial Statement Analysis and Valuation. Cambridge Business Publishers. Easton, P., Monahan, S., 2005, An evaluation of the reliability of accounting based measures of expected returns: A measurement error perspective, The Accounting Review 80, 501-538. Easton, P. and G. Sommers, 2007. Effect of analysts' optimism on estimates of the expected rate of return implied by earnings forecasts. Journal of Accounting Research 45, 983-1016. Easton, P., Taylor, G., Shroff, P., Sougiannis, T., 2002, Using forecasts of earnings to simultaneously estimate growth and the rate of return on equity investment, Journal of Accounting Research 40, 657-676. Fairfield, P. M., J. S. Whisenant, and T. L. Yohn, 2003. Accrued earnings and growth: implications for future profitability and market mispricing. Accounting Review 78, 353-371. Fama, E., 1970. Efficient capital markets: a review of theory and empirical work. Journal of Finance 25, 383-417. Fama, E. and K. French, 1992. The cross-section of expected stock returns. Journal of Finance 47, 427-465. Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 356. Fama, E., French, K., 1995. Size and book-to-market factors in earnings and returns. Journal of Finance 50, 131155.
97

Fama, E., French, K., 1997, Industry costs of equity, Journal of Financial Economics 43, 153-193. Fama and French, 2000. Forecasting profitability and earnings. Journal of Business 72, 161-175. Fama, E., and K. French, 2008. Dissecting anomalies. Journal of Finance 63, 1653 1678. Feltham, G., Ohlson, J., 1995. Valuation and clean surplus accounting for operating and financial activities. Contemporary Accounting Research 11, 689731. Foster, G., Olsen, C., Shevlin, T., 1984. Earnings releases, anomalies and the behavior of security returns. The Accounting Review 59, 574603. Francis, J., P. Olsson, and D. Oswald, 2000. Comparing the accuracy and explainability of dividend, free cash flow and abnormal earnings equity valuation models. Journal of Accounting Research 38, 45-70. Francis, J., R. LaFond, P. Olsson, and K. Schipper, 2007, Information uncertainty and post earnings-announcement-drift. Journal of Business, Finance and Accounting 34, 403. Francis, J., R. LaFond, P. Olsson, K. Schipper, 2005. The market pricing of accruals quality. Journal of Accounting and Economics 39, 295-327. Francis, J., I. Khurana, and R. Pereira, 2005. Disclosure incentives and effects on cost of capital around the world. The Accounting Review 80, 1125. Gebhardt, W., Lee, C., Swaminathan, B., 2001, Toward an implied cost of capital, Journal of Accounting Research 39, 135-176. Goh, B. W., J. Ng, and K. O. Yong. 2009. Market pricing of banks fair value assets reported under SFAS 157 during the 2008 economic crisis. Working paper. Goldman Sachs, 2004. Introducing wavefronts: A new way of trading macro equity views: Equity trading strategies. Greene, J., J.Hand, and M. Soliman, 2009. The demise of the accruals anomaly. University of North Carolina, working paper. Griffin, J, P. Kelly,. and F. Nardari, 2007. Measuring short-term international stock market efficiency, Working paper. Griffin and Lemmon, 2002. Book-to-market equity, distress risk, and stock returns. Journal of Finance 57, 2317-2336. Grinhold, R., and R. Kahn, 2000. Active Portfolio Management: A Quantitative Approach for Producing Superior Returns and Controlling Risk, 2nd ed, New York: McGraw-Hill. Grossman, S., and J. Stiglitz, 1980. On the impossibility of informationally efficient markets. American Economic Review 70, 393-408.
98

Guay, W., 2000. Discussion of value investing: The use of historical financial statement information to separate winners from losers. Journal of Accounting Research 38 (Supplement), 43-51. Guay, W, Kothari, S. and S. Shu, 2005. Properties of implied cost of capital using analysts' forecasts, MIT Sloan Working Paper No. 4422-03. Hail, L., and C. Leuz, 2006. International differences in the cost of equity capital: Do legal institutions and securities regulation matter? Journal of Accounting Research 44, 485-531. Hesford, J., S. Lee, W. Van der Stede, and S. Young, 2007. Management accounting: a bibliographic study. In: Chapman, Christopher S. and Hopwood, Anthony G. and Shields, Michael D., (eds.) Handbook of management accounting research. Elsevier, pp. 3-26. Hirshleifer, D., 2001. Investor psychology and asset pricing. Journal of Finance 56, 1533-1597. Hirshleifer, D., and S.H. Teoh. 2006. Limited investor attention and stock market misreactions to accounting information. Working Paper, Ohio State University. Hirshleifer, D., K. Hou, and S. Teoh, 2006., The accrual anomaly: risk or mispricing? Working paper, Ohio State University, 2006. Hirshleifer, D., K. Hou, S. Teoh, and Y. Zhang, 2004. Do investors overvalue firms with bloated balance sheets? Journal of Accounting and Economics 38, 297331. Hirshleifer, D., K. Hou, and S.H. Teoh, 2008, Accruals and aggregate stock market returns, Journal of Financial Economics forthcoming. Hirshleifer, D., J. Myers, L. Myers, and S. Teoh, 2008. Do individual investors drive post-earnings announcement drift? Working paper, Ohio State University. Hong, H., Lim, T., Stein, J.C., 2000. Bad news travels slowly: Size, analyst coverage, and the profitability of momentum strategies. Journal of Finance 55, 265295. Hou, K. and D.T. Robinson, 2006, Industry concentration and average stock returns. Journal of Finance 61, 19271956. Hou, K, and M. van Dijk, 2008 Resurrecting the size effect: Firm size, profitability shocks, and expected stock returns, Working paper. Hou, K., M. van Dijk, and Y. Zhang, 2009. The implied cost of capital: A new approach, Working paper. Ibbotson, R., 1975. Price performance of common stock new issues. Journal of Financial Economics 3, 235-272. Ikenberry, D., J. Lakonishok, and T. Vermaelen, 1995. Market underreaction to open market share repurchases. Journal of Financial Economics 39, 181-208. Jegadeesh, N., and S. Titman, 1993. Retuens to buying winners and selling losers: implications for stock market efficiency. Journal of Finance 48, 65-91.
99

Jegadeesh, N., and S. Titman, 2001. Profitability of Momentum Strategies: An Evaluation of Alternative Explanations, Journal of Finance 56, 699-720. Juettner-Nauroth, B, and K. Skogsvik, 2005. A Note on the Impact of Conservative Accounting Principles in Abnormal Earnings, working paper. Ke, B., and S. Ramalingegowda, 2005. Do institutional investors exploit the postearnings announcement drift? Journal of Accounting and Economics 39, 25-53. Keim, D. and W. Ziemba, 2000. Security market imperfections in world wide equity markets. University Press, Cambridge, UK. Keloharju, M., 2008. Whats new in finance? European Financial Management 14, 564-608. Khan, M., 2008. Are accruals mispriced evidence from tests of an intertemporal capital asset pricing model. Journal of Accounting and Economics 45, 55-77. Kimbrough, M. 2005. The effect of conference calls on analyst and market underreaction to earnings announcements. The Accounting Review 80, 189-219. Kothari, S., 2001. Capital markets research in accounting. Journal of Accounting and Economics 31, 105-231. Kothari, S., E. Loutskina and V. Nikolaev, 2007. Agency theory of overvalued equity as an explanation for the accrual anomaly. MIT working paper. Kraft, A., A. Leone, and C. Wasley, 2007. Regression-based tests of the market pricing of accounting numbers: The Mishkin test and ordinary least squares. Journal of Accounting Research 45, 1081-1114. Kuhnen, C., 2008. Business networks, corporate governance and contracting in the mutual fund industry. Journal of Finance, Forthcoming. LaFond, R. 2005. Is the accrual anomaly a global anomaly?, Working Paper Sloan School of Management. Lakonishok, J, A. Shleifer, and R. Vishny, 1994. Contrarian investment, extrapolation and risk. Journal of Finance 49, 1541-1578. Lakonishok, J. and T. Vermaelen, 1990. Anomalous price behavior a round repurchase tender offers. Journal of Finance 45, 455-477. Larcker, D., and S. Richardson, 2004. Fees paid to audit firms, accrual choices and corporate governance. Journal of Accounting Research 42, 625-658. Larcker, D., S. Richardson, A. Seary and I. Tuna. 2008. Director networks,executive compensation, and firm performance. Working paper. Larcker, D., S. Richardson and I. Tuna, 2007. Corporate governance, accounting outcomes, and organizational performance. The Accounting Review 82, 960-1008.

100

Lee, C., 2001. Market efficiency and accounting research: a discussion of capital market research in accounting by S.P. Kothari. Journal of Accounting and Economics 31: 233-253. Leippold, M., and H. Lohre (2008). 'Data snooping and the global accrual anomaly', Working paper. Imperial College London. Lev, B., and D. Nissim, 2006. The persistence of the accruals anomaly. Contemporary Accounting Research 23, 193-226. Lev, B., and S. R. Thiagarajan, 1993. Fundamental information analysis. Journal of Accounting Research 31, 190-215. Levi, S. 2008. Voluntary disclosure of accruals in earnings press releases and the pricing of accruals. Review of Accounting Studies 13, 1-21. Lewellen, J., 2004, Predicting returns with financial ratios. Journal of Financial Economics 74, 209-235. Li, F., 2008. Annual report readability, current earnings and earnings persistence. Journal of Accounting and Economics 45, 221-247. Liang, L., 2003. Post-earnings-announcement-drift and market participants information processing biases. Review of Accounting Studies, 321-345. Libby. R., R. Bloomfield and M. Nelson, 2002. Experimental research in financial accounting. Accounting Organizations and Society, 775-810. Liu, W., N. Strong, and X. Xu (2003). Post-earnings-announcement drift in the UK. European Financial Management 9, 89-116. Livnat, J. and R. Mendenhall, 2006. Comparing the post-earnings announcement drift for surprises calculated from analyst and time series forecasts. Journal of Accounting Research 44, 177-205. Lo, A. 2007. The adaptive markets hypothesis: Market efficiency from an evolutionary perspective. Journal of Portfolio Management - 30th Anniversary Issue, 15-29. Lo, A., and A. Khandani, 2007. What happened to quants in August 2007? Working paper. Loughran, T., and J. Ritter, 1995. The new issues puzzle. Journal of Finance 50. Loughran, T., and A. Vijh, 1997. Do long-term shareholders benefit from corporate acquisitions? Journal Of Finance 52. Lundholm, R., and T. OKeefe. 2001. Reconciling value estimates from the discounted cash flow model and the residual income model. Contemporary Accounting Research 18, 31135. Madhavan, A., 2000. Market Microstructure: A survey. Journal of Financial Markets.

101

Manning, C., and H, Schutze. 1999. Foundations of statistical natural language processing. MIT Press. Mashruwala, C., S. Rajgopal, and T. Shevlin (2006). 'Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs'. Journal of Accounting & Economics 42, 3-33. McConnell, J. and Ovtchinnikov, 2004. Predictability of long-term spinoff returns. Journal of Investment Management 2. McRae, T., 1974. A citiational analysis of the accounting information network. Journal of Accounting Research 12,: 80-92. Mendenhall, R. 2004. Arbitrage risk and post-earnings-announcement drift. Journal of Business 77. Michaely, R., R. Thaler, and K. Womack, 1995. Price reactions to dividend initiations and omissions: Overreaction or drift? The Journal of Finance 50, 573-608. Mohanram, P. 2005. Separating winners from losers among low book-to-market stocks using financial statement analysis. Review of Accounting Studies 10, 122170. Moreno, J., 1934. Who Shall Survive? Foundations of Sociometry, Group Psychotherapy, and Sociodrama. Washington D.C.: Nervous and Mental Disease Publishing Co. (Beacon House). Nagel, S., 2005. Short sales, institutional investors, and the cross section of stock returns. Journal of Financial Economics 78, 277-309 Narayanamoorthy, G., 2006. Conservatism and cross-sectional variation in the postearnings announcement drift. Journal of Accounting Research 44, 763789. Ng J., T. Rusticus and R. Verdi, 2008. Implications of transaction costs for the postearnings-announcement drift. Journal of Accounting Research 46. Ohlson, J, 1980. Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research 18, 109131. Ohlson, J., 1990. A synthesis of security valuation theory and the role of dividends, cash flows, and earnings. Contemporary Accounting Research 6, 648676. Ohlson, J., 1995. Earnings, book-values, and dividends in equity valuation. Contemporary Accounting Research 11, 661687. Ohlson, J., 2001., Earnings, book values and dividends in equity valuation: An empirical perspective. Contemporary Accounting Research 18, 107-120. Ohlson, J., 2005. On accounting-based valuation formulae. Review of Accounting Studies 10: 323-347. Ohlson, J., 2009. Accounting data and value: The basic results. Contemporary Accounting Research (26), 231-259.
102

Ohlson, J., and Z. Gao, 2006. Earnings, earnings growth and value. Foundations and Trends in Accounting 1, 170. Ohlson, J. and Juettner-Nauroth, 2005. Expected eps and eps growth as determinants of value. Review of Accounting Studies 10, 349-365. Ou J. and Penman, S., 1989. Financial statement analysis and the prediction of stock returns. Journal of Accounting and Economics 11, 295329. Pastor, L., M. Sinha, and B. Swaminathan, 2007. Estimating the intertemporal riskreturn tradeoff using the implied cost of capital, Journal of Finance. Penman, S., 2001. On comparing cash flow and accrual accounting models for use in equity valuation: A response to Lundholm and O'Keefe (CAR, Summer 2001), Contemporary Accounting Research 18, 681-692. Penman, S., 2009. Financial Statements Analysis and Security Valuation, Irwin McGraw Hill, 4th ed. Penman, S., 2009. The design of financial statements. Center for Excellence in Accounting and Security Analysis. Columbia Business School. Penman, S., S. Richardson and I. Tuna, 2007. The book-to-price effect in stock returns: Accounting for leverage. Journal of Accounting Research 45, 427-467. Penman, S., and X. Zhang, 2002. Accounting conservatism, the quality of earnings, and stock returns. The Accounting Review 77, 237-264. Penman, S., and X. Zhang, 2006. Modeling sustainable earnings and P/E ratios with financial statement information. Working Paper. Pincus, M., S. Rajgopal, and M. Venkatachal, 2007. The accrual anomaly: international evidence. Accounting Review 82, 169-203. Piotroski, J., 2000. Value investing: The use of historical financial statement information to separate winners from losers. Journal of Accounting Research 38, 1-41. Pontiff, J., 2006. Costly arbitrage and the myth of idiosyncratic risk. Journal of Accounting and Economics 42, 35-52. Rakhlin, D. and G. Sofianos, 2006. The impact of an increase in volatility on trading costs. Journal of Trading 1, 4350. Rau, P. R., and T. Vermalen. 1998. Glamour, value and the post-acquisition performance of acquiring firms. Journal of Financial Economics 49, 223-253. Reed, A., 2007, Costly short-selling and stock price adjustments to earnings announcements, Working paper, University of North Carolina. Reinganum, M., 1981. Misspecification of the capital asset pricing: empirical anomalies based on earnings yields and market values. Journal of Financial Economics 9, 29-46.
103

Richardson, S. and R. Sloan, 2003. External financing and future stock returns. Rodney L. White Center for Financial Research Working Paper No. 03-03 Richardson, S., R. Sloan, M. Soliman, and I. Tuna, 2005. Accrual reliability, earnings persistence and stock prices. Journal of Accounting & Economics 39, 437-485. Richardson, S., R. Sloan, M. Soliman, and I. Tuna, 2006. The implications of firm growth and accounting distortions for accruals and profitability. The Accounting Review 81, 713-743. Richardson, S., R. Sloan, and I. Tuna, 2006. Balance sheet information and future stock returns. Working paper. Richardson, S, S. Teoh, and P. Wysocki, 2003. The walk-down to beatable analyst forecasts: The role of equity issuance and insider trading incentives. Contemporary Accounting Research 21, 885-924. Richardson, G., and S. Tinaikar, 2004. Accounting based valuation models: What have we learned? Accounting and Finance 44, 223-255. Ritter, J., 2003. Differences between European and American IPO markets. European Financial Management 9, 421-434. Ross, S., 1976. The arbitrage theory of capital asset pricing. Journal of Economic Theory 13, 341-360. Sadka, R., 2006. Momentum and post-earnings-announcement drift anomalies: The role of liquidity risk. Journal of Financial Economics 80, 309-349. Schwert, G.W., 2003. Anomalies and market efficiency. Chapter 15 in Handbook of the Economics of Finance, Edited by G.M. Constantinides, M. Harris and R. Stulz. Elsevier Science B.V., Netherlands. Schwert, G.W., 2007. Report of the editorial office for the year 2006. Journal of Financial Economics. Scott, J., 2000. Social network analysis: a handbook. SAGE Publications, London. Shanthikumar, D., 2004. Small and large trades around earnings announcements: Does trading behavior explain post-earnings announcement drift? Working paper, Harvard Business School. Shivakumar, L. 2006. Accruals, cash flows and the post-earnings-announcement drift. Journal of Business Finance & Accounting 33, 1-25. Sloan, R, 1996. Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review 71, 289-316. Soliman, M. 2008 The use of DuPont analysis by market participants. The Accounting Review 8, 823-853. Spiess, K., and J. Affleck-Graves, 1999. The long-run performance of stock returns following debt offerings. Journal of Financial Economics 54, 45-73.
104

Subrahmanyam, A., 2007. Behavioral finance: a review and synthesis. European Financial Management 14, 12-29. Titman, S., J. Wei, and F. Xie. 2004. Capital investments and stock returns. Journal of Financial and Quantitative Analysis 39, 677-700. Vassalou, M., and Y. Xing, 2004. Default risk in equity returns. Journal of Finance 59, 831-868. Vuolteenaho, T., 2002, What drives firm-level stock returns? Journal of Finance 57, 233-264. Watts. R and J. Zimmerman, 1986. Positive Accounting Theory, Prentice-Hall, London. Wu, J., Zhang, L., and Zhang, F., 2008 The q-theory approach to understanding the accrual anomaly. Journal of Accounting Research, Forthcoming. Xie, H. 2001. The mispricing of abnormal accruals. The Accounting Review 76, 357373. Zach, T., 2005. Evaluating the accrual-fixation hypothesis as an explanation for the accrual anomaly. Washington University working paper. Zhang, F., 2007. Accruals, investment, and the accruals anomaly. The Accounting Review 82, 1333-1364. Zhang, L. 2005. The value premium. Journal of Finance 60, 67-103. Zhang, X., 2006, Information uncertainty and analyst forecast behavior. Contemporary Accounting Research 23, 56590.

105

2.5

log (Cumulative Total Return)

1.5

0.5

0 197301 197501 197701 197901 198101 198301 198501 198701 198901 199101 199301 199501 199701 199901 200101 200301 200501 200701 Month Excluding Transaction Costs Including Transaction Costs

Figure 1: Cumulative log total returns to a full risk model based NOA characteristic portfolio for US data (1973-2008). The blue line reflects the portfolio that is ignorant of transaction costs and the red line the portfolio that is transaction cost aware.
106

Figure 2: What is the impact of risk and transaction cost forecasting?

Panel A: Correlations (excl. transaction costs)
1 1

Panel B: Correlations (incl. transaction costs)

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0 197302

197602

197902

198202

198502

198802

199102

199402 Spearman

199702

200002

200302

200602

0 197303

197603

197903

198203

198503

198803

199103

199403

199703

200003

200303

200603

Pearson

Pearson

Spearman

Panel C: Scatter plot (excl. transaction costs)

0.06

Panel D: Scatter plot (incl. transaction costs)

0.05

0.04 0.04 0.03

0.02 0.02
dNOA (portfolio weight) dNOA (portfolio weight)

0.01

0 -4 -3 -2 -1 0 1 2 3

0 -4 -3 -2 -1 -0.01 0 1 2 3

-0.02 -0.02

-0.03 -0.04 -0.04

-0.06 dNOA (standardized measure)

-0.05 dNOA (standardized measure)

107

0.1 0 0 -0.1 -0.2

Cumulative Return

10

13

14

15

16

17

20

21

22

23

24

27

28

29

30

31

-0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 Day in August 2007

Figure 3: Daily returns to characteristic portfolios built on dNOA for a set of large capitalization US equity securities for the month of August 2007.
108

Figure 4: Graphical representation of linkages between board director members in Australia (ASX 200 companies)

109

Figure 5: Graphical representation of linkages between Japanese firms as of 2008 based on cross-equity ownership.
110

Table 1 Summary of Results of Questionaire/Survey of Investment Professionals and Academics Opinions on Academic Research on Fundamental Analysis and Equity Market Anomalies The samples consist of (i) 201 practitioner responses to the questionnaire, and (ii) 63 academic responses to the questionnaire. The academic response numbers for each question is listed below each sub-table. The differences across the sample mean for each answer is calculated using a chi-square test of populations of unequal size and unequal variance. The p-values are adjusted using Cochran-Coxs approximation of the degrees of freedom for the unmatched samples. Q1: Which risk model is most appropriate for risk calibration of an equity trading strategy? Practitioner Opinions 35% 24% Academic Opinions 7% ** 22%

CAPM with size & industry adjustments Fama-French 3-factor model (Market, Size, Book Value/Market Value) Multifactor model 12% 4% ** Other model 11% 15% CAPM 10% 4% * Fama-French 3-factor model plus other factors 5% 33% ** CAPM with size adjustments 4% 15% ** * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (61 academic responses). Q2: Which risk model(s) have you used in the last 12 months for risk calibration of an equity trading strategy? Practitioner Opinions 29% 10% 16% 8% 23% 4% 11%

CAPM with size & industry adjustments Fama-French 3-factor model (Market, Size, Book Value/Market Value) Multifactor model Other model CAPM Fama-French 3-factor model plus other factors CAPM with size adjustments

Q3: What effect do you think the current financial market will have on the use and/or demand for each of the following? (Practitioner BLACK / Academic RED UNDERLINE) Increase No Decrease use/demand effect use/demand Practitioner demand for new academic research 69% 22% 9% on fundamental analysis/anomalies 55% * 37% * 8% Risk models used in investment management 68% 17% 14% (general) 68% 24% 8% Techniques used in fundamental analysis and 55% 31% 14% quant fund management 62% 34% 4% ** Risk models used in investment management 58% 19% 23% (quant funds specifically) 60% 29% 21% PhDs in quant fund management 26% 38% 37% 27% 49% 24% * * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (63 academic responses). Q4: For the following equity trading strategies, please indicate how successful each has been over the past decade. (Practitioner BLACK / Academic RED UNDERLINE) Successful Neutral Unsuccessful 61% 28% 11% 22% ** 65% ** 13% Value strategies (for example, book value 56% 28% 16% multiples) 52% 35% 13% Growth strategies (for example, earnings 57% 25% 18% growth) 22% ** 52% ** 26% Return momentum 47% 35% 18% 70% ** 26% 4% ** Misreaction to earnings announcements or 40% 41% 18% management forecasts 52% 44% 4% ** Accounting quality (for example, accruals 41% 37% 23% anomaly) 74%** 22% ** 4% ** Misreaction to analyst forecasts 29% 45% 26% 35% 70% ** 5% ** * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (57 academic responses). Earnings or cash flow momentum

112

Q5: For the following equity trading strategies, how frequently each will be used over the next 5 years. (Practitioner BLACK / Academic RED UNDERLINE) Frequently Infrequently Never 70% 27% 3% 46% ** 54% ** 0% * Value strategies (for example, book value 58% 39% 3% multiples) 83% ** 13% ** 4% Growth strategies (for example, earnings 53% 39% 7% growth) 30% ** 61% ** 9% Return momentum 48% 48% 4% 74% ** 22% ** 4% Misreaction to earnings announcements or 37% 59% 3% management forecasts 71% ** 29% ** 0% * Accounting quality (for example, accruals 37% 57% 5% anomaly) 71% ** 29% ** 0% ** Misreaction to analyst forecasts 32% 59% 9% 50% ** 46% * 4% * * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (55 academic responses). Earnings or cash flow momentum Q6: Over the last 12 months, how often have you used the following valuation techniques in your work? (Practitioner BLACK / Academic RED UNDERLINE) Never Earning multiples 3% 13% ** Book value multiples 7% 12% Cash flow multiples 8% 46% ** Discounted free cash flow model 14% 4% ** Discounted dividend model 31% 29% Residual income (Economic Profit) model 38% 12% ** Other multiples 50% 36% * Other valuation models 52% 33% ** * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (60 academic responses). Frequently Infrequently 74% 23% 54% ** 33% 52% 41% 38% * 50% 53% 39% 25% ** 29% 59% 28% 58% 38% 26% 43% 21% 50% 16% 46% 71% ** 17% ** 26% 25% 23% 41% * 25% 22% 29% 38% **

113

Q7: How frequently do you read or reference the following academic and practitioner research for your work? (Practitioner BLACK / Academic RED UNDERLINE) Regularly Sometimes Never 10% 44% 46% 83% ** 14% ** 3% ** Journal of Financial and Quantitative Analysis 5% 32% 63% 8% 67% ** 25% ** Journal of Financial Economics 6% 26% 68% 72% ** 28% 0% ** Review of Financial Studies 6% 23% 71% 51% ** 45% ** 4% ** Journal of Banking and Finance 6% 23% 71% 0% ** 32% 68% Journal of Accounting and Economics 3% 16% 81% 88% ** 8% 4% ** Contemporary Accounting Research 0% 18% 82% 48% ** 48% ** 4% ** The Accounting Review 1% 14% 85% 92% ** 8% 0% ** Journal of Accounting Research 1% 11% 88% 92% ** 8% 0% ** CFA Magazine 48% 44% 8% 0% ** 32% 68% ** Financial Analysts Journal 49% 37% 14% 24% ** 60% ** 16% CFA Institute Conference Proceedings 28% 46% 26% Quarterly 0% ** 17% ** 83% ** Journal of Portfolio Management 12% 34% 54% 4% ** 44% 52% Journal of Investment Management 6% 31% 63% 0% ** 11% ** 89% ** Journal of Investing 3% 28% 69% 0% * 16% ** 84% * Journal of Fixed Income 2% 22% 76% 0% 8% ** 92% ** Journal of Applied Corporate Finance 2% 18% 80% 4% 72% ** 24% ** European Financial Management 1% 10% 89% 0% 7% 93% * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (63 academic responses). Journal of Finance

114

Q8: How frequently do you read or reference the following new, unpublished academic research for your work? (Practitioner BLACK / Academic RED UNDERLINE) Regularly Sometimes Never Papers posted on specific university department 7% 39% 54% or business school websites 20% ** 52% 28% ** Papers posted on specific faculty member or 6% 38% 56% researcher websites 21% ** 71% ** 8% ** Papers posted on "Social Sciences Research 9% 18% 73% Network" (http://www.ssrn.com) 96% ** 6% ** 0% ** Papers posted on "EconPapers" 4% 12% 84% (http://econpapers.repec.org) 16% ** 41% ** 43% ** * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (63 academic responses). Q9: How important is it that future academic research on fundamental analysis and anomalies focus on the following? (Practitioner BLACK / Academic RED UNDERLINE) Not Important Empirical tests of investor behavior 62% 22% 16% 92% ** 8% ** 0% ** Empirical research on forecasting firm and 59% 25% 16% industry fundamentals 75% * 15% 10% Empirical tests of asset pricing, risk, and factor 62% 20% 18% models 52% 44% ** 4% ** Empirical discovery and investigation of new 55% 25% 19% "anomalies" or signals 55% 36% 9% ** Theoretical models of investor behavior 51% 29% 20% 68% * 28% 4% ** Empirical implementation of trading strategies 51% 26% 23% 63% 28% 9% ** Theoretical asset pricing, risk, and factor models 44% 31% 25% 56% 37% 7% ** Theoretical models of trading strategies 27% 43% 30% 48% ** 48% 4% ** * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (59 academic responses). Important Neutral

115

Q10: Overall, do academic research studies about anomalies/trading strategies have appropriate emphasis on: (Practitioner BLACK / Academic RED) Not Enough Emphasis Theoretical foundations of a strategy? 21% 61% 18% 4%** 40% ** 56% ** Empirical tests of a strategy? 11% 64% 25% 24% ** 40% ** 36% Possible (alternative) risk-based explanations? 7% 54% 38% 24% ** 40% * 36% Potential market impact of executing a 10% 44% 46% strategy? 4% 28% * 68% ** Economic/psych. origins of an anomaly that 8% 43% 50% leads to a strategy? 3% * 65% ** 32% * Real world transactions & trading costs for a 9% 41% 50% strategy? 5% 36% 59% Applicability of strategy to other markets 4% 45% 51% (countries/types of markets)? 12% ** 47% 41% * and ** indicate difference in means across practitioner and academic sample answers are significant at 5% and 1% levels, respectively (59 academic responses). Too Much Appropriate Emphasis Emphasis

116

Table 2: Returns to Accrual Anomaly Through Time Panel A: Monthly returns (ignoring transaction costs) 1973-2008 #Average Std. Dev. Test statistic H0: Average = 0 Annualized Sharpe Ratio 7.98 4.51 4.14 5.05 2.23 7.90 0.0126 0.0328 1970s 0.0163 0.0330 1980s 0.0128 0.0339 1990s 0.0151 0.0327 2000s 0.0067 0.0310 1973-1999 0.0146 0.0331

1.33

1.72

1.31

1.60

0.75

1.52

Panel B: Monthly returns (including transaction costs) 1973-2008 Average Std. Dev. Test statistic H0: Average = 0 Annualized Sharpe Ratio 9.21 4.84 4.16 6.35 3.12 8.79 0.0139 0.0314 1970s 0.0170 0.0319 1980s 0.0128 0.0337 1990s 0.0177 0.0306 2000s 0.0086 0.0286 1973-2000 0.0157 0.0321

1.54

1.84

1.32

2.01

1.05

1.69

117

Table 2 - Continued Panel C: Regression of accrual anomaly decline (ignoring transaction costs) RdNOAt = + TIMETIMEt + TIME_SQTIME2t + t Coefficient T-statistic Coefficient T-statistic 0.0176 5.59 0.0121 2.56 TIME -0.0023 -1.84 0.0053 1.05 -0.0018 -1.56 0.0089 TIME_SQ Adjusted R2 0.0056

Panel D: Regression of accrual anomaly decline (including transaction costs) RdNOAt = + TIMETIMEt + TIME_SQTIME2t + t Coefficient T-statistic Coefficient T-statistic 0.0177 5.85 0.0120 2.64 TIME -0.0018 -1.44 0.0062 1.27 -0.0018 -1.68 0.0067 TIME_SQ Adjusted R2 0.0025

Portfolios are formed from the 1,000 US largest securities (as measured by market capitalization) from February 1973 to December 2008. Portfolios are rebalanced monthly to achieve a target annualized risk of 10 percent. Portfolios are fully invested with individual positions limited to be no more than 5% of the total portfolio.

118

Table 3: Ex Post Return Analysis of Accrual Anomaly Panel A: Explaining returns to accruals with known characteristics (ignoring transaction costs) RdNOAt = + MKTRMKTt + SIZERSIZEt + B/PRB/Pt + MOMRMOMt + t Coefficient T-statistic Coefficient T-statistic 0.0125 7.48 0.0112 6.53 MKT 0.0056 0.16 0.0047 0.14 -0.0377 -0.82 0.1129 2.50 0.1210 3.12 0.0269 SIZE B/P MOM Adjusted R2 -0.0023

Panel B: Explaining returns to accruals with known characteristics (including transaction costs) RdNOAt = + MKTRMKTt + SIZERSIZEt + B/PRB/Pt + MOMRMOMt + t Coefficient T-statistic Coefficient T-statistic 0.0141 8.78 0.0119 7.14 MKT -0.0082 -0.25 -0.0143 -0.45 -0.0830 -1.91 0.1690 3.73 0.1223 3.22 0.0513 SIZE B/P MOM Adjusted R2 -0.0022

Panel C: Fama-Macbeth characteristic regressions XRETi,t+k = + NOAi,t+ i,t+k Number of months ahead k=1 FamaMacbeth T-Stat 0.0126 k=3 0.0119 k=6 0.0094 k=9 0.0066 k=12 0.0068

7.88

7.33

5.53

4.09

4.30

119

RNOA is the monthly return to the NOA characteristic portfolio. NOA is measured as in section 5.1. RMOM is the monthly return to the MOM characteristic portfolio. MOM is measured as the total equity return over the previous 12 months ignoring the most recent month. RB/P is the monthly return to the B/P characteristic portfolio. B/P is measured as the ratio of common equity to market capitalization. RSIZE is the monthly return to the SIZE characteristic portfolio. SIZE is measured as the average of the log of total assets and the log of market capitalization as per the BARRA USE3 risk model. All of the characteristic portfolios are formed with a 10% annualized risk target. XRET is the excess return. It is the regression residual from the first stage of the full risk model estimation. Specifically, total returns are projected onto a set of 13 characteristics and 55 industries. The residual from this weighted least squares regression is the excess return.

120

Table 4: Ex Post Return Analysis (Fama-French) of Accrual Anomaly Panel A: Characteristic portfolio returns (ignoring transaction costs) RdNOAt = + MKTRMKTt + SMBSMBt + HMLHMLt + t Coefficient T-statistic 0.0126 7.72 MKT 0.0056 0.15 SMB 0.0050 0.10 HML -0.0004 -0.01 Adjusted R2 -0.0069

Panel B: Characteristic portfolio returns (including transaction costs) RdNOAt = + MKTRMKTt + SMBSMBt + HMLHMLt + t Coefficient T-statistic 0.0136 8.71 MKT 0.0150 0.41 SMB 0.0017 0.03 HML 0.0703 1.26 Adjusted R2 -0.0032

Panel C: Equal weighted zero-cost investment (extreme deciles) RdNOAt = + MKTRMKTt + SMBSMBt + HMLHMLt + t Coefficient T-statistic 0.0149 3.52 MKT 0.1615 1.62 SMB 0.0864 0.64 HML 0.1187 0.78 Adjusted R2 0.0012

Panel D: Value weighted zero-cost investment (extreme deciles) RdNOAt = + MKTRMKTt + SMBSMBt + HMLHMLt + t Coefficient T-statistic 0.0079 3.92 MKT 0.0277 0.58 SMB -0.0648 -1.01 HML 0.0422 0.58 Adjusted R2 -0.0032

121