Analysts forcards and stock returns

Analyst Forecasts and Stock Returns* James S Ang Florida State University Stephen J Ciccone University of New Hampshire Abstract This study seeks to determine the relation between stock returns and analyst forecast properties, specifically, the dispersion and error of annual earnings forecasts The results of portfolio sorts, Fama-MacBeth cross-sectional regression models, and Fama and French (1993) factor models indicate firms with low dispersion or error outperform firms with high dispersion or error Robustness tests show the results are not explained by liquidity, momentum, industry, post-earnings announcement drift, or traditional risk measures An investment strategy based on forecast properties is shown to produce zero-cost returns of 13% per year, yielding positive returns in all 19 years using an error measure This paper can be downloaded from the Social Science Research Network Electronic Paper Collection: http://papers.ssrn.com/paper.taf?abstract_id=271713 _ * This paper is taken from Ciccone’s dissertation The authors thank Gary A Benesh, William A Christiansen, Christopher Gadarowski, Elton Scott, Thomas W Zuehlke and seminar participants at Florida State University, the University of New Hampshire, the Financial Management Association Doctoral Student Seminar, and the Northern Finance Association Meeting for providing valuable comments The data was provided by the Institutional Brokers Estimate System, Inc (IBES) Correspondence can be sent to Stephen Ciccone, Whittemore School of Business and Economics, 15 College Road, McConnell Hall, Durham, NH 03824-3593; phone: (603) 862-3343; email; sciccone@cisunix.unh.edu Analyst Forecast Properties and Stock Returns Abstract This study seeks to determine the relation between stock returns and analyst forecast properties, specifically, the dispersion and error of annual earnings forecasts The results of portfolio sorts, Fama-MacBeth cross-sectional regression models, and Fama and French (1993) factor models indicate firms with low dispersion or error outperform firms with high dispersion or error Robustness tests show the results are not explained by liquidity, momentum, industry, post-earnings announcement drift, or traditional risk measures An investment strategy based on forecast properties is shown to produce zero-cost returns of 13% per year, yielding positive returns in all 19 years using an error measure I Introduction Analyst earnings forecast properties have been extensively studied in the finance and accounting literature Research has found analyst forecasts or recommendations to be informative, valuable, and, in the case of the forecasts, accurate relative to time-series earnings models.1 Because of these characteristics, not only are determinants of analyst forecasts considered an important research topic, but also analyst forecasts are often used empirically to proxy for information quality, investor beliefs, expected growth rates, or disparity of opinion However, there are limitations to analyst forecasts Research has uncovered optimistic biases, irrationality, under/over reaction to earning changes and “herding” behavior in the forecasts.2 Analyst forecasts have also been linked to financial distress and earnings volatility (e.g., Brown (1998), Das, Levine, and Sivaramakrishnan (1998), Ciccone (2001)) The potential qualities, whether good or bad, of analyst earnings forecasts suggest possible relations to stock returns For example, if analyst forecast properties represent information quality, then stock returns may differ between firms with high or low information quality Low information firms may have greater risk, which should translate into greater returns.3 Contrarily, firms generating greater amounts of information might cause more volatile stock prices as prices are updated more frequently to reflect the new information (e.g., Veronesi (2000)) Although the above contentions represent theories consistent with market efficiency, a contrarian-value strategy based on information quality may potentially exist Down-market investors might exhibit a “flight-to-quality” and buy firms with high quality information environments and avoid or sell firms with low quality information environments Thus, high information firms would 11 See, for example, Brown and Rozeff (1978), Crichfield, Dyckman, Lakonishok (1978), Givoly and Lakonishok (1979), Fried and Givoly (1982), and Lys and Sohn (1990) See, for example, O’Brien (1988), DeBondt and Thaler (1990), Butler and Lang (1991), Abarbanell and Bernard (1992), Trueman (1994)), and Welch (2000) A related line of research is based on estimation risk, the risk from uncertain return parameters such as beta (e.g., Barry and Brown (1985), Clarkson, Guedes, and Thompson (1996)) Low information firms are believed to have greater estimation risk be superior performers in down-markets while low information firms would be superior performers in up-markets While information risk arguments represent one line of reasoning for a relation between analyst forecast properties and stock returns, the well-known biases in the forecasts may result in different stock return influences For example, if analysts are systematically optimistic regarding certain types of firms, then these firms may consistently disappoint investors resulting in persistently lower returns (consistent with Lakonishok, Shleifer, and Vishny (1994) theory on low returns to growth stocks) Potential biases are also consistent with the optimism framework of Miller (1977) and the investor psychology models of Barberis, Schleifer, and Vishny (1998) and Daniel, Hirshleifer and Subrahmanyam (1998) As it is known empirically that firms having highly dispersed forecasts tend to have forecasts that are optimistic (e.g., Ciccone (2001)), firms with high dispersion should be the firms disappointing investors Thus, these firms should have lower returns Ackert and Athanassakos (1997) and Scherbina (2001) find empirical support for this theory, but these studies rely mainly on the relation of stock returns to dispersion, rather than to a direct optimism measure However, if this theory is indeed true, forecast optimism should explain returns, subsuming most of the relation between dispersion and returns The purpose of this study is to examine the relation between analyst forecast properties and stock returns The primary motivation is to establish whether the forecast properties have predictive abilities Attempts are made, however, to attribute the results to an acceptable theory The forecast properties used are the dispersion and the error of analyst annual earnings forecasts, measured at a period prior to the return period The main testing uses portfolio sorts and two commonly applied empirical models: a Fama and MacBeth (1973) cross-sectional regression model and a Fama and French (1993) factor model Dispersion and error are used in conjunction with market, size, and book-to-market variables to determine any relation to stock returns For convenience, the term “transparency” is used throughout this research This terminology is consistent with an interpretation that firms with lower dispersion or error are more easily understood financially or are more transparent to investors Thus, a firm is called “transparent” if it has low forecast dispersion or low forecast error A firm is called “opaque” if it has high forecast dispersion or high forecast error The portfolio sorts and Fama and MacBeth (1973) cross-sectional regressions show that the stock returns of transparent firms outperform the stock returns of opaque firms This “transparency component” of stock returns is separate from the size and book-to-market components Additional robustness tests show the transparency component of stock returns occurs independently of industry, liquidity, momentum, and traditional risk measures The component is also not explained by post-earnings announcement drift A zero-cost trading rule based on transparency earns an average return of about 13% per year, about twice the magnitude of the average returns earned by similar trading rules using size and book-to-market ratio The rule yields positive returns in 17 of 19 years using a dispersion measure and in all 19 years using an error measure Transparency also appears to be a systematic risk factor, similar to size and book-to-market ratio Zero-cost Fama and French (1993) factors designed to mimic the risk arising from the analyst forecast properties have large, significant returns Furthermore, these factors are important in explaining variations in stock returns; the factors have large standard deviations and are significant in time series regressions The results are not attributable to the common theories regarding analyst forecasts Information risk theories in which low information firms have greater risk are immediately eliminated as the high information firms outperform the low information firms Volatility theories are also eliminated as the results occur independently of traditional volatility measures (beta and return volatility) Flight to quality theories are shown to be invalid as high information firms outperform low information firms in both up and down markets The results not appear attributable to behavioral theories based on investor under/over- reaction as forecast optimism only seems to explain a small portion of the results, about 2% per year Thus, the reasons behind the transparency component of stock returns remain elusive The findings herein have important implications as additional evidence of non-risk related stock return predictors is produced Furthermore, although the anomaly may be related to behavioral theory, most present theory relies on investor optimistic biases Perhaps a new framework of investor behavior needs to be developed The rest of this paper is organized as follows Section II describes the methodology used Section III presents and discusses the results Section IV concludes II Methodology In this section, the transparency measures and asset pricing models used for testing are explained The asset pricing models include Fama and MacBeth (1973) (hereafter, Fama-MacBeth) cross-sectional models and Fama and French (1993) time series models A pictorial representation of the variable calculation dates is also provided A Transparency Measures Two measures are constructed from analyst earnings forecast properties: forecast dispersion and forecast error The dispersion is measured every fiscal year for each sample firm using annual earnings forecasts The dispersion is computed as the standard deviation of all individual forecasts available for a firm in the last fiscal month of the corresponding fiscal year.4 A firm must have at least two individual forecasts for this measure to be computed Like the dispersion measure, the error is measured every fiscal year for each sample firm using annual earnings forecasts The forecast error is computed as the absolute value of the difference between the actual earnings and the mean of the individual earnings forecasts made in the last fiscal month of the corresponding fiscal year One forecast is required for this measure to be computed Degeorge, Patel, and Zeckhauser (1999) use this error measure in their study of earnings management.5 These forecasts are shown to be more accurate than earlier forecasts (e.g., Crichfield, Dyckman, and Lakonishok (1978) and O’Brien (1988)) and using forecasts as of the same date standardizes any biases Using these “unscaled” measures may result in biases if firms with higher absolute values of earnings necessarily have higher dispersion and error measures A quick check is performed by sorting firms into five portfolios based on the absolute value of actual earnings and then sorting each of these five portfolios into five An important aspect of this research is the relation between the transparency measures and forecast optimism A forecast is defined as optimistic when the mean annual earnings forecast is greater than the corresponding actual earnings concurrently with the dispersion and error Optimism is computed at fiscal year end, Thus, optimism surrounding a particular stock is determined before the stock returns are computed This may present difficulties if concurrent optimism is more relevant in the testing of behavioral models However, as the primary purpose of this study is to find stock return predictors, optimism must be calculated ex ante The I/B/E/S United States (IBES) summary database is used to gather the earnings forecast data This database covers the period 1976 through 1997 and contains mean analyst earnings forecasts, the forecast standard deviation, the actual earnings, and the number of analysts The IBES data is matched to the Center for Research in Security Prices (CRSP) database to obtain stock return data and to the COMPUSTAT database to obtain annual accounting data COMPUSTAT data begins in 1977.6 The samples not necessarily reflect an intersection of the three databases Most return periods cover 1978 through 1996 B Fama-MacBeth Regressions The Fama-MacBeth cross-sectional model is reproduced but with the added dispersion and error terms This model uses monthly returns as the dependent variable and individual firm characteristics as the independent variables Firms are assigned the characteristics at the beginning of the return period and then monthly returns are measured over the subsequent year The regression is performed each month The time-series averages of the slopes from the month-by-month regressions are divided by their time-series standard errors to compute t-values more portfolios based on either dispersion or error (see Table A1 in the appendix) The dispersion and error measures remain quite constant across the actual earnings quintiles with the exception of firms in the highest actual earnings quintile Thus, due to the limited amount of firms with this potential bias and due to the results being similar for the firms unaffected by the bias, the unscaled measures are not considered problematic The COMPUSTAT survivorship bias is not considered a problem because all the sample years, except for one, occur after 1978 expansion Fama and French (1992) show that the CAPM beta does not predict as well as size or bookto-market Based on these results, the dispersion and error are used with size and book-to-market The size (price times shares) is computed at the end of June The book-to-market is computed using the market value at the end of the previous December and the book value at the end of the previous fiscal year Stocks with negative book-to-market values are not included in the sample The transparency measures are computed at the end of the previous fiscal year The natural logarithm of the characteristics is taken for use in the regressions Returns are measured from July of year t to June 30 of year t+1 C Fama and French (1993) Factor Model The factor model is motivated by Fama and French’s (1993) study using variables thought to proxy for size and book-to-market factors: SMB (small minus big) and HML (high minus low), respectively These factors are created by sorting stocks into portfolios based on the underlying characteristics and computing the difference between the extreme portfolio means These two factors and a market factor capture stock return variation in their sample Further research by Fama and French (1996) shows that the three-factor model is robust to a variety of portfolio formations and can explain most stock return anomalies The asset model constructed herein includes the Fama and French (1993) factors plus an added factor constructed using the transparency measures At the end of June of year t, two portfolios are formed based on size, three based on book-to-market, and three based on each of the transparency measures.7 For each characteristic (size, book-to-market, or transparency), the portfolios are formed independently of the other characteristics The difference between the two Consistent with Fama and French (1993), size is defined as the per share price times the shares outstanding Book value is defined as the book value of stockholders’ equity plus balance-sheet deferred taxes and investment tax credits minus the book value of preferred stock The market value of each individual firm is computed on the same date as the portfolio formation (the end of June of year t) for the size portfolios but at the end of December of year t-1 for the book-to-market portfolios The transparency measures and book values are computed as of fiscal year end t-1 Negative book-to-market stocks are not used in the portfolio formation extreme groups’ equal-weighted return is computed each month.8 These monthly differences serve as the factors used in a time-series regression equation Consistent with Fama and French (1993) terminology, the factors are called SMB, HML, and TMO (transparent minus opaque) A market factor, RM – RF is also used, where RM is the value-weighted market return and RF is the T-bill rate The dependent variables are computed by grouping the stock sample into portfolios based on their ranking of certain characteristics that may predict stock return variation Similar to Fama and French (1993), 25 portfolios are formed from rankings of the transparency measures and the book-to-market ratio Only New York Stock Exchange breakpoints are used to allocate the stocks to the portfolios The value-weighted monthly excess returns of the portfolios are computed from July of year t to June of year t+1 These monthly returns, Rt, less the T-bill rate, RFt, serve as the dependent variables The empirical regression equation is as follows: Rt – RFt = a + b[RMt - RFt] + sSMBt + hHMLt + mTMOt + εt (1) Additional testing evaluates the factor means and standard deviations in a manner similar to Chan, Karceski, and Lakonishok (1998) D Pictorial Representation of Variable Calculations Because the variables are computed on several dates, the following pictorial representation is used to facilitate understanding The time line shows certain key dates and, beneath the dates, the variables that are computed on those dates The fiscal year end in year t-1 is often the same as the calendar year end in year t-1, December 31 An important note is that all the variables are computed prior to the return period Fama and French (1993) use value-weighted portfolio returns However, to represent a simple zero-cost investment strategy, equally-weighted portfolio returns are used herein, similar to Chan, Karceski, and Lakonishok (1998) The portfolio returns used as dependent variables in the time-series regressions are valueweighted The results are qualitatively and quantitatively similar if equally-weighted portfolio returns are used as the dependent variables  Fiscal year end Calendar year end June 30 July June 30 year t-1 December 31 year t year t year t+1 year t-1 dispersion, error book value optimism market value (for book-to-market) market value (for size) | return period | III Results The first series of tests examines the relationship between transparency and stock returns using portfolio sorts and Fama-MacBeth cross sectional regressions while controlling for size and book-to-market ratio After the transparency component is established by the procedures described above, the next series of tests examines a transparency-related factor, TMO This factor is analyzed in a manner similar to and along with the previous factors of Fama and French (1993), RM-RF, SMB, and HML The results are then shown to be unrelated to other existing anomalies Finally, a trading rule is established A Portfolio Sorts Table presents the equally-weighted raw returns for dispersion and error portfolios.9 An interesting pattern emerges Contrary to most risk theory, transparent firms outperform opaque firms by an average of 0.86% per month when using either dispersion or error This amount is not only large, but also appears to occur independently of the size and book-to-market effect The transparent firms, although they have the highest returns, tend to be large, growth firms Further tests, shown in Table 2, directly control for size and book-to-market by using twodimensional portfolios sorts These results corroborate the previous findings: the transparency component of stock returns not only exists independently of the size and book-to-market components but also appears larger than those components The average difference between the transparency and opaque portfolios is around 0.90% per month The affect of size on the returns is For this initial portfolio sort and for the Fama and French (1993) factor model sorts, NYSE deciles are used for breakpoints However, as many of the smallest stocks on Nasdaq are not included in the IBES database, the use of this procedure is not necessary Thus, in this study, most sorts using dispersion or error not use the NYSE breakpoints 10 also includes the period after 1992, when Brown (1998) finds that managers dramatically increased both their tendency to warn analysts of impending losses and to report profits that meet or slightly beat analyst estimates The results are similar when stocks are sorted by error The transparent firms outperform the opaque firms by an average of 13.12% per year Perhaps most striking is that positive returns are earned in every sample year A similar trading rule based on buying small stocks and selling short large stocks over the same time period earns 5.41% per year and generates positive returns in only of the 19 years A similar rule based on buying value stocks and selling short growth stocks yields 6.72% per year, generating positive returns in 11 of the 19 years IV Conclusions Much controversy surrounds the use of analyst forecasts This paper seeks to provide an indication of the relation between analyst forecast properties and stock returns Using portfolio sorts, Fama-MacBeth cross-sectional regressions, and Fama and French (1993) factor models, the results indicate that firms with lower dispersion or error have greater stock returns This analyst forecast component of stock returns is not related to size or book-to-market, but is a separate predictor of stock returns The component is not related to industry, liquidity, momentum, postearnings announcement drift, or traditional risk measures Optimistic forecasts only explain a small portion of the component Present theories of analyst forecasts are not adequate to explain this anomaly The anomaly is not related to risk-based theory for two reasons: 1) the seemingly riskier firms have lower stock returns and 2) the higher returns are not explained by higher volatility Furthermore, the results are not attributable to contrarian-value strategies as there appears to be no flight to quality Firms with low dispersion or error outperform regardless of market conditions Finally, optimism only explains a small part of the results ruling out most optimistic behavioral theories 20 References Abarbanell, J S and Bernard, V L 1992 Tests of analysts’ overreaction/underreaction to earnings information as an explanation for anomalous stock price behavior Journal of Finance 47: 11811207 Ackert, L., and Athanassakos, G 1997 Prior uncertainty, analyst bias, and subsequent abnormal returns Journal of Financial Research 20: 263-273 Ball, R and Brown, P 1968 An empirical evaluation of accounting income numbers, Journal of Accounting Research 6: 159-178 Barberis, N., Shleifer, A., and Vishny, R 1998 A model of investor sentiment Journal of Financial Economics 49: 307-343 Barry, C B and Brown, S J 1985 Differential information and security market equilibrium Journal of Financial and Quantitative Analysis 20: 407-422 Beaver, W H., Clarke, R., and Wright, W F 1979 The association between unsystematic security returns and the magnitude of earnings forecast errors Journal of Accounting Research 17: 316-340 Bernard, V L and Thomas, J K 1990 Evidence that stock prices not fully reflect the implications of current earnings for future earnings Journal of Accounting and Economics 13: 305340 Brown, L D 1998 Managerial behavior and the bias in analysts’ earnings forecasts, Working Paper Atlanta: Georgia State University Brown, L D and Rozeff, M S 1978 The superiority of analyst forecasts as measures of expectations: Evidence from earnings Journal of Finance 33: 1-16 Butler, K C and Lang, L H 1991 The forecast accuracy of individual analysts: Evidence of systematic optimism and pessimism Journal of Accounting Research 29: 150-156 Chan, L., Karceski, J and Lakonishok, J 1998 The risk and return from factors Journal of Financial and Quantitative Analysis 33: 159-188 Ciccone, S J 2001 Analyst forecast properties, financial distress, and business risk, Working paper Durham: University of New Hampshire Clarkson, P., Guedes, J and Thompson, R 1996 On the diversification, observability, and measurement of estimation risk Journal of Financial and Quantitative Analysis 31: 69-84 Crichfield, T., Dyckman, T., and Lakonishok, J 1978 An evaluation of security analysts’ forecasts Accounting Review 53: 651-668 Daniel, K., Hirshleifer, D., Subrahmanyam, A 1998 Investor psychology and security market underand overreactions Journal of Finance 53: 1839-1886 Das, S., Levine, C., and Sivaramakrishnan, K 1998 Earnings predictability and bias in analysts’ earnings forecasts Accounting Review 73: 277-294 21 DeBondt, W F M., and Thaler, R H 1990 Do security analysts overreact? American Economic Review 80: 52-57 Degeorge, F., Patel, J and Zeckhauser, R 1999 Earnings management to exceed thresholds Journal of Business 72: 1-33 Fama, E F and French, K R 1992 The cross-section of expected stock returns Journal of Finance 47: 427-465 Fama, E F and French, K R 1993 Common risk factors in the returns on stocks and bonds Journal of Financial Economics 33: 3-56 Fama, E F and French, K R 1996 Multifactor explanations of asset pricing anomalies Journal of Finance 51: 55-84 Fama, E F and MacBeth, J 1973 Risk, return, and equilibrium: Empirical tests Journal of Political Economy 81: 607-636 Fried, D and Givoly, D 1982 Financial analyst’s forecasts of earnings: A better surrogate for market expectations Journal of Accounting and Economics 4: 85-107 Givoly, D and Lakonishok, J 1979, The information content of financial analysts’ forecast of earnings Journal of Accounting and Economics 1: 1-21 Jegadeesh, N and Titman, S 1993 Returns to buying winners and selling losers: Implication for stock market efficiency Journal of Finance 48: 65-91 Kahle, K M and Walking, R A 1996 The impact of industry classifications on financial research Journal of Financial and Quantitative Analysis 31: 309-335 Keim, D B 1983 Size-related anomalies and stock return seasonality: Further empirical evidence Journal of Financial Economics 12: 13-32 Lakonishok, J., Shleifer, A and Vishny, R 1994 Contrarian investment, extrapolation, and risk Journal of Finance 49: 1541-1578 Lys, T and Sohn, S 1990 The association between revisions of financial analysts’ earnings forecasts and security-price changes Journal of Accounting and Economics 13: 341-363 Miller, E., 1977 Risk, uncertainty, and divergence of opinion Journal of Finance 32: 1151-1168 O’Brien, P C., 1988 Analysts’ forecasts as earnings expectations Journal of Accounting and Economics 10: 53-83 Scherbina, A 2001 Stock prices and optimism: Empirical evidence that prices reflect optimism Working paper Evanston: Northwestern University Trueman, B 1994 Analyst forecasts and herding behavior Review of Financial Studies 7: 97-124 Veronesi, P 2000 How does information quality affect stock returns? Journal of Finance 55: 807-837 Welch, I 2000 Herding among security analysts Journal of Financial Economics 58: 369-396 22 23 TABLE Summary Statistics of Stocks Sorted into Quintiles by Size, Book-to-Market, and Transparency Quintiles Mean Monthly Return Size (Millions) Book-to-Market Firm-years Transparent Dispersion 1.65 1413 0.57 7515 1.25 1306 0.70 7515 1.09 1373 0.80 7408 Error Opaque Transparent Opaque 1.04 1537 0.89 5780 0.79 1252 1.05 5513 1.71 1530 0.58 9705 1.36 1129 0.70 7798 1.26 1050 0.77 8424 1.14 1021 0.88 7937 0.85 739 1.03 7975 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1978 through 1996 At the end of June of year t, stocks are ranked by the following attributes: 1) dispersion defined as the standard deviation of analyst forecasts and 2) error defined as the absolute value of the actual earnings less the average earnings forecast The stocks are placed into portfolios based on their rankings using NYSE breakpoints Five portfolios are formed for each attribute Each attribute uses its complete sample of firms, regardless of the other attributes The transparency measures are computed as of the previous fiscal year end Equally-weighted monthly raw returns are measured from July of year t to June of year t+1 All returns are in percent 31 TABLE Stock Returns and Summary Statistics of 25 Book-to-Market/Transparency Measure Sorted Portfolios Transparent Size Small Large Difference (S-L) 1.86 1.56 1.64 1.71 1.60 0.26 Opaque Difference (T-O) Transparent Panel A: Portfolios Sorted by Size and Transparency Measures Mean Equally-Weighted Monthly Raw Return Dispersion 1.54 1.44 1.16 0.92 0.94 1.94 1.27 0.97 0.97 0.37 1.19 1.66 1.33 0.97 0.92 0.72 0.92 1.89 1.31 1.17 1.01 0.84 0.87 1.68 1.35 1.34 1.07 1.09 0.51 1.70 0.19 0.10 0.09 -0.17 0.24 Avg.=0.09 Avg.=0.89 1.49 1.33 1.31 1.40 1.34 0.15 Opaque Difference (T-O) Mean Equally-Weighted Monthly Raw Return Error 1.62 1.23 1.06 0.88 1.15 0.98 0.46 1.20 1.17 1.08 0.63 1.26 1.25 1.10 0.92 0.76 1.28 1.14 1.19 0.51 0.34 0.09 -0.13 Avg.=0.14 Avg.=0.92 Panel B: Portfolios Sorted by Book-to-Market Ratio and Transparency Measures Mean Equally-Weighted Monthly Raw Return Mean Equally-Weighted Monthly Raw Return Book-toMarket Low High Difference (H-L) 1.48 1.65 1.96 1.90 1.98 0.50 Avg.=0.61 1.20 1.27 1.55 1.70 1.58 0.38 Dispersion 0.93 1.12 1.19 1.35 1.47 0.54 0.59 1.04 1.22 1.24 1.38 0.79 0.38 0.60 0.94 0.98 1.22 0.84 1.10 1.05 1.02 0.92 0.76 1.66 1.87 1.82 1.83 1.93 0.27 Avg.=0.97 Avg.=0.45 1.50 1.38 1.68 1.53 1.81 0.31 Error 1.07 1.34 1.43 1.51 1.66 0.59 0.87 1.09 1.20 1.38 1.48 0.61 0.65 0.73 1.03 1.10 1.14 0.49 1.01 1.14 0.79 0.73 0.79 Avg.=0.89 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1978 through 1996 At the end of June of year t, stocks are ranked by the following attributes: 1) size defined as market value; 2) book-to-market defined as the book value of common equity relative to market value; 3) dispersion defined as the standard deviation of analyst forecasts; and 4) error defined as the absolute value of the actual earnings less the average earnings forecast Portfolios are sorted by either size and transparency measure (Panel A) or book-to-market ratio and transparency measure (Panel B) Equally-weighted, monthly raw returns are measured from July of year t to June of year t+1 All returns are in percent 32 TABLE Average Slopes and t-values of Monthly Cross-Sectional (Fama-MacBeth) Regressions Size -0.18 (-3.20) -0.02 (-0.28) -0.06 (-1.17) Book-to-Market Dispersion Error Average Coefficients from Monthly Cross-Sectional Regressions with t-values in Parentheses Panel A: Each independent variable is used separately in the cross-sectional regressions (univariate) 0.23 -0.24 -0.15 (2.44) (-5.70) (-4.90) Panel B: The independent variables in each cross-sectional regression are size, book-to-market, and a transparency measure (multivariate) 0.51 -0.32 (3.74) (-7.99) 0.36 (2.88) Optimism dummy -0.23 (-3.36) -0.21 (-7.24) Panel C: The independent variables in each cross-sectional regression are size, book-to-market, a transparency measure, and the optimism dummy (multivariate) -0.02 0.53 -0.32 -0.14 (-0.35) (3.87) (-8.03) (-2.62) -0.07 (-1.25) 0.36 (2.93) -0.20 (-7.15) -0.13 (-2.37) Each month, a cross-sectional regression is run using stock returns as the dependent variable and size, book-to-market, transparency measures, and optimism as the independent variables All domestic stocks listed on the NYSE, Amex, and Nasdaq during the period 1978 through 1996 from either the CRSP, Compustat, and IBES databases are selected At the end of June of year t, stocks are assigned the natural logarithm of the following attributes: 1) size defined as the market value; 2) book-to-market defined as the book value of common equity relative to market value; 3) dispersion defined as the standard deviation of analyst forecasts; and 4) error defined as the absolute value of the actual earnings less the average earnings forecast An optimism dummy variable equal to one if the mean forecast is greater than the actual earnings and equal to zero otherwise is also used The transparency measures are computed as of the previous fiscal year end The full equation is as follows: R = a + b log (size) + c log (book/market) + d log (transparency measure) + e optimism dummy The numbers below represent the average slope from the monthly cross-sectional regressions and the corresponding t-value (in parentheses) Panel A displays the results when the regressions include only one independent variable Panel B displays the results when the regressions include size, book-to-market, and one transparency measure Panel C displays the results when the regressions include size, book-to-market, one transparency measure, and the optimism dummy variable The returns are measured from July of year t to June of year t+1 The returns used as dependent variables are in percent 33 TABLE Factor RM – RF SMB HML TMO: Dispersion Error Mean Monthly Returns and Summary Statistics of Zero-Cost Portfolios All Months 0.70 0.36 0.46 t-value for All Months 2.51 1.95 2.52 Factor Standard Deviation 4.30 2.86 2.74 January 1.79 3.86 0.99 February to November 0.55 0.14 0.48 December 1.08 -0.93 -0.29 Up-Market Months 3.29 0.35 -0.54 Down-Market Months -3.33 0.37 2.07 0.70 0.67 6.39 6.18 1.70 1.69 -1.20 -1.71 0.83 0.81 1.34 1.73 0.66 0.67 0.76 0.69 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1977 through 1996 (beginning 1978 for HML) At the end of June of year t, stocks are ranked by the following attributes: 1) size defined as the market value; 2) book-to-market defined as the book value of common equity relative to market value; 3) dispersion defined as the standard deviation of analyst forecasts; 4) error defined as the absolute value of the actual earnings less the average earnings forecast The transparency measures are computed as of the previous fiscal year end The stocks are placed into portfolios based on their rankings using NYSE breakpoints Two portfolios are formed for size and three portfolios are formed for all the other variables The monthly equally-weighted mean return is computed for each portfolio from July of year t to June of year t+1 Every month factor risk premiums are computed as the difference between the high and low portfolios for each attribute SMB is computed as the small stock returns minus the large stock returns HML is computed as the high book-to-market stock returns minus the low book-to-market stock returns TMO is computed as the transparent stock returns minus the opaque stock returns A market factor, RM – RF, is also computed as the difference between the CRSP value-weighted market return and the risk-free rate An up-market month is one during which the value weighted CRSP return less the risk free rate is positive All returns are in percent 34 TABLE Time Series Regression Results Using 25 Dispersion/Book-to-Market Sorted Portfolios as Dependent Variables and Market, Size, Book-to-Market, and Transparency (Dispersion) Factors as Independent Variables Book-to-Market Quintile Transparency Quintile Low High Low High 0.71 -0.65 -0.75 0.22 1.91 t(a) -0.79 -0.04 -0.16 -0.43 0.34 -0.13 -0.01 -0.02 -0.08 0.08 -0.31 0.31 -0.02 -0.26 -0.17 0.19 -0.12 0.02 -0.29 -0.01 0.78 -0.81 -2.95 0.37 0.83 -1.27 1.54 -0.11 -1.68 -0.88 0.52 -0.55 0.09 -1.46 -0.03 1.10 1.11 1.04 0.91 1.05 1.09 0.91 0.89 1.05 1.11 1.04 1.02 1.01 1.03 1.19 29.42 22.64 21.22 14.21 13.20 30.80 26.93 19.40 19.71 16.70 t(b) 28.05 27.07 29.32 22.25 20.54 19.52 19.47 25.65 29.10 24.53 15.67 20.22 19.15 22.88 23.52 0.13 0.00 -0.14 -0.17 -0.23 0.20 -0.05 -0.16 -0.07 -0.07 0.22 -0.05 0.09 0.03 0.20 -1.95 -0.07 1.22 -2.83 -0.16 -1.67 -0.46 -1.11 -1.71 -1.92 2.69 -0.79 -3.36 -1.53 -1.11 2.52 -0.79 1.24 0.54 2.93 0.38 0.15 0.16 -0.04 -0.09 0.44 0.25 0.26 0.40 0.20 0.44 0.61 0.63 0.58 0.44 -6.67 -4.59 -2.04 -6.14 -3.49 2.24 -1.42 -2.30 -2.91 -3.65 4.62 3.10 4.33 6.50 2.63 3.89 7.05 7.03 7.60 5.07 a Transparent Opaque 0.11 -0.15 -0.63 0.10 0.28 0.06 -0.12 -0.17 0.05 0.48 Transparent Opaque 0.98 0.99 1.04 0.89 1.02 1.07 1.09 1.03 0.96 0.97 Transparent Opaque -0.09 -0.00 0.08 -0.24 -0.02 -0.08 -0.02 -0.08 -0.11 -0.15 Transparent Opaque -0.38 -0.34 -0.17 -0.66 -0.46 0.13 -0.10 -0.21 -0.24 -0.36 b s t(s) h 2.50 0.08 -2.95 -3.13 -3.31 t(h) 35 5.71 2.21 2.65 -0.61 -0.98 TABLE (cont.) Book-to-Market Quintile Transparency Quintile Low High Low m Transparent Opaque 0.42 0.29 0.37 -0.61 -1.33 0.36 0.29 -0.24 -0.26 -0.83 Transparent Opaque 0.89 0.82 0.78 0.70 0.65 0.86 0.84 0.74 0.76 0.72 High 6.61 2.45 -0.81 -2.11 -5.43 4.49 0.31 -0.35 1.04 -5.16 3.47 2.49 1.31 -0.15 -2.32 t(m) 0.63 0.24 -0.07 -0.21 -0.67 0.60 0.04 -0.03 0.09 -0.56 0.56 0.30 0.17 -0.02 -0.28 0.81 0.82 0.84 0.77 0.76 0.67 0.68 0.79 0.82 0.79 0.56 0.66 0.63 0.72 0.76 5.19 2.76 3.14 -4.00 -7.11 4.23 2.98 -1.85 -2.23 -5.92 R2 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1978 through 1996 Time series regressions are run using the excess returns on 25 dispersion and book-to-market portfolios as the dependent variables and market, size, book-to-market, and transparency factors as the independent variables Stocks are placed into the 25 portfolios based on their dispersion and book-to-market ranking using NYSE breakpoints Excess returns are computed as the monthly value-weighted portfolio return less the T-bill rate The market factor, RM–RF, is computed as the value weighted CRSP return less the T-bill rate The size, book-to-market, and transparency factors are computed by assigning stocks to portfolios based on their size, book-to-market, or transparency ranking (as applicable) using NYSE breakpoints Two portfolios are formed for size and three portfolios are formed for book-to-market and transparency The monthly equallyweighted mean return is computed for each portfolio from July of year t to June of year t+1 Every month factor risk premiums are computed as the difference between the high and low portfolios for each attribute SMB is computed as the small stock returns minus the large stock returns HML is computed as the high book-to-market stock returns minus the low book-to-market stock returns TMO is computed as the transparent firms (by dispersion) minus the opaque firms All returns are in percent R2 is adjusted for degrees of freedom The time series regression is as follows: R(t) – RF(t) = a + b[RM(t) – RF(t)] + sSMB(t) + hHML(t) + mTMO(t) + e(t) 36 TABLE Low High Average B: Mining D: Manufacturing E: Transportation F: Wholesale Trade G: Retail Trade H: Finance I: Services Average Mean Monthly Returns for Portfolios Sorted by Transparency and Volume, Momentum, or Industry Monthly Return Difference between Transparent and Opaque Portfolios (Transparent less Opaque) in Percent Volume Momentum Dispersion Error Dispersion 0.51 0.34 Losers 0.71 0.53 0.33 0.27 0.44 0.46 0.27 0.50 0.48 0.53 0.32 0.45 0.27 0.57 0.63 0.77 0.58 0.53 0.44 0.74 0.56 0.51 0.55 0.45 0.38 0.05 0.14 Winners 0.48 0.47 0.44 Industry Dispersion 0.58 0.56 0.33 0.27 0.52 0.47 0.72 Error 0.47 0.41 0.27 0.63 0.68 0.44 0.77 0.49 0.52 Average Low Volatility High Volatility Average Error 0.48 0.42 0.40 0.12 0.30 0.64 0.44 0.55 0.54 0.71 0.46 0.46 Earnings volatility Dispersion 0.66 0.42 0.36 0.07 0.24 0.04 0.11 0.19 0.28 0.67 Error 0.47 0.17 0.24 -0.02 -0.05 0.17 0.04 -0.21 0.17 0.53 0.30 0.15 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1977 through 1996 Stocks are sorted into deciles based on volume, momenturm, industry, and earnings volatility At the end of June of year t, the stocks within each portfolio are ranked by the following attributes: 1) dispersion defined as the standard deviation of analyst forecasts and 2) error defined as the absolute value of the actual earnings less the average earnings forecast Two portfolios are formed for each attribute, a transparent portfolio and an opaque portfolio The transparency measures are computed as of the previous fiscal year end Equally-weighted, monthly raw returns are measured from July of year t to June of year t+1 All returns are in percent Volume is defined as average monthly volume per CRSP Momentum return is defined as their six-month return from the beginning of January of year t to the end of June of year t The industry groups are as follows: industry A (SIC codes 01-09) is agriculture, forestry, and fishing; industry B (SIC codes 10-14) is mining; industry C (SIC codes 15-17) is construction; industry D (SIC codes 20-39) is manufacturing; industry E (SIC codes 40-49) is transportation, communication, electric, gas, and sanitary services; industry F (SIC codes 50-51) is wholesale trade; industry G (SIC codes 52-59) is retail trade; industry H (SIC codes 60-67) is finance, insurance, and real estate; industry I (SIC codes 70-89) is services; and industry J (SIC codes 91-97) is public administration Included industries must have an average of at least 25 stocks per year Industries A, C, and J did not meet this criteria Earnings volatility is defined as the standard deviation of actual annual earnings in years t-1, t-2, and t-3 37 TABLE Mean Monthly Returns for Transparency Portfolios Sorted by Earnings Characteristics Transparency Quintile Transparent Pessimistic (positive surprise) Optimistic (negative surprise) Difference 1.76 1.43 0.33 1.37 1.26 0.11 Profit Loss Difference 1.65 2.08 -0.43 1.29 1.56 -0.27 Pessimistic (positive surprise) Optimistic (negative surprise) Difference 1.75 1.55 0.20 1.45 1.40 0.05 Profit Loss Difference 1.72 1.64 0.08 1.39 1.39 0.00 Panel A: Portfolios Sorted by Dispersion Mean Equally-Weighted Monthly Raw Returns 1.23 1.05 1.15 1.05 0.08 0.00 1.09 0.87 0.22 0.99 1.13 -0.14 Panel B: Portfolios Sorted by Error Mean Equally-Weighted Monthly Raw Returns 1.29 1.27 1.19 1.09 0.10 0.18 1.18 1.58 -0.40 1.06 1.43 -0.37 Transparent minus Opaque Opaque Average of Quintiles 1.00 0.67 0.33 1.28 1.11 0.17 0.76 0.76 0.75 0.71 0.04 1.15 1.27 -0.12 0.90 1.37 1.01 0.78 0.23 1.35 1.20 0.15 0.74 0.77 0.74 0.81 -0.07 1.22 1.37 -0.15 0.98 0.83 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1977 through 1996 At the end of June of year t, stocks are ranked by the following attributes: 1) dispersion defined as the standard deviation of analyst forecasts and 2) error defined as the absolute value of the actual earnings less the average earnings forecast The transparency measures are computed as of the previous fiscal year end Each stock is placed into one of five portfolios based on its rankings The portfolios are further sorted by the direction of the forecast error or the sign of the earnings per share A mean forecast is a negative surprise when it is greater than the actual earnings A mean forecast is pessimistic (a positive surprise) when it is equal to or less than actual earnings A loss occurs when actual earnings are less than A profit occurs when actual earnings are greater than or equal to zero Equally-weighted raw returns are measured from July of year t to June of year t+1 All returns are in percent 38 TABLE Traditional Risk Measures for Transparency Portfolios Transparent Opaque 1.37 0.98 4.39 Portfolios Sorted by Dispersion 1.30 0.90 4.08 Mean Value-Weighted Raw Return β Standard Deviation of Mean Returns 1.62 1.04 4.75 1.12 0.90 4.10 1.07 1.04 4.77 Mean Value-Weighted Raw Return β Standard Deviation of Mean Returns 1.57 0.97 4.41 1.38 1.01 4.49 Portfolios Sorted by Error 1.30 0.91 4.08 1.21 0.97 4.38 1.01 1.03 4.69 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1977 through 1996 At the end of June of year t, stocks are ranked by the following attributes: 1) dispersion defined as the standard deviation of analyst forecasts and 2) error defined as the absolute value of the actual earnings less the average earnings forecast The transparency measures are computed as of the previous fiscal year end Each stock is placed into one of five portfolios based on its rankings Value-weighted raw returns are measured from July of year t to June of year t+1 All returns are in percent Beta (β) is measured by regressing the mean portfolios returns against CRSP value-weighted excess returns (less the T-bill rate) each month for the sample period The standard deviation of the mean value-weighted monthly raw returns is also computed for the sample period 39 TABLE Year t 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Average of all years Annual Buy-and-Hold Returns of Transparent and Opaque Firms Portfolios Sorted by Dispersion Opaque Transparent (Bottom Quintile) Difference (T – O) (Top Quintile) Annual buy-and-hold return from July of year t to June of year t+1 12.98 16.44 -3.46 16.38 20.19 14.03 6.16 20.68 50.64 27.43 23.21 45.82 -7.86 -24.83 16.97 -6.12 97.67 71.75 25.92 98.51 -10.03 -18.36 8.33 -7.37 36.44 8.80 27.64 37.21 39.44 18.46 20.98 44.02 11.15 12.02 -0.87 14.45 -4.16 -17.07 12.91 -3.32 17.90 13.21 4.69 16.50 17.69 -12.75 30.44 18.16 12.78 -12.51 25.29 13.11 19.11 18.50 0.61 22.91 25.95 25.04 0.91 26.37 6.46 0.61 5.85 9.62 36.37 13.50 22.87 34.04 27.69 20.65 7.04 32.85 19.04 7.53 11.51 17.00 Transparent (Top Quintile) 22.60 9.60 13.00 23.73 Portfolios by Error Opaque (Bottom Quintile) Difference (T – O) 13.45 13.30 37.66 -20.83 85.59 -17.94 7.44 20.48 9.16 -17.68 7.72 -12.01 -14.14 20.44 23.37 1.98 16.22 22.15 5.17 2.93 7.38 8.16 14.71 12.92 10.57 29.77 23.54 5.29 14.36 8.78 30.17 27.25 2.47 3.00 7.64 17.82 10.70 11.83 10.61 13.12 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1978 through 1996 At the end of June of year t, stocks are sorted into quintiles based on rankings of the following attributes: 1) dispersion defined as the standard deviation of analyst forecasts and 2) error defined as the absolute value of the actual earnings less the average earnings forecast The stocks are placed into the quintiles based on their rankings using NYSE breakpoints The transparency measures are computed as of the previous fiscal year end Annual equally-weighted mean buy-and-hold returns are computed for the top and bottom quintiles Returns are measured from July of year t to June of year t+1 All returns are in percent 40 APPENDIX TABLE A1 Transparency Quintile Transparent Opaque Summary Statistics of Portfolios Sorted by Actual Earnings and Dispersion or Error Low 0.01 0.02 0.04 0.08 0.29 0.01 0.02 0.03 0.06 0.24 Mean Dispersion 0.01 0.02 0.04 0.06 0.27 0.00 0.01 0.04 0.09 0.52 Mean Error 0.01 0.02 0.05 0.11 0.53 Actual Earnings Quintile High Low Portfolios Sorted by Dispersion and Actual Earnings 0.02 0.03 0.05 0.09 0.31 0.03 0.07 0.11 0.21 1.98 0.19 0.19 0.17 0.17 0.16 High Mean Actual Earnings (Absolute Value) 0.47 0.80 1.29 0.44 0.81 1.34 0.49 0.81 1.36 0.47 0.82 1.37 0.47 0.82 1.37 2.40 2.63 2.79 3.33 12.25 Mean Actual Earnings (Absolute Value) 0.46 0.80 1.32 0.47 0.82 1.33 0.47 0.80 1.34 0.47 0.82 1.37 0.47 0.82 1.35 2.57 2.57 2.89 3.43 14.96 Portfolios Sorted by Error and Actual Earnings Transparent Opaque 0.00 0.01 0.04 0.11 0.56 0.01 0.03 0.06 0.14 0.69 0.02 0.06 0.15 0.45 8.05 0.16 0.19 0.17 0.17 0.16 Portfolios are constructed using all domestic stocks listed on the NYSE, Amex, and Nasdaq from the period 1977 through 1996 Stocks are sorted into quintiles based on the absolute value of actual earnings in year t and either the dispersion or error measured as of the same date Dispersion is defined as the standard deviation of analyst forecasts and error is defined as the absolute value of the actual earnings less the average earnings forecast The mean absolute actual earnings and corresponding transparency measure are shown for each portfolio 41 ... sorts and Fama and MacBeth (1973) cross-sectional regressions show that the stock returns of transparent firms outperform the stock returns of opaque firms This “transparency component” of stock returns. .. (1978), Givoly and Lakonishok (1979), Fried and Givoly (1982), and Lys and Sohn (1990) See, for example, O’Brien (1988), DeBondt and Thaler (1990), Butler and Lang (1991), Abarbanell and Bernard... between the high and low portfolios for each attribute SMB is computed as the small stock returns minus the large stock returns HML is computed as the high book-to-market stock returns minus the