Does Book-to-Market Equity Proxy for Distress Risk or Overreaction? John M. Griffin and Michael L. Lemmon *

Transcription

1 Does Book-to-Market Equity Proxy for Distress Risk or Overreaction? by John M. Griffin and Michael L. Lemmon * December * Assistant Professors of Finance, Department of Finance- ASU, PO Box , Tempe, AZ , Griffin can be reached at Office: (602) , Fax: (602) , john.griffin@asu.edu and Lemmon can be reached at (602) , mlemmon@asu.edu. Part of this research was conducted while Griffin was a Dice Visiting Scholar at the Ohio State University. We thank Hank Bessembinder, Jim Booth, Kent Daniel, Mike Hertzel, Grant McQueen, René Stulz (the editor), Sheridan Titman, Russ Wermers, two anonymous referees, and seminar participants at Arizona State University, Southern Methodist University, and the 1999 WFA meetings for helpful comments. We acknowledge Lalitha Naveen and Kelsey Wei for research assistance.

2 Does Book-to-Market Equity Proxy for Distress Risk or Overreaction? Abstract This paper tests risk and overreaction explanations of the book-to-market equity (BE/ME) premium in returns by focusing on the joint relationship between distress and BE/ME. Within the most distressed firms, the difference in returns between high and low book-to-market securities is more than twice as large as that in non-distressed firms, and is largely driven by extremely low returns on firms with low BE/ME. These large return differentials cannot be explained by risk as captured by the Fama and French three-factor model, nor differences in economic fundamentals, such as profitability or the likelihood of delisting. In contrast, predictions of the overreaction hypothesis are borne out. Distressed firms exhibit the largest return reversals around earnings announcements, and the book-to-market return premium is largest in small firms with low analyst coverage.

3 Fama and French (1992) document that firms with small market capitalization and high book-to-market equity (BE/ME) earn higher returns. While the magnitude of the size effect appears to be small, 1 the book-to-market equity effect is both robust and economically large. 2 The two most widely accepted explanations of the BE/ME premium in returns focus on risk and investor overreaction. 3 This paper provides new evidence on these explanations by examining the empirical relation between a direct measure of distress, BE/ME, and stock returns. Proponents of the distress risk interpretation of book-to-market equity argue that firms with high BE/ME tend to be distressed firms with poor earnings prospects, and that investors rationally attach a high discount rate to these firms. Fama and French (1995) show that BE/ME captures differences in relative profitability. Furthermore, Fama and French (1993, 1996) show that a three-factor model using the market portfolio and factor mimicking portfolios based on size and book-to-market works quite well at explaining differences in average stock returns across portfolios formed on variables known to cause problems for the Capital Asset Pricing Model (CAPM). Alternatively, Lakonishok, Shleifer and Vishny (1994) suggest that the relation between book-to-market and stock returns is the result of investor overreaction. They argue that some investors get overly excited and buy up stocks that have done very well in the past (low BE/ME stocks). Similarly, they tend to overreact, and sell stocks that have done poorly in the past (high BE/ME stocks). This phenomenon leads to underpricing of high book-to-market or value stocks 1 Using a sample period from 1926 to 1995, Davis, Fama, and French (1999) show that the size effect in returns is quite small, averaging 0.2 percent per month, after controlling for book-to-market. 2 Davis (1994) documents that a book-to-market equity effect exists in US returns from 1940 to Chan, Jegadeesh and Lakonishok (1995) show that this effect is not driven by selection bias caused by the intersection of the CRSP and COMPUSTAT data. Barber and Lyon (1997) provide evidence that the value premium exists in financial firms as well. In addition, international evidence by Chan, Hamao, and Lakonishok (1991), Capaul, Rowley and Sharpe (1993), Hawaini and Keim (1997), and Fama and French (1998), among others, document that small market capitalization and high book-to-market equity stocks also have higher returns in many foreign markets. 3 Other explanations of the book-to-market return premium include sample selection bias (Kothari, Shanken, and Sloan (1995)) and data snooping (MacKinlay (1995)). However, the robustness of the domestic and international evidence noted above, suggests that these biases are at best only partial explanations. 1

4 and overpricing of low book-to-market or glamour stocks. Consistent with overreaction explanations of the value premium, Daniel and Titman (1997) provide evidence that firm characteristics (i.e., size and BE/ME) explain returns better than loadings on the Fama and French factors. However, Davis, Fama and French (1999) argue that Daniel and Titman s results are sub-sample specific. To shed additional light on the competing explanations of the value premium we focus on the joint relationship between book-to-market equity and economic fundamentals predicted to be associated with distress and overreaction. Similar to Chan and Chen (1991) and Fama and French (1995), we believe that differences in risk should be related to differences in economic fundamentals. We identify firms whose returns should be most sensitive to common factor risk related to distress using Ohlson s (1980) measure of financial distress. This variable, denoted by O-score, is based on a weighted sum of accounting ratios and was constructed to estimate the probability of a firm filing for bankruptcy in the year after the variable is calculated. Dichev (1998) uses this distress measure to examine whether bankruptcy risk is priced and if it is correlated with book-to-market equity, but does not focus on the more general issue of the underlying rationale for the BE/ME premium. 4 We show that O-score captures differences in profitability, leverage, and other characteristics hypothesized to be related to differences in risk between high and low book-to-market firms. In addition, distressed firms tend to have characteristics that may be related to mispricing, such as small market capitalization, high amounts of information asymmetry, and low analyst coverage. Our sorts thus produce variation 4 We focus on using Ohlson's measure because, as shown by Dichev (1998), it predicts CRSP delistings better than of Altman s (1968) measure (Z-score). However, we also find similar results using Altman's measure. Shumway (1996) also examines the association between book-to-market and distress risk using a hazard model to predict distress. He finds that distressed firms tend to earn higher returns than nondistressed firms, but that distress risk cannot account for the book-to-market effect in returns. 2

5 in firm characteristics that allow us to distinguish between the risk and behavioral explanations of the value premium. We find that the book-to-market return premium is largest in the most distressed firms. Within the highest O-score quintile, the book-to-market effect is twice as large as in any other quintile, averages 14.4 percent per year, and is driven largely by extremely low returns on firms with low BE/ME. Because the most distressed firms exhibit the largest return differential between high and low BE/ME stocks, any differences in risk should be quite evident in these firms. Nevertheless, the three-factor model of Fama and French (1993) cannot explain the extremely low returns of low BE/ME firms in distress. Pricing errors from the three-factor model average 9.6 percent on an annual basis for these firms. To provide further evidence about differences in risk we also examine earnings performance and the likelihood of delisting as a function of our distress variable. For stocks in the highest O-score quintile, we find results contrary to those of Fama and French (1995) in that low BE/ME firms experience persistently low earnings that are slightly lower than those of high BE/ME firms. The fraction of firms delisted from CRSP for performance reasons is also similar across high and low BE/ME securities. In sum, even though the BE/ME return premium is largest in the high distress quintile, we find no evidence in support of the view that distressed low BE/ME firms are less risky than their high BE/ME counterparts. To assess whether the return patterns are consistent with investor overreaction we investigate abnormal returns around earnings announcements and the association between analyst coverage and subsequent returns. Following Chopra, Lakonishok, and Ritter (1992), La Porta (1996), and La Porta, Lakonishok, Shleifer, and Vishny (1997), we examine returns around subsequent earnings announcements. Consistent with the large return differential in the high O- 3

6 score quintile, we find that the difference in abnormal earnings announcement returns between high and low BE/ME stocks is largest for the most distressed firms. We also sort firms based on firm size and analyst coverage to examine whether overreaction is most pronounced in firms with slower rates of information diffusion, which may be more difficult for investors to value, and where rational arbitrage is less likely to be effective. Both size and analyst coverage play an import role in explaining the book-to-market effect. Small firms with low analyst coverage exhibit a return difference between high and low BE/ME firms of percent per year, as compared to percent per year for large firms with high analyst coverage. Similar to our findings based on O-score, the lowest returns occur on firms with low BE/ME and low analyst coverage. In sum, our results are most consistent with the overreaction hypothesis. However, unlike the usual version of the overreaction hypothesis, in which investors extrapolate past performance too far into the future (e.g., Lakonishok, Shleifer, and Vishny, (1994)), we find the strongest evidence of mispricing in firms with weak current fundamentals. The extremely low returns that we document for distressed low BE/ME stocks suggest that investors overestimate the payoffs from future growth opportunities for these firms. The remainder of the paper is organized as follows. Section II develops our hypotheses relating both risk and overreaction explanations for the BE/ME effect to our measure of distress. Section III describes the sample and provides summary statistics for portfolios formed according to distress probability, size, and BE/ME. Section IV, documents return patterns for these portfolios. Section V examines whether three-factor regressions can explain the returns on distressed firms, and if patterns in earnings and exchange delistings are consistent with a distress risk explanation for the BE/ME effect. Section VI examines whether overreaction is driving the 4

7 BE/ME effect in high O-score firms by examining abnormal stock price performance around subsequent earnings announcements and the relation between analyst coverage and returns. Section VII discusses our findings. A brief conclusion follows in Section VIII. II. Book-to-Market Equity and Distress In this section, we discuss some of the testable implications of the distress risk and overreaction explanations for the BE/ME effect in stock returns. A. Risk and the Book-to-Market Effect Fama and French (1993, 1996) develop a three-factor model that does quite well at capturing patterns in returns known to cause problems for the CAPM. The three factors are the excess market return (MTB), the difference in returns between portfolios of small stocks and large stocks (SMB), and the difference in returns between portfolios of high and low BE/ME securities (HML). One interpretation is that SMB and HML are factor mimicking portfolios in a multifactor equilibrium pricing model in the spirit of Merton (1973) or Ross (1976). Nevertheless, as noted by Fama and French (1995, p. 131): But returns tests cannot tell a complete economic story. Size and BE/ME remain arbitrary indicator variables that, for unexplained economic reasons, are related to risk factors in returns. To shed light on the economic rationale for the value premium, Fama and French (1995) examine profitability, and show that low BE/ME firms typically have persistently strong earnings, while high BE/ME firms are associated with persistently low earnings. Along similar lines, Chan and Chen (1991) argue that small firms (which tend to have high BE/ME) tend to have characteristics associated with distress, such as low profitability, low cash flow, and high financial leverage. Taken together, the evidence is consistent with the view that the return 5

8 premium on small firms and stocks with high BE/ME is compensation for risk related to relative distress. Our goal here is similar. We examine whether BE/ME captures differences in risk by focusing on the relation between BE/ME and underlying economic fundamentals related to relative distress. To capture differences in fundamentals we use a measure of financial distress proposed by Ohlson (1980). This variable, which we denote by O-score, is a weighted average of accounting variables that are available prior to the date that we assign firms to portfolios, and was constructed to estimate the probability of a firm filing for bankruptcy in the following year. By classifying firms based on O-score, we attempt to identify firms whose returns should be particularly sensitive to common factor risk associated with distress. 5 Using a measure of overall economic strength, like O-score, should have substantial benefits for segmenting firms compared to using any single measure of economic stability alone. 6 Assuming that the book-to-market return premium is driven by differences in relative distress, and that a firm s BE/ME ratio and O- score both capture unique information related to distress, the risk-based model predicts that average returns should be increasing in both O-score and BE/ME. B. Investor Overreaction and the Book-to-Market Effect Another explanation for the profitability of value-based investment strategies is that investors tend to overreact. For example, De Bondt and Thaler (1985, 1987) find that stocks with the lowest returns over the last three to five years (high BE/ME) outperform the market over the next three to five years. While their findings have attracted considerable scrutiny (Chan (1988), Ball and Kothari (1989), Conrad and Kaul (1993), and Ball, Kothari and Shanken (1995)), the 5 This effect will be further magnified if distressed firms also tend to have high financial leverage. Galai and Masulis (1976) show that the factor sensitivities increase in absolute value as the firm approaches bankruptcy. 6 For example, high leverage may be a sign of relative distress for many firms, but not for an efficiently run firm in a growing industry. A similar argument is made by Cleary (1999) in his analysis of the relationship between investment and financial status of firms. 6

9 large returns to past losers can not be explained by market risk (Chopra, Lakonishok, and Ritter (1992)) or microstructure effects (Loughran and Ritter (1996)). 7 Lakonishok, Shleifer, and Vishny (1994) test a specific version of the overreaction hypothesis. They suggest that investors place too much emphasis on past fundamentals, such as earnings or sales growth. When future news does not meet investors expectations, prices subsequently adjust, leading to the underperformance of glamour stocks and positive abnormal returns to value stocks. 8 There are other places to look for evidence of overreaction as well. It is likely that mispricing will be more prevalent in small, illiquid stocks with large amounts of uncertainty about future cash flows and low analyst coverage. Since distressed firms are likely to posses these characteristics then, holding other factors constant, it seems plausible that reversals in returns high returns for high BE/ME firms and low returns for low BE/ME firms will be most pronounced for firms in distress. III. Data and Methodology The risk and overreaction explanations yield separate predictions about the joint relationship between BE/ME, O-score, and returns. To test these predictions, we segment firms by forming portfolios from independent ranks on O-score, BE/ME, and size and then examine patterns in returns, factor loadings, and other firm characteristics across these portfolios. The sample is constructed in a manner similar to Fama and French (1992). Non-financial NYSE, NASDAQ, and AMEX stocks with monthly returns from CRSP and with book-values available from COMPUSTAT are examined from June 1965 to June Stocks are ranked each June according to their previous December book-to-market equity ratio and June market 7 Chopra, Lakonishok, and Ritter (1992) argue that time -varying market risk premia (as proposed by Chan (1988) and Ball and Kothari (1989)) cannot fully explain the effect. Loughran and Ritter (1996) demonstrate that the overreaction effect is not driven by low priced stocks with large bid-ask spreads (as proposed by Conrad and Kaul (1993) and Ball, Kothari, and Shanken (1995)). 7

10 capitalization. Ohlson s (1980) measure of the probability of financial distress (O-score) is also calculated using accounting values from the previous December for the June rankings. The definition of O-score is given in the Appendix. Portfolios are formed from three independent rankings on BE/ME, five rankings on O-score, and two rankings on market capitalization (size). The three rankings on BE/ME use 30 and 70 percentile breakpoints. Breakpoints for BE/ME, market value, and O-score are formed using all NYSE, NASDAQ, and AMEX securities. For brevity, we mainly report size-adjusted data for returns and other characteristics. The sizeadjusted variables are formed from a simple average of the means or medians of the small and large firm groups. Using these two size groups performs a simple control on size, but more detailed controls for size are also performed and discussed throughout the paper. When a stock is delisted we adjust the returns using the methodology of Shumway (1997). To avoid problems due to bid-ask bounce we calculated annual value-weighted buy-and-hold portfolio returns by taking the holding period return for each security over the year and calculating a value-weighted average across securities within each portfolio. Details are provided in the appendix. Table I presents size-adjusted median firm characteristics for the stocks in each group. Several interesting findings emerge. Within the highest quintile of O-score, low BE/ME firms have the highest probability of bankruptcy (26.6%), while high BE/ME firms have lower probabilities of bankruptcy (12.8%). These findings imply that, within the highest quintile of distress, a negative relationship exists between BE/ME and O-score. Over the entire sample, the Spearman correlation between O-score and BE/ME is Dichev (1998) also finds that firms with a high probability of distress have low BE/ME. 9 8 Haugen (1995) makes a similar argument. Recent papers by Daniel, Hirshleifer, and Subrahmanyam (1998a, 1998b), Barberis, Shleifer, and Vishny (1998), and Hong and Stein (1999) provide theoretical models of underreaction and overreaction in stock prices resulting from specific behavioral assumptions. 9 Dichev also finds a small positive correlation between O-score and BE/ME. Dichev s sample includes firms with negative BE/ME and covers only the 1981 to 1995 period. 8

11 Within high BE/ME firms, the median book-to-market ratio increases monotonically from to as firms move from low to high O-score. Low BE/ME firms with high O-score actually exhibit the lowest median book-to-market ratio (0.267). Thus, the dispersion in BE/ME is largest in the high O-score group. In addition, distressed firms tend to have smaller market capitalization. As expected, the second row in Table I shows that both O-score and BE/ME capture significant variation in underlying firm fundamentals that have been hypothesized to be related to risk. The return on assets, measured as the ratio of income before extraordinary items to total book assets, decreases as firms become more distressed, and actually becomes negative in the highest O-score quintile. The return on assets is generally inversely related to the firm s book-tomarket ratio, except in the high O-score quintile, where low BE/ME firms are less profitable than those with high BE/ME. Firms with high O-score and high BE/ME also tend to have high leverage. 10 To examine whether distressed firms are associated with characteristics predicted to be related to overreaction, Table I reports the percentage of firms with analyst coverage from the Institutional-Brokers-Estimates-System (IBES), and a measure of residual analyst coverage. 11 Consistent with our arguments in Section II, high O-score firms tend to be less likely to be followed by analysts and have lower residual analyst coverage. Within quintiles of O-score, firms with high BE/ME are more likely to be followed by analysts than low BE/ME firms are. Table I also shows that, consistent with conventional wisdom, low BE/ME firms spend more 10 The fact that O-score is strongly related to earnings and leverage is not too surprising given that O-score is partially constructed using accounting ratios related to these variables. 11 Residual analyst coverage is the difference between the analyst coverage for a firm and the average analyst coverage for firms in the same size quintile. We get similar patterns after controlling for size with a cross-sectional regression approach similar to the procedure of Hong, Lim, and Stein (1999). Since 1976 is the first year that firms are covered by the IBES database, the statistics on IBES coverage considers firm years from 1976 to

12 heavily on R&D. Across O-score quintiles, there is little difference in R&D expenditures for low BE/ME firms. In summary, our sorts reveal that characteristics associated with distress, such as low return on assets and high leverage, are increasing in both O-score and BE/ME, suggesting that both measures provide unique information about underlying fundamentals hypothesized to be related to differences in risk. Furthermore, distressed firms tend to be small and have low analyst coverage characteristics which may be related to the amount of mispric ing. In the next section, we examine whether firms with high O-score and BE/ME are rewarded with higher returns as predicted by the risk model, or whether the patterns in returns are more consistent with investor overreaction. IV. Distress, Book-to-Market Equity, and Returns Table II displays the annual buy-and-hold returns for each distress and book-to-market portfolio. We report results separately for the small and large market capitalization groups, as well as size-adjusted returns. We assess statistical significance using p-values calculated from the time-series of monthly returns. The book-to-market effect in returns tends to increase across O-score quintiles. The average annual size-adjusted percentage return differential between the portfolio of high and low BE/ME securities is 3.87, 3.25, 5.49, 10.62, and within O-score quintiles one through five, respectively. Similar patterns hold for both the small and large firm portfolios separately. For small (large) stocks the return spread between high and low BE/ME stocks is 4.32 (3.42) percent per year for low O-score firms and (15.33) for high O-score firms. In both size groups, the return differentials in the low O-score quintile are not significantly different from zero, while those in the high O-score group are highly significant. 10

13 These findings reveal that Dichev s (1998) evidence that distressed firms earn low average returns is largely driven by the under-performance of low BE/ME stocks. Indeed, the most striking result in the table is the extremely low returns on low book-to-market firms in the high O-score group. The size-adjusted average return on this group of firms is 6.36 percent, which is approximately twice as small as the average return on any of the other portfolios. In fact, this return is slightly lower than the risk-free rate of return over our sample period. Finally, these low returns are not due only to small firms. Using NYSE size breakpoints, the returns of low BE/ME stocks in the highest O-score group average 8.2, 7.1, and 4.2 percent in the first, second, and third NYSE size quintiles, respectively. Distressed low BE/ME stocks also have low returns in the top two NYSE size quintiles, but inference is not reliable because these portfolios contain few and sometimes no stocks in some years of the sample. We perform several robustness checks on our results. First, we repeat our returns sorts using Altman s Z-score to measure distress (see Appendix). In the highest quintile of financial distress according to Z-score, firms with low BE/ME earn 5.95 percent, while high BE/ME firms earn percent. The patterns found with Z-score confirm that the BE/ME effect is most extreme in the highest quintile of financial distress. Second, we verify that our results are not driven by differences in leverage between low and high BE/ME firms. One reason that distressed firms may exhibit a large dispersion in book-to-market ratios and returns is that their returns are more volatile. In this case, large price swings could result in large changes in leverage, leading to large changes in risk. 12 As shown in Table 1, however, distressed low BE/ME firms have significantly higher market leverage than other low BE/ME firms, suggesting that they should actually earn higher returns. To explore this further, we form portfolios by sorting on market leverage and BE/ME. Unlike the O-score results, the book-to-market premium is lowest in the 11

14 highest leverage portfolio. Third, it is possible that the low BE/ME firms in the highest distress quintile are heavily weighted toward firms that have recently gone public. If this is the case, our results could be driven by the low returns following initial public offerings documented by Ritter (1991). We investigate this possibility by repeating our returns sorts after removing firms that have been on the CRSP tapes for less than five years. The results are similar to those reported in Table II. In particular, the size-adjusted returns to low BE/ME firms in the highest distress quintile are 3.48%, indicating that IPO firms are not driving the results. We also investigate the effects of excluding negative BE/ME firms from our analysis. Negative book equity (BE) is one of the predictors in O-score. Interestingly, in the construction of O-score, negative BE results in a lower probability of bankruptcy (see Appendix). Of more concern for our analysis is how to rank negative BE firms in terms of BE/ME. 13 To address this, we rank firms using the book-value of total assets divided by book assets minus book equity plus the market value of equity. The results obtained using this measure of value instead of BE/ME, and including both positive and negative BE firms yields results that are qualitatively similar to those in Table II. Finally, we assess whether the results are affected by the stock exchange of listing, time period, or industry effects. Across exchanges (NYSE/AMEX and NASDAQ), time periods (formation years from 1965 to 1979 and 1980 to 1995), and controlling for industry effects using industry adjusted book-to-market ratios, 14 the spread between high and low BE/ME firms is consistently the largest for firms with high O-score. 12 We thank an anonymous referee for pointing this out. 13 One might think of book-to-market equity is a measure of intrinsic value relative to market value, but once bookvalue becomes negative the market value of the firm becomes irrelevant in the BE/ME ratio. For this reason, Fama and French (1992) exclude these firms, arguing that negative BE firms actually behave more like high BE/ME stocks. 14 The procedure for forming the industry-adjusted BE/ME is discussed in Daniel, Grinblatt, Titman, and Wermers (DGTW) (1997). We thank Kent Daniel for generously supplying the industry-adjusted book-to-market equity ratios. 12

15 In sum, the BE/ME return premium is largest in the most distressed firms and this finding is quite robust. Additionally, the majority of the large return differential between high and low BE/ME securities in distress is driven by the underperformance of low BE/ME securities relative to all other groups. While this finding is consistent with the predictions of the overreaction models, it does seem inconsistent with O-score being correlated with a priced underlying risk factor. Nevertheless, it is possible that book-to-market equity is associated with loadings on a priced risk factor that is not related to O-score. Indeed, the large dispersion in BE/ME ratios for high O-score firms suggests that a risk-based model has a chance to explain the large differences in returns between high and low BE/ME firms in distress. V. Distress, Book-to-Market Equity, and Risk In this section, we examine whether the large return differential between high and low BE/ME firms in the highest O-score quintile can be explained with a risk-based model. A. Distress, BE/ME, and the Three-Factor Model Within the highest quintile of O-score, both low and high BE/ME stocks have poor earnings fundamentals, yet exhibit a wide dispersion in book-to-market ratios. It is thus interesting to see how O-score and the book-to-market ratio relate to factor loadings from the three-factor model of Fama and French (1993). To investigate this, Table III reports post-ranking factor loadings for the portfolios in each book-to-market, O-score, and size group. The factor loadings are estimated from time-series regressions of monthly value-weighted portfolio excess returns on the value weighted market index (MTB), size (SMB), and book-to-market (HML) factors of Fama and French over the entire June 1963 to July 1996 period. As seen in Table III, the loadings on the market and size factors increase monotonically as firms become more distressed (as measured by O-score). The increases in the factor loadings on the MTB and SMB 13

16 factors across O-score groups are considerable. In spite of the fact that they have the lowest book-to-market ratios, large, low BE/ME stocks in the highest O-score group tend to have HML loadings that are similar to those on other low BE/ME firms. Small firms in this same category actually have small positive loadings on the HML factor. The patterns in SMB and HML factor loadings indicate that both O-score and BE/ME are positively correlated with the Fama and French factor loadings. If the multi-factor model adequately describes differences in returns, the intercepts from the time-series regressions reported in Table III should be indistinguishable from zero. For the low BE/ME portfolios, pricing errors become more negative for the higher O-score quintiles, and the pricing errors are statistically significant for the two highest O-score groups. This is true for both the small and large firm portfolios. The portfolio of small (large) firms in the highest O- score quintile with low BE/ME has a negative abnormal return of (-0.87) percent per month, which is both economically and statistically significant with a t-statistic of 3.73 (-4.42). It is important to note that these results are not simply a reaffirmation of the negative regression intercept documented in Fama and French (1993) for the portfolio of stocks with low BE/ME in the smallest NYSE size quintile. First, the magnitude of the intercept is over two times as large as any pricing error in Fama and French (1993). Second, the rejection is not limited to the smallest firms. We re-estimate the three-factor regressions above for portfolios based on NYSE size quintiles. In the first size quintile, the regression intercept for the low BE/ME high O- score portfolio is 0.78 with a t-statistic of In the second (third) size quintile, the intercept is 0.81 (-0.83) percent with a t-statistic of 3.40 (-2.77). Intercepts are also negative in the largest two size quintiles (-0.66, and -1.01), but statistical tests lack power because there are few firms in these portfolios. 14

17 These results suggest that, although the three-factor model does a reasonable job of explaining returns for many firms, it cannot explain the low returns to distressed low BE/ME firms. The large negative abnormal returns, which average 9.6 [12*( )/2] percent per year, for these firms present a significant challenge to the three-factor model. These rejections are even more surprising in light of arguments by Loughran and Ritter (2000) that the threefactor model has low power to reject deviations due to mispricing since the same firms that are being examined are also used in forming the factor mimicking portfolios. B. Distress, BE/ME, and Earnings The main connection between book-to-market equity and economic fundamentals is found in Fama and French (1995). They demonstrate that the behavior of earnings is consistent with book-to-market equity being associated with relative distress in that high BE/ME firms have persistently higher earnings then low BE/ME firms. Given that the difference in returns between high and low BE/ME firms is particularly large for firms in distress, the traces of BE/ME risk in earnings should be most evident in this group of firms. Specifically, high BE/ME firms in distress should display persistently lower earnings than distressed low BE/ME firms. Table IV displays the median earnings relative to book-assets (BA) in the year prior to ranking, the year of ranking, and the year after ranking for each size-adjusted portfolio. 15 In O- score quintiles one through four we find results consistent with those reported by Fama and French (1995); earnings are significantly lower for high BE/ME firms. However, the earnings differential between low and high BE/ME firms decreases monotonically as the probability of distress increases. It is interesting that the relationship between earnings and BE/ME becomes 15 The earnings reported is the median percentage earnings in the year prior to ranking (E(t-1)) scaled by book assets in from the previous year (BA(t-2)). Because changes in earnings directly affect book-equity, a major component of book-assets (BA), all earnings numbers are scaled by BA from year t-2. Scaling by BA in year t yields similar patterns. Measuring profitability as the sum of the earnings of all firms divided by the sum of the book assets for all firms also yields similar results. 15

18 weaker as firms become more distressed, given that the return differentials between high and low BE/ME are increasing in the likelihood of distress. Surprisingly, in the highest quintile of O- score, the relationship is reversed, with low BE/ME firms exhibiting lower earnings than high BE/ME firms in the year prior to ranking. Examination of subsequent years reveals that distressed low BE/ME firms have lower earnings than high BE/ME financially distressed firms in the year of ranking and the year following ranking as well. Although many distressed firms exit from the sample, in unreported results we find that the mean and median profitability of surviving low BE/ME, high O-score firms remains below that of high BE/ME, high O-score firms for five years following portfolio formation. C. Distress, BE/ME, and CRSP Performance-Related Delistings Table IV also reports the percentage of firms in each size-adjusted portfolio with CRSP performance-related delistings. CRSP performance-related delistings include failure to meet minimum exchange requirements, pay fees, file reports, and filing for bankruptcy (delisting codes 500, ). Shumway (1997) finds that these delistings are on average associated with a 30 percent return. O-score is highly correlated with CRSP performance-related delistings. The percentage of firms delisting for performance-related reasons increases nearly monotonically across O-score quintiles. Examining the high O-score quintile, we observe that 6.58 percent of low BE/ME firms are delisted as compared to 5.28 percent of high BE/ME firms. In the next highest distress quintile there is no difference in the delisting probability between high and low BE/ME firms. Within the high O-score quintile, firms in the middle BE/ME group have fewer delistings than either high or low BE/ME firms. 16

19 D. Distress, BE/ME, and Returns During Risky Periods If high BE/ME stocks are riskier and earn higher returns during most states of the world there should be some bad states in which high BE/ME stocks earn low returns. This should be particularly true for distressed firms if they are more sensitive to aggregate economic factors. Following Lakonishok, Shleifer, and Vishny (1994), we examine the return differential between high and low BE/ME securities in periods of high and low market returns and positive and negative GNP growth. 16 Table V presents size-adjusted return differentials for portfolios of high minus low BE/ME stocks in the highest and lowest O-score quintiles across different economic regimes. The percentage return differential for the high O-score portfolio is 22.99, 10.47, 24.47, and 4.05 when moving from periods of low to high market movements. These returns are significant for all periods except the period containing the highest market movements, and are larger than the corresponding returns for the low O-score portfolio. Similar patterns are obtained when the data is divided into periods of negative and positive GNP movements. In general, distressed high BE/ME firms earn higher returns than distressed low BE/ME firms in both good and bad economic states. Overall, the evidence presented in this section does not support the idea that differences in risk can explain the large book-to-market premium in returns for distressed firms. Intercepts from three-factor regressions are economically large and statistically significant for firms in the low BE/ME, high O-score portfolios. In addition, there is no evidence that high BE/ME firms in the highest O-score quintile are riskier than their low BE/ME counterparts in terms of either 16 It is important to note, however, that Fama and French (1996) argue that the relative distress factor captured by HML is not related to market and GNP risk. 17

20 profitability or the likelihood of delisting. Finally, distressed high BE/ME stocks significantly outperform distressed low BE/ME stocks across all economic states. VI. Book-to-Market Equity, Distress, and Investor Overreaction To assess whether the return patterns documented previously are consistent with investor overreaction, we investigate abnormal returns around earnings announcements and the association between analyst coverage and subsequent returns. First, similar to Chopra, Lakonishok, and Ritter (1992), La Porta (1996), and La Porta, Lakonishok, Shleifer, and Vishny (1997), we examine abnormal stock price performance around earnings announcements to assess whether systematic expectational errors can explain the return differentials between value and growth portfolios. If distressed firms are more prone to investor overreaction, and if investors revise their expectations when earnings news is released, then reversals in returns around subsequent earnings announcements should be most pronounced in distressed firms. Second, we examine the relationship between analyst coverage and returns. Similar to distressed firms, we argue that small firms with low analyst coverage are likely to be associated with more information asymmetry. Investors may find these firms more difficult to value leading to more mispricing in small firms with low analyst coverage. 17 This argument suggests that the BE/ME premium in returns should be largest for firms with low analyst coverage. A. Overreaction and Abnormal Earnings Announcement Returns To test the first hypothesis, we examine the three-day abnormal returns for our stocks around quarterly earnings announcements. Following the methodology La Porta, et al. (1997), we benchmark all earnings announcements relative to other securities that are in the same size decile and have a BE/ME in the middle 40 percent of our sample. We compute the average 18

21 abnormal return over the four quarterly earnings announcements in the year following portfolio formation and annualize this number by multiplying by four. Table VI presents the equallyweighted and value-weighted annualized abnormal returns for our portfolios from 1971 to Consistent with La Porta, et al. (1997), the equally-weighted earnings announcement returns are negative for low BE/ME portfolios and positive for high BE/ME firms portfolios. Examining the performance across O-score quintiles reveals that the high BE/ME firms in the highest O-score quintile have the largest positive abnormal earnings surprises. High O-score firms also exhibit the largest difference in abnormal returns between high and low BE/ME stocks, averaging 3.39 percent (p-value= 0.000). In comparison, the difference in abnormal returns between high and low BE/ME stocks in the lowest O-score quintile is 2.14 percent (pvalue= 0.000). Value-weighted earnings announcement returns exhibit a similar pattern. These findings provide additional support for the hypothesis that distressed firms are more difficult to value, and that overreaction is most pronounced in high O-score firms. 19 B. Overreaction and Analyst Coverage We form portfolios based on the prior year values of residual analyst coverage, size, and book-to-market equity with size quintiles based on NYSE breakpoints in Table VII. Consistent with mispricing, both size and analyst coverage play an important role in explaining the book-tomarket equity return premium. Within each size quintile, the difference in value-weighted returns between high and low book-to-market stocks is larger for firms with low analyst coverage than 17 The model of Hong and Stein (1999) also provides a link between analyst coverage and mispricing. They argue that both momentum and overreaction should be most pronounced in those stocks where information diffuses more slowly. Using analyst coverage as a proxy for the rate of information flow, Hong, Lim, and Stein (1999) find that momentum is strongest in small firms with low analyst coverage. They do not examine overreaction. 18 Earnings announcement dates are reported on COMPUSTAT beginning in

22 for high analyst coverage stocks. The small firm portfolio with low analyst coverage displays a return difference between high and low BE/ME of percent per year (p-value= 0.000), while the large, high analyst coverage portfolio displays a return difference between high and low BE/ME of percent per year (p-value= 0.591). 20 Given that distressed firms have low analyst coverage, these findings provide some support for the conjecture that overreaction is most pronounced in firms with weak current fundamentals and large information asymmetries. VII. Discussion and Interpretation The evidence presented so far is most consistent with investor overreaction and suggests that mispricing is most prevalent in distressed firms and in small firms with low analyst coverage, which are likely to be firms that are difficult for investors to value and where rational arbitrage is less likely to be effective. One interesting issue that remains is to examine what is driving the overreaction. A common version of the overreaction hypothesis argues that investors extrapolate past performance too far into the future. For example, Lakonishok, Shleifer, and Vishny (1994) show that past growth in sales and estimates of future growth potential, like cash flow and earnings to price ratios, are useful for predicting future returns. As illustrated in Table IV, however, low BE/ME firms in distress exhibit low profitability both before and after portfolio formation. The weak current earnings of these firms raise the question as to why they have high market values relative to their book values of equity. 19 An alternative interpretation is that earnings announcements are more informative for distressed firms. If this is the case, then overreaction will appear larger around earnings announcements for distressed firms because earnings announcements contain more information for these firms. Conversely, firms in distress may be more likely to postpone or pre-announce bad news, which would make earnings announcements less informative. Consistent with this view, we find that firms in the high O-score quintile have more variation in the time between earnings announcements, as compared to low O-score firms. This would tend to mitigate the finding that the largest reversals occur around earnings announcements for distressed firms. 20 Similar to our O-score findings, the three-factor model fails to explain the low returns to low analyst coverage, low BE/ME firms. 20

23 One possibility is that the low book-to-market ratios in distressed firms are driven by low book values rather than high market values. Specifically, because negative shocks to earnings directly affect the book value of equity, it may be the case that distressed firms have low BE/ME not because investors are awarding them high market values, but rather because these firms have low book values. To address this issue, Table VIII reports the growth in retained earnings two years prior and in the year prior to portfolio formation ranking for firms ranked according to O- score and BE/ME. Consistent with the negative earnings of these firms shown in Table IV, retained earnings growth is generally negative for financially distressed firms. In the high O- score quintile, however, retained earnings growth in the year prior to ranking is larger (less negative) for low BE/ME firms than that for high BE/ME firms, suggesting that negative shocks to book equity can not be a total explanation for the low BE/ME ratios of these firms. Further support for this explanation comes from examining the equally weighted buy-and-hold stock returns of our portfolios over the one and three years prior to portfolio formation. 21 Over both time horizons, and across all O-score groups, the average past returns of low BE/ME stocks are significantly higher than those of high BE/ME stocks. Among all low BE/ME groups, high O- score firms have the largest twelve-month prior returns, and the lowest three-year prior return. The prior return results also suggest that our findings are not due to momentum since negative momentum would be needed to explain the low returns to distressed low BE/ME firms. Table VIII also reports the percentage growth in sales in the year prior to ranking. The median growth rate in sales for distressed low BE/ME firms (15.3 percent) is much larger than that in distressed high BE/ME firms (1.6 percent), but still smaller than for other non-distressed low BE/ME firms. The prior earnings performance and growth in sales for these firms do not 21 The previous month s return is skipped to avoid bid-ask bias as discussed in Jegadeesh (1990) and Jegadeesh and Titman (1993). 21

24 appear to tell a complete story of why investors award these firms with low multiples of BE/ME. Thus, in contrast to previous versions of the overreaction hypothesis, our results seem to suggest that investors are overly optimistic about the future growth potential of high O-score, low BE/ME firms, despite the fact that they have low current earnings. Table VIII shows that distressed firms with low BE/ME have the highest capital expenditures as a percent of book assets. These results are similar to those from Table I, which show that R&D expenditures as a percentage of book assets for high O-score low BE/ME firms is the second highest among all the groups. These findings indicate that, in spite of their poor current earnings, these firms continue to invest heavily in future growth opportunities. We also examine the industry concentrations of these firms. In unreported results, we find that the low BE/ME firms in distress have the second highest percentage of firms in growth-oriented industries. 22 One possible interpretation of our findings is that investors are overly excited about the growth potential of these firms, despite their poor current fundamentals. Taken together with the returns around earnings announcements and the results for analyst coverage, the findings support the conjecture that investors may overestimate the payoffs from future growth opportunities for distressed growth oriented firms. VIII. Conclusion In this paper, we use a direct proxy for distress to examine whether the relationship between book-to-market equity and returns is compensation for risk or is due to investor overreaction. Distressed firms tend to have low return on assets, high leverage, and higher SMB and HML factor loadings than their non-distressed counterparts. Distressed firms also have characteristics, such as low analyst coverage, that should be associated with the degree of 22

25 mispricing. To test the risk and overreaction hypotheses, we sort stocks independently on bookto-market equity and distress. These sorts reveal that the book-to-market equity effect in returns is largest in distressed firms and is largely driven by very low returns on low BE/ME stocks in this group. If risk can explain the book-to-market effect, then the large return differential between high and low BE/ME firms in distress suggests that any differences in risk should be most evident in this subset of firms such is not the case. The three-factor model of Fama and French leaves economically large average pricing errors of 9.6 percent per year on low BE/ME stocks in financial distress. Contrary to Fama and French (1995), for firms in the highest quintile of distress, high BE/ME firms actually have higher earnings than low BE/ME firms. In addition, for firms in distress, low BE/ME firms are somewhat more likely to delist for performance-related reasons than high BE/ME firms are. Consistent with predictions of the overreaction hypothesis, abnormal returns around subsequent earnings announcements are largest for firms in financial distress. In addition, we directly test whether mispricing is most prevalent in firms with few analysts. Both analyst coverage and size play an important role in explaining the BE/ME effect the return differential between high and low BE/ME firms is percent per year for small firms with low analyst coverage, but percent per year for large firms with high analyst coverage. Overall, our analysis does not support a risk-based explanation of the book-to-market premium in returns but does find substantial evidence for investor overreaction. The evidence against a risk-based interpretation is strongest in firms where the book-to-market premium is largest. It is important to note that the evidence of overreaction presented here is somewhat 22 We define growth industries subjectively, using the industry definitions in Fama and French (1997). We classify the following as growth industries: Healthcare, Medical Equipment, Pharmaceuticals, Chemicals, 23