EarningsBased and AccrualBased Market Anomalies: One Effect or Two?
Daniel W. Collins*
Henry B. Tippie Research Chair in Accounting
Paul Hribar
Ph.D. Student
College of Business Administration
University of Iowa
Iowa City, IA 522461000
May 25, 1999
Abstract: This paper investigates whether the accrual pricing anomaly documented by Sloan (1996) for annual data holds for quarterly data and whether this form of market mispricing is distinct from the postearnings announcement drift anomaly. We find that the market appears to overestimate (underestimate) the persistence of the accrual (cash flow) component of quarterly earnings and, therefore, tends to overprice (underprice) accruals (cash flows). Moreover, the accrual (cash flow) mispricing appears to be distinct from postearnings announcement drift. A hedge portfolio trading strategy that exploits both forms of market mispricing generates abnormal returns in excess of those based on unexpected earnings, accruals, or cash flow information alone.
Key Words: Market Efficiency, Accruals, PostEarnings Announcement Drift
JEL Classification: G14, M41
*Corresponding Author. Phone (319) 3350910; Fax: (319) 3351956; email: danielcollins@uiowa.edu.
We gratefully acknowledge the insightful comments and suggestions made by Kevin Den Adel, Bruce Johnson, Mort Pincus, Shyam Sunder, Richard Tubbs, Charles Wasley, Greg Waymire and workshop participants at the University of Utah and Carnegie Mellon University.
Earningsbased and accrualbased market anomalies: one effect or two?
The efficiency of capital markets has received a great deal of attention in both the finance and accounting literature. Research in this area most commonly examines market anomalies, which are defined simply as predictable abnormal returns (Ball 1992). Prominent examples of market anomalies include postearnings announcement drift (e.g. Foster et al. 1984; Bernard and Thomas 1989, 1990), the booktomarket anomaly (Stattman 1980), the earningstoprice anomaly (Basu 1977), and more recently the accrual anomaly (Sloan 1996). This study compares two prominent earningsbased accounting anomalies  postearnings announcement drift and the accrual anomaly  to determine whether they capture the same market inefficiency or whether they represent different anomalies that in combination reveal more extreme market mispricing than has been documented in the literature to date.
The literature on postearnings announcement drift demonstrates that prices continue to drift in the direction of the initial market response to quarterly earnings surprises (standardized unexpected earnings or SUEs) for at least 120 trading days following the earnings announcement, with much of the price adjustment occurring in the days surrounding the subsequent two quarters’ earnings releases (e.g. Foster, Olsen, Shevlin 1984; Rendleman, Jones, Latane 1987; Bernard and Thomas 1989; 1990; Freeman and Tse 1989). The results of this stream of research suggest that the market fails to fully appreciate and price the future earnings
implications of current earnings surprises. Essentially, the postearning announcement drift literature suggests that the market underreacts to earnings surprises.
The accrual anomaly documented by Sloan (1996) shows that the level of standardized accruals is negatively related to abnormal returns over the following year. Sloan’s results suggest that the market fails to appreciate that the accrual component of earnings is less persistent than the cash flow component. Consequently, the market appears to overreact to earnings that contain a large accruals component. This result holds for both extreme positive and negative accruals. The overreaction is subsequently reversed when earnings are reported in the following year and the market learns that the earnings of the previous period are not sustainable.
The goal of this paper is to provide additional insight into the relation between the earningsbased and accrualbased anomalies (and the closely related cashflow anomaly). To address this issue, three primary questions are answered. First, does the accrual anomaly hold on a quarterly basis? That is, do portfolios formed on the basis of quarterly accruals provide evidence that the market overreacts to earnings surprises with a large accrual component? Second, if the accrual anomaly holds on a quarterly basis, does it represent a form of mispricing distinct from postearnings announcement drift? That is, can undereaction to earnings surprises and overreaction to accruals coexist? Finally, if postearnings announcement drift and the accrual anomaly represent distinctly different forms of market mispricing, can they be combined into a joint strategy that exploits both anomalies to earn even larger abnormal returns than those associated with the individual anomalies? For example, will the amount of postearnings announcement drift be less prominent or even move in the opposite direction for firms that report extreme positive (negative) unexpected earnings due to extreme positive (negative) accruals? Conversely, will the mispricing be greater when positive (negative) earnings surprises contain a large negative (positive) accrual component? If so, hedge portfolios based on various combinations of unexpected earnings and accrual rankings should generate abnormal returns that exceed those that can be earned by exploiting any one of the anomalies in isolation. Analogous predictions hold for combinations of unexpected earnings and cash flow strategies.
Our results provide evidence of statistically significant abnormal returns associated with quarterly accrual and cash flowbased trading strategies as well as unexpected earningsbased strategies. More importantly, the unexpected earnings and accrual (cash flow) strategies appear to capture different mispricing phenomenon. Thus, combining the earningsbased SUE strategy with either the accrual or the cash flowbased strategy significantly increases the magnitude of abnormal returns that can be earned. Moreover, there does not appear to be significant additional risk involved with the combined strategies in terms of magnitude or frequency of losses. In summary, we find the unexpected earnings and accruals/cash flow anomalies, when combined, reveal a more extreme form of market mispricing than previously documented in the literature. This mispricing can be exploited to generate abnormal returns in excess of those based on unexpected earnings, accruals or cash flow rankings alone.
The remainder of the paper proceeds as follows. Section two summarizes the extant research on earningsbased and accrualbased (cash flow) anomalies and develops the hypotheses. Section three explains sample selection procedures and provides descriptive statistics. Section four outlines the Mishkin test (1983) and hedge portfolio tests for market mispricing and gives the empirical results. Section five provides diagnostic tests, and Section six gives our summary and conclusions and discusses the future research opportunities suggested by our findings.
2. EarningsBased and AccrualBased Market Anomalies:
Postearnings announcement drift  the phenomenon where stock prices continue to drift in the direction of the initial price response to an earnings announcement  is one of the most prominent and perplexing market anomalies documented in the accounting literature. More generally, postearnings announcement drift can be viewed as a manifestation of what Bernard, Thomas and Whalen (1997) label the standardized unexpected earnings (SUE) effect. This broader characterization refers to all market anomalies designed and tested using the SUE metric as a proxy for unexpected earnings. Therefore, it includes the literature on postearnings announcement drift as well as the related work on the time series properties of SUEs (e.g. Bernard and Thomas 1990, Ball and Bartov 1996). In the current paper, these anomalies are referred to as earningsbased SUE anomalies.
The central puzzle underlying this group of anomalies is that stock prices act as if investors use a simple seasonal random walk with drift expectations model in forming quarterly earnings expectations. However, earnings forecast errors conditional on such a model exhibit strong and predictable autocorrelation patterns. Thus, when subsequent quarterly earnings are announced, stock prices adjust to a component of the earnings surprise that should have been predictable given the past time series of earnings. The upshot of these findings is that stock prices do not appear to fully impound the implications of current quarterly earnings surprises for future earnings. While there is some skepticism that SUEbased anomalies truly reflect a departure from market efficiency, rather than just an artifact of research design, Bernard et al.
(1997) argue that of six prominent anomalies, this group is the most likely to reflect market mispricing.
Sloan’s (1996) results suggest the market fails to properly price the accruals component of earnings. He shows that the market erroneously overestimates the persistence of the accruals component of annual earnings while underestimating the persistence of the cash flow component. Moreover, accruals exhibit negative serial correlation or mean reversion tendencies. Consequently, the market responds as if surprised when seemingly predictable earnings reversals occur in the following year.
In order to exploit this overreaction to accruals, Sloan forms zero net investment hedge portfolios that take a long position in firms with the largest negative accruals (standardized by total assets) and an offsetting short position in firms with the largest positive accruals. This strategy earns positive annual excess returns of 10.4% and incurs losses in only two of the thirty years examined. Although Sloan’s hypotheses are stated in terms of accounting accruals, he notes that similar predictions could be made based on investors underestimating the persistence of cash flows (p.292, footnote 4). Thus, a trading strategy taking a long position in firms with the highest level of operating cash flows and an offsetting short position in firms with the lowest level of operating cash flows also appears to generate positive abnormal returns.
One complication that arises when comparing the earningsbased anomaly with an accrual or cash flow anomaly is that the SUEbased anomaly has been examined on a quarterly basis while Sloan documents the accrualbased anomaly only for annual data. Accordingly, to facilitate a comparison between the unexpected earnings and accrual anomalies, we first
implement the accrualbased and related cash flowbased trading strategies on a quarterly basis to determine whether the mispricing documented by Sloan generalizes to quarterly data. In a quarterly setting, it is unclear how long it will take for any mispricing which may be associated with quarterly accruals and cash flows to be ‘corrected’ when subsequent quarters’ earnings are realized. Because postearnings announcement drift largely manifests itself in the two quarters immediately following the earnings announcement, the primary tests will focus on the two quarters immediately following an extreme quarterly accrual or cash flow realization.
A second objective of this study is to determine if the unexpected earningsbased anomaly is distinct from the accrual or cash flow anomalies and whether one dominates or is subsumed by the other. We do this by forming portfolios based on rankings along both dimensions (unexpected earnings and either accruals or cash flows). The resultant contingency tables allow us to examine conditional and marginal frequencies to determine the degree of overlap between the unexpected earnings and accruals/cash flow rankings and to test predictions about the abnormal return performance of various portfolio subgroups within the contingency table.
If, in fact, unexpected earnings and accruals/cash flow anomalies reflect different forms of mispricing and are largely independent of one another, we predict that a new trading strategy which takes advantage of both forms of mispricing will yield even larger abnormal returns than those associated with each individual anomaly. Specifically, formulating a strategy that takes a long (short) position in firms with extreme positive (negative) unexpected earnings and extreme negative (positive) accruals is expected to yield larger abnormal returns than a strategy that exploits only one of the anomalies. Similarly, a strategy that takes a long (short) position in firms with extreme positive (negative) unexpected earnings and extreme high (low) operating cash flows is expected to dominate a strategy that exploits only the SUE or cash flow strategy alone.
Quarterly COMPUSTAT and daily CRSP data are collected for all NYSE/AMEX firms on either the Primary, Supplementary and Tertiary Industrial File or the Research File over the years 19881997. In contrast to Sloan (1996), we estimate quarterly accruals as the difference between earnings and cash flows from operations. Specifically, the accruals component of quarterly earnings is computed as follows:
Accruals_{t} = Earnings_{t}  CFO_{t} (1)
where
Earnings_{t} = earnings from continuing operations (COMPUSTAT #8)
CFO_{t} = cash flow from operations (COMPUSTAT #108).
As in Sloan (1996), earnings, cash flows and accruals are all standardized by average total assets to enhance crosssectional comparability.
To provide evidence on postearnings announcement drift and the earningsbased SUE anomalies, expected earnings are calculated as a seasonal random walk with drift
(Bernard and Thomas’ 1989, 1990):
E_{ }(Q_{i,t}) = Q_{i,t4} + d _{i} (2)
where d _{i} is the drift term estimated using a minimum of 12 quarters and a maximum of 36 quarters. The SUE for firm i in quarter t is then calculated as:
SUE_{i,t} = (Q_{i,t} – E(Q_{i,t})) / s [Q_{i,t} – E(Q_{i,t})] (3)
where s [Q_{i,t} – E(Q_{i,t})] is the standard deviation of the forecast error. Note that equation (2) excludes the firstorder autoregressive term used in Bernard and Thomas (1989). Foster, Olsen, Shevlin (1984), however, report no substantive difference from using either estimation equation, and Bernard, Thomas, Whalen (1997) use equation (2) in their analysis of the SUEbased anomaly. Additionally, while the standard deviation of the forecast error is used as a scaling variable when replicating prior studies, later empirical tests use unexpected earnings standardized by average total assets in order to maintain a consistent scaling variable across all variables. The final sample of quarterly accruals, operating cash flows and SUE realizations contains 41,237 firmquarters.
Tables 1a and 1b provide descriptive statistics for the final sample based on quarterly partitions of both accruals and cash flows. In Table 1a, quarterly accruals are used as the partitioning variable. The negative correlation between accruals and cash flows is evident in the mean cash flow realizations, which appear to be monotonically decreasing with respect to the accrual ranking. Similarly, in Table 1b when cash flows are used as the partitioning variable, mean accrual realizations appear to be monotonically decreasing with respect to the cash flow ranking. This is confirmed by a strong negative Spearman rank correlation of –0.75 (pvalue < 0.001) between accrual and cash flow decile classifications.
The SUE variable appears to be positively correlated with both accruals and cash flows, although the relation is not linear. This is confirmed by Spearman rank correlations of 0.078 (pvalue < 0.001) between SUE and accrual deciles, and 0.125 (pvalue < 0.001) between SUE and cash flow deciles. It should be noted that because of the documented postearnings announcement drift effect, the correlation between unexpected earnings and either accruals or cash flows might confound a trading strategy based on these latter variables. For example, a quarterly accrual strategy predicts positive (negative) excess returns for firms in the lowest (highest) accrual decile. The positive correlation of accruals with SUE, however, implies that there will be a disproportionate number of firms with large negative accruals that will belong to the lowest unexpected earnings decile, for which we predict negative excess returns. Thus, quarterly accrual trading strategies that do not control for unexpected earnings will likely be understated. With cash flows the relation is just the opposite. Cash flow deciles tend to be positively correlated with unexpected earnings deciles. Thus, the highest cash flow decile will tend to be populated by a disproportionate number of firms in the highest unexpected earnings decile. Accordingly, the abnormal returns to a cash flow strategy that fails to control for unexpected earnings are likely to be overstated. Our subsequent joint tests control for both SUE decile membership and accrual/cash flow decile membership.
With respect to the risk proxies listed in Panel B of Tables 1a and 1b, the extreme accrual and cash flow deciles on average contain smaller firms as well as firms with slightly higher betas. Table 1a reveals that the mean betas for the first and tenth accrual deciles are 1.08 and 1.10 respectively, while the mean beta for firms in accrual deciles 2 through 9 is 1.04. Similarly, mean betas for the first and tenth cash flow deciles are 1.12 and 1.08 respectively, while the mean beta for firms in cash flow deciles 2 through 9 is 1.08. The average size based on market value decile ranking for the first and tenth accrual deciles are 6.6 and 6.6 respectively, while the average for accrual deciles 2 through 9 is 7.7. Similarly, the average market value decile ranking for the first and tenth cash flow deciles are 6.3 and 7.3 respectively, while the average for deciles 2 through 9 is 7.7.
As noted in Sloan (1996), the symmetric relationship between firm characteristics in the extreme accrual and cash flow deciles results in negligible exposure to either beta or size pricing effects when offsetting long and short positions are taken in firms in these extreme deciles. Nevertheless, to control for potential risk or sizebased explanations for our results, returns will be adjusted for both size and beta before computing excess returns
3.1 Abnormal return calculations
Previous anomaly studies typically calculate abnormal returns using a companion portfolio approach (e.g. Foster, Olsen and Shevlin, 1984; Bernard and Thomas, 1989; Sloan 1996). Under this approach, sizeadjusted abnormal returns are computed as:
AR_{i,t} = R_{i,t}  R_{p,t} (4)
where AR_{it} = sizeadjusted abnormal return for firm i, day t.
R_{it} = the raw return for firm i, on day t.
R_{pt} = value weighted mean return on the NYSE/AMEX
size decile that firm i is a member of in the quarter examined.
CRSP computes size decile returns (R_{p,t}) by ranking firms on market capitalization at the beginning of the portfolio formation year and then dividing these firms into ten equal portfolios. The decile classifications are rebalanced annually. As indicated in Table 1, the extreme accrual (cash flow) deciles (1 and 10) tend to be composed of smaller firms and firms with higher betas. Moreover, the full sample has a mean beta exceeding 1, indicating that on average, the entire sample contains somewhat riskier firms (mean beta = 1.04). While there is a known negative relation between size and beta, basic exploratory tests for the sample show statistically positive abnormal returns for the entire sample of approximately 1.8% over the two years following portfolio formation when using equation (4) to compute abnormal returns. This suggests that implicitly assuming a coefficient of one on R_{p,t} in equation (4) may ignore firmspecific covariation with the an appropriately chosen benchmark (e.g. the size control portfolio). Therefore, to control for both size and covariation as sources of risk, the following regression is estimated over the 60 months prior to the month of the portfolio formation:
where the variables are as defined above. Abnormal returns are then computed as prediction errors using the following equation:
Using size portfolio average returns as the independent variable controls for the wellknown size effect (Banz, 1981 and Reinganum, 1982). Moreover, allowing the coefficient on the size portfolio return to vary by firm helps control for firmspecific systematic risk. Using equation (6) results in abnormal returns that are insignificantly different from zero for the entire sample over the two years following portfolio formation, suggesting that risk and size effects have been effectively controlled.
Abnormal returns are cumulated from 18 days after the earnings announcement for quarter t to 17 days after the earnings announcement for quarter t+2 (approximately 120 trading days). The hedge portfolio return is computed by subtracting the average abnormal return on the short portfolio from the average abnormal return on the long portfolio. Thus, for example, taking the average cumulative abnormal return of firms in the highest SUE decile and subtracting the cumulative abnormal return of firms in the lowest SUE decile produces the SUE hedge portfolio return.
4. Tests of Market Mispricing
Following Sloan (1996), we use the Mishkin (1983) test and hedge portfolio tests to determine whether the market efficiently impounds accounting information (e.g., earnings surprises (SUEs), accruals, or cash flows from operations) into the price structure.
4.1 Mishkin test
Mishkin (1983) develops a framework to test the rational expectations hypothesis in macroeconomics. We adapt this framework to examine whether the market’s valuation of quarterly unexpected earnings, accruals or cash flows rationally anticipates the implications of these signals for future earnings. Mispricing is indicted if the weight the market assigns to these items in valuation is significantly larger (smaller) than the weight that these items receive in predicting future earnings.
In the present setting, an earnings forecasting equation is combined with a rational pricing model to yield the following system of equations that are jointly estimated:
Q_{t+1} = Q_{t3} + a (Q_{t} – Q_{t4}) + d + u _{t} (7)
CSAR_{t+1 }= R_{t+1}  E_{t}(R_{t+1½ q t}) = b [Q_{t+1} – Q_{t3}  d  a ^{*} (Q_{t} – Q_{t4})] + e _{t}, (8)
where q _{t} is the set of information available to the market at the end of period t.
Notice that the forecasting equation is essentially Foster’s (1977) model of quarterly earnings. Conditional on equation (7) being correctly specified, market efficiency implies that expected sizeadjusted abnormal returns (CSAR_{t+1 }) should be zero. Therefore, market efficiency imposes the constraint that a = a ^{*}. This nonlinear constraint requires that the market correctly anticipate the implications of seasonal differences for the most recent quarter for updating forecasts of next quarter’s earnings. The evidence on postearnings announcement drift suggests that the market systematically fails to do this and, therefore, that a ^{*} < a .
To test whether Sloan’s (1996) findings of the market’s overpricing of accruals carries over to a quarterly setting, the current quarter’s earnings (Q_{t}) are decomposed into its accruals and cash flow components and the above system of equations is rewritten as follows:
Q_{t+1} = Q_{t3} + g _{1}Accruals_{t} + g _{2} CFO_{t} – a Q_{t4} + d + u _{t} (9)
CSAR_{t+1 }= b [Q_{t+1} – Q_{t3}  d  g _{1}^{*}_{ }Accruals_{t}  g _{2}^{*} CFO_{t} + a ^{*}Q_{t4}] + e _{t}, (10)
The benefits of this specification are twofold. First, decomposing the current quarter’s earnings into accrual and cash flow components allows the coefficients on the two components to differ, thereby highlighting any differential persistence between the accrual and cash flow
components. Second, allowing the a coefficient to vary between the two equations in essence captures the effect of postearnings announcement drift. Thus, we are able to investigate if accruals and/or cash flows are systematically mispriced on a quarterly basis after controlling for postearnings announcement drift. If the market appropriately prices the future earnings implications of current accruals and cash flows, then we expect g _{1} = g _{1}^{*} and g _{2} = g _{2}^{*}. If, however, Sloan’s findings that the market overprices accruals and underprices cash flows carries over to a quarterly setting, then we expect g _{1} < g _{1}^{*} and g _{2} > g _{2}^{*}. Moreover, if postearnings announcement drift exists after allowing the coefficients on accruals and cash flows to differ, we expect a > a ^{*}.
Mishkin shows that the following likelihood ratio statistic is distributed asymptotically c ^{2}(q) under the null hypothesis of market efficiency:
2*n*ln(SSR^{c}/SSR^{u}),
where
q = the number of constraints imposed by market efficiency,
n = the number of observations in the sample,
SSR^{c} = the sum of squared residuals from the constrained weighted system, and
SSR^{u} = the sum of squared residuals from the unconstrained weighted system.
Market efficiency is rejected in favor of market mispricing if the likelihood ratio statistic is sufficiently large; i.e. if imposing the market efficiency constraint substantially increases SSR^{c} as compared to SSR^{u}.
Table 2 presents the results of the Mishkin tests. Panel A provides the results of estimating equations (7) and (8). Consistent with the evidence in Bernard and Thomas (1989, 1990) and Ball and Bartov (1996), the market appears to systematically underreact to quarterly earnings surprises. Specifically, the a _{1} coefficient in the forecasting equation equals 0.306 while the a _{1}^{*} coefficient in the returns equation equals 0.093. The likelihood ratio statistic for the test of market efficiency (a _{1} = a _{1}^{*}) is 43.55, which is significant at the a =0.001 level.
Table 2, Panel B presents the results of estimating equations (9) and (10) where we decompose the current quarter’s (Q_{t}) earnings into its cash flow and accruals component. This analysis allows us to determine if the accrual or cash flow mispricing is distinct from the postearnings announcement drift effect captured by parameters a _{1} and a _{1}^{*}. Consistent with Sloan’s annual results, Panel B shows in the forecasting equation that quarterly accruals (g _{1} = 0.234) are less persistent than quarterly cash flows (g _{2} = 0.247) with respect to next quarter’s earnings. However, the difference is less pronounced than on an annual basis. More importantly, the market appears to systematically overestimate the persistence of accruals (g _{1}^{*} = 0.295), and underestimate the persistence of cash flows (g _{2}^{*} = 0.134). The likelihood ratio test statistic for market efficiency, which is a joint test of whether g _{1} = g _{1}^{*} and g _{2} = g _{2}^{*}, is 113.44 which is significant at the 0.001 level. Finally, the unexpected earnings coefficient in the forecasting equation (a _{1} = 0.301) is still larger than the unexpected earnings coefficient implied in the returns equation (a _{1}^{*} = 0.092). This result demonstrates that postearnings announcement drift is still present after allowing for differential pricing implications of the accrual and cash flow components. In summary, it appears that postearnings announcement drift as well as accrual and cash flow mispricing occurs on a quarterly basis. These results provide the basis for hedge portfolio tests that attempt to simultaneously exploit both of these anomalies.
SUE and accrual or cashbased hedge portfolios are constructed to determine if the abnormal returns documented in Bernard and Thomas (1990) can be replicated, and whether or not the accrual anomaly documented by Sloan (1996) holds on a quarterly basis. We form hedge portfolios taking corresponding long and short positions in firms from extreme decile rankings for either SUE, accruals or cash flows. As noted earlier, the abnormal returns are calculated from 18 days after the earnings announcement date for quarter t to 17 days after the earnings announcement date for quarter t+2. SUE deciles are numbered 1 through 10 with SUE1 representing firms with the most negative unexpected earnings and SUE10 representing firms with the most positive unexpected earnings. Likewise, accrual deciles are numbered 1 through 10 with ACC1 representing firms with the largest income decreasing (negative) accruals and ACC10 representing firms with the largest income increasing (positive) accruals. For cash flow portfolios, CFO1 represents firms with the lowest level of operating cash flows and CFO10 represents firms with the highest level of operating cash flows. For the tests using quintiles rather than deciles, a similar convention is used with 1 and 5 representing the lowest and highest level of the respective partitioning variable.
To ensure the hedge portfolio strategy is implementable, the entire distribution of the partitioning variable must be known prior to the portfolio formation date. For example, the distribution of SUEs is unknown until the last earnings announcement has been made in a given quarter, and the distributions of accruals and cash flows are unknown until all financial statements have been publicly released in a given quarter. To mitigate this concern, the decile cutoffs for the quarterly SUE portfolios are based on the previous quarter’s SUE distribution as in Bernard and Thomas (1989). For the accrual or cash flow classifications, firms are classified into deciles or quintiles based on the distribution of accruals from the same quarter of the previous year. This better reflects the effect of the ‘integral’ approach to quarterly reporting, whereby certain expenses are estimated during the first three quarters and the cumulative effect is corrected in the fourth quarter (Rangan and Sloan 1998). The primary results are replicated using an expost distribution of SUE, accruals and cash flows with no qualitative change in results.
The hedge portfolio tests examine if SUEs, accruals, and cash flows are efficiently priced by taking a long position in firms with extreme positive (negative) SUEs or cash flows (accruals) and a short position in firms with extreme negative (positive) SUEs or cash flows (accruals). If the hedge portfolio yields consistently positive returns in subsequent quarters, this indicates market mispricing of the current accounting information in the portfolio formation quarter.
4.2.1 Replication of postearnings announcement drift
As a check on the validity of the data and to provide a benchmark for our subsequent combined SUE and accrual (cash flow) strategies, we first implement a SUE hedge portfolio strategy in an attempt to exploit the postearnings announcement drift phenomenon. Taking a long position in firms with large positive unexpected earnings (SUE10) and a corresponding short position in firms with large negative unexpected earnings (SUE1) earns positive excess returns of 6.88% over the following two quarters, or approximately 13.8% annualized. A time plot of the average abnormal returns for each of the 36 quarters in our sample period is depicted in Figure 1. As suggested by Bernard and Thomas, if risk is an omitted variable that is driving the results then it should manifest itself as either a higher incidence of losses or as a few extremely large losses. The incidence of losses in Figure 1 is 19.4% (7 out of 36 quarters),
which is somewhat higher than the frequency of losses reported by Bernard and Thomas (1989). As noted in their study, this loss frequency compares favorably to that yielded by a unit beta portfolio, where the incidence of loss quarters is 39% over the years 19351968 (see Fama and MacBeth 1973). Moreover, for the current sample the most extreme loss is only – 2.1%, which is smaller than the most extreme loss in Bernard and Thomas (1989). Finally, losses occur only once in adjacent quarters, and the cumulative abnormal return over these two quarters is –2.85%. Thus, there is not a sustained negative effect of implementing this strategy over an extended period. The sum of all negative quarters is only 11.6% while the sum of all positive quarters is approximately 262%. These results corroborate the Bernard and Thomas (1989, 1990) findings of postearnings announcement drift that is not likely explained by omitted risk factors.
4.2.2 Examination of quarterly accrual and cash flow anomalies
To provide for a direct comparison between the SUE and accrual (cash flow) anomalies, we next attempt to replicate Sloan's (1996) annual accrual (cash flow) pricing anomalies with quarterly data.
Hedge portfolios are formed using quarterly decile classifications based on the magnitude of accruals or operating cash flows with abnormal returns accumulated over the two quarters after the portfolio formation date. The hedge portfolios are formed by taking a long position in ACC1 firms and a short position in ACC10 firms. Figure 2a depicts these twoquarter abnormal returns associated with each of the thirtysix quarters in our sample period. The average (median) twoquarter excess return accruing to this strategy is 5.56% (5.50%) with a standard error of 0.771% (significant at .001). As shown, the accrualsbased hedge portfolio yields negative returns in only 5 of 36 quarters or 13.9% of the time. Moreover, the largest negative return is 7.95% and the sum of all negative returns is only 10.7%, while the sum of all positive returns is 211%. Thus, it appears the market misprices quarterly accruals and this mispricing can be exploited to generate significantly positive abnormal returns on average.
Figure 2b presents returns to a quarterly cash flow strategy where we take a long position in CFO10 firms and a short position in CFO1 firms. The mean (median) excess return accruing to this strategy is 3.77% (3.64%) with a standard error of 0.848% (significant at .01). Figure 2b reveals that this strategy generates negative excess returns in 7 of 36 quarters or 19.4% of the time. While considerably less viable than the quarterly accruals strategy, these results nevertheless suggest that the market fails to fully impound into prices the future earnings implications of current operating cash flow information.
4.2.3 Joint SUE/accrual and SUE/cash flow strategies.
To this point, the SUE, accruals, and cash flow strategies have been examined independently of one another. If, indeed, the market’s mispricing of accruals (cash flows) is distinct from the postearnings announcement drift phenomenon, then it should be possible to form trading strategies that capitalize on both forms of mispricing to yield even larger excess returns than previously documented.
For example, some firms with large positive unexpected earnings shocks may attempt to mitigate these shocks by creating large negative (income decreasing) accruals that shift expenses forward or delay revenues. Given the negative autocorrelation tendencies for large accruals coupled with Sloan’s and our finding of market overreaction to accruals, these firms would likely exhibit positive earnings changes and positive abnormal returns in the subsequent quarter(s). Conversely, firms faced with large negative unexpected earnings shocks may attempt to smooth earnings and/or make their situation look better by using large income increasing accruals. If the market overestimates the persistence of these income increasing accruals, then these firms would be expected to exhibit negative abnormal returns in the subsequent quarter(s). Note that in both of these situations, the subsequent market correction for accruals mispricing reinforces the postearnings announcement drift that would be present in the absence of large accruals.
The results of the joint strategies are summarized in two contingency tables in Table 3 and graphically in Figure 3. Panel A of Table 3 presents the abnormal returns earned from portfolios constructed by grouping firms according to unexpected earnings and accrual realizations, while Panel B presents results where firms are partitioned by unexpected earnings and cash flow realizations. The top number in each cell represents the mean cumulative size and riskadjusted return earned over the two quarters subsequent to the portfolio formation date averaged across the 36 quarters in our sample period. The number in parentheses represents the number of firm/quarters that comprise each cell. To simplify our presentation, quintiles 2 through 4 have been condensed into a single cell, while the extreme quintiles (1 and 5) are presented separately. Shaded cells in Table 3 represent cells where both of the partitioning variables give congruent signals for future earnings and, hence, would be used in a hedge portfolio strategy. This presentation format allows for several insights into the individual and joint anomalies.
First, it is immediately apparent that joint strategies will be more profitable than the individual strategies outlined in the previous section. For example, in the unexpected earnings and accruals matrix in panel A, the average abnormal returns to the long position (SUE5/ACC1 = 5.84%) and the short position (SUE1/ACC5 = 6.11%) are each roughly as large as the total hedge portfolio returns generated by the individual strategies reported earlier (6.88% for SUE, 5.56% for accruals). Thus a hedge portfolio strategy formed by taking a long position in SUE5/ACC1 firms and a short position in SUE1/ACC5 firms will earn an abnormal return of 11.94% that is roughly double that of either individual strategy.
Second, notice that postearnings announcement drift is not present when the accruals ranking signals mispricing in the opposite direction. Specifically, firms’ returns in the lowest SUE group do not drift downward if earnings contain large income decreasing accruals (SUE1/ACC1 = 0.29%, insignificantly different from zero). Similarly, firms’ returns in the highest SUE group do not drift upward if earnings contain large income increasing accruals (SUE5/ACC5=0.76%, insignificantly different from zero). Thus, the effect of postearnings announcement drift is affected by the magnitude of accruals associated with the earnings that are announced.
Finally, note that the average abnormal returns associated with the firms falling into a given SUE grouping (column) always decrease as we move down the table while the abnormal returns for a given ACC grouping (row) always increase as we move from left to right in the table. The consistency in the pattern of abnormal returns across cells suggests that the differences in returns are more likely due to the mispricing associated with the joint SUE/accruals signals than to some omitted variable(s).
Similar observations can be made for the abnormal returns associated with unexpected earnings and cash flow groups reflected in Panel B of Table 3. Consistent with our earlier evidence that the cash flow mispricing is less pronounced, the average abnormal return to the long position (SUE5/CFO5) is 4.10% while average return associated with the short position (SUE1/CFO1) is 4.43%. Both of these average returns are smaller than their respective counterparts in Panel A. Nevertheless, a strategy combining unexpected earnings with cash flow rankings earns average abnormal returns of 8.53% over the subsequent two quarters which is larger than the abnormal returns generated by either of the individual strategies. Moreover, as was the case for accruals, the effect of postearnings announcement drift is affected by the magnitude of cash flows. As shown, firms with large positive unexpected earnings but large negative cash flows earn returns insignificantly different from zero (SUE5/CFO1 = 0.99%) as do firms with large negative unexpected earnings but large positive cash flows (SUE1/CFO5 = 1.10%).
The effect of accruals mispricing on the magnitude of the postearnings announcement drift is even more evident in Figure 3, which plots cumulative daily abnormal returns over 120 trading days starting 18 days past the quarterly earnings announcement. The dark solid lines in Figure 3 depict the standard postearnings announcement drift in the highest and lowest SUE quintiles, yielding a hedge portfolio abnormal return of approximately six percent. The abnormal return plots change substantially, however, if accruals are added as an additional partitioning variable. Heavy dashed lines in Figure 3 depict the average abnormal return associated with firmquarters in the highest (lowest) SUE quintile and the largest income decreasing (increasing) accrual quintile. Notice that firms with large positive unexpected earnings that use incomedecreasing accruals to mitigate the magnitude of the positive earnings surprise exhibit substantially larger upward drift than the average for all SUE5 firms. Similarly, firms with large negative unexpected earnings that use incomeincreasing accruals to mitigate the level of the negative earnings surprise experience a significantly larger downward drift than the average for all SUE1 firms. Finally, the thin dashed lines in Figure 3 depict the abnormal returns accruing to firms in the highest (lowest) SUE quintile and the largest income increasing (decreasing) accrual quintile. In these cases, where the earnings surprise is driven by a large accruals component, there is essentially no postearnings announcement drift. Thus, it is clear that the level of accruals embedded within an earnings surprise significantly alters the abnormal return behavior over the following quarters. As shown in Figure 3, the mispricing of accruals can either magnify or mitigate the drift in prices subsequent to earnings announcements.
Figures 4a and 4b graphically present the twoquarter average abnormal returns to implementable hedge portfolio strategies based jointly on unexpected earnings and either accruals or cash flow rankings across the 36 quarters in our sample. Figure 4a depicts quarterly abnormal returns to the joint SUE and accrual strategy (SUE/ACC strategy). As shown in the contingency table, taking a long position in firms with positive unexpected earnings and income decreasing accruals (i.e. SUE5/ACC1) and an offsetting short position in firms with negative unexpected earnings and income increasing accruals (i.e. SUE1/ACC5) earns average abnormal returns of 11.94% with a standard error of 1.68% (significant at .0001). Of the 36 quarters in our sample, 32 quarters yield positive abnormal returns while negative abnormal returns occur in only four quarters. Moreover, the sum of all positive quarters is 448% while the sum of all negative quarters is 22.5% with the bulk of the negative abnormal returns coming in the fourth quarter of 1993 (15.0%). Thus, combining the two individual anomalies clearly generates additional abnormal returns, with minimal (if any) increase in risk.
The abnormal returns accruing to the joint SUE and cash flow strategy (SUE/CFO strategy) presented in Figure 4b, while significantly positive, are less dramatic than those
observed for the joint SUE/accruals strategy. Taking a long position in firms with large positive unexpected earnings and high cash flows (i.e. SUE5/CFO5) and a short position in firms with large negative unexpected earnings and low cash flows (i.e. SUE1/CFO1) yields average abnormal returns of 8.53%, with a standard error of 1.46% (significant at .001). This strategy is negative in only 5 of the 36 quarters suggesting no increase in risk over the individual SUE or
CF strategies reported earlier. Moreover, the sum of the excess returns across all positive quarters is 324%, while the sum across all negative quarters is only –31.9%. Again, the largest single loss occurs in the fourth quarter of 1993 (16.9%). Thus, the joint SUE/CFO strategy also appears to be viable, although not as impressive as the SUE/ACC strategy. This suggests that the market mispricing of cash flow information is not as pervasive and/or as dramatic as the mispricing of accruals.
4.3 Regression based portfolio tests
A comparison between the individual and joint strategies can also be made in a regression framework. To examine the hedge portfolios using a regression approach, the following models are estimated:
CSAR_{t, t+2} = a + b _{1} SUE1 + b _{2} SUE5 + b _{3} ACC1 + b _{4} ACC5 + b _{5} SUE1*ACC1
+ b _{6} SUE1*ACC5 + b _{7} SUE5*ACC1 + b _{8} SUE5*ACC5 (11)
CSAR_{t, t+2} = a + b _{1} SUE1 + b _{2} SUE5 + b _{3} CFO1 + b _{4} CFO5 + b _{5} SUE1*CFO1
+ b _{6} SUE1*CFO5 + b _{7} SUE5*CFO1 + b _{8} SUE5*CFO5 (12)
where SUE1(5) = 1 if firm j is in the lowest (highest) SUE quintile, zero otherwise.
ACC1(5) = 1 if firm j is in the lowest (highest) accrual quintile, zero otherwise.
CFO1(5) = 1 if firm j is in the lowest (highest) cash flow quintile, zero otherwise.
The rationale for equations (11) and (12) is that the indicator variables pick up the mean effect of being in a specific quintile. Thus, by taking appropriate linear combinations of the parameters, different hedge portfolios can be examined. The benefit to this approach is that additional factors posited to affect the excess returns can be added as control variables to rule out other potential explanations for our results. Furthermore, because the regression equation examines the effect of SUE, accruals, and cash flows simultaneously, estimated abnormal returns can be computed after controlling for the correlation among the partitioning variables.
Equations (11) and (12) are estimated on a quarterly basis and the mean values over time are reported in Table 4. To mitigate potential bias, tstatistics are computed using the sampling distribution of the parameter estimates over time, and pvalues are computed using a tdistribution with 35 degrees of freedom.
The main effects that are presented would be the same as the accrual hedge portfolio if there were no significant interaction effects (although based on quintiles instead of deciles). With respect to the accrual and cash flow strategy in Panel A, the magnitude of the joint SUE and accrual strategy (11.94%) is comprised of both SUE and accrual main effects. Moreover, as indicated by the b _{6} coefficient, there is a significantly negative interaction effect of approximately –1.89% for firms with large negative unexpected earnings and large income increasing accruals. Similarly, in Panel B the magnitude of the joint SUE and cash flow strategy (8.53%) is comprised of both SUE and cash flow main effects, and there is a significantly negative interaction (b _{5} = –2.10%) for firms with large negative unexpected earnings and large negative cash flows. We also see a significantly negative interaction (b _{7} =–2.00%) for firms with large positive unexpected earnings but large negative cash flows. Thus, it appears that significant postearnings announcement drift documented in earlier studies is mitigated to a large degree when the mispricing associated with accruals (cash flows) moves in the opposite direction.
5. Diagnostic Tests
As mentioned previously, 18 days after the earnings announcement was chosen as the start of the accumulation period. While this is a conservative estimate, there may be some concern that the returns to either the cash flow based or accrual based strategy are earned by trading on information before it is actually known. If this is indeed the case, the bulk of the abnormal returns should accrue at the start of the accumulation period when seemingly unknowable information is being traded on. To explore this issue, daily cumulative abnormal returns for the joint SUE and accrual strategy are plotted over the following two quarters. The results (not presented) show a smooth upward and downward drift that appears unaffected by the starting date chosen. Results for the joint SUE and cash flow anomaly are qualitatively similar.
A problem in formulating an implementable trading strategy arises when using the hedge portfolio approach. As noted in Bernard and Thomas (1989), implementing this strategy requires taking new positions in size control portfolios on a daily basis. This raises the question of whether the abnormal returns would remain under an easily implementable zero investment strategy. To address this concern, Bernard and Thomas (1989) also use an alternative to the companion portfolio approach. Specifically, they compute abnormal returns using what they call a continuously balanced approach. This approach begins by identifying all firms that announce earnings on a given day. If firms in both extreme SUE, accruals or cash flow deciles report earnings on the same day, appropriately weighted offsetting long and short positions are taken in these firms. If firms from only one extreme portfolio announce on a given day, then no position is taken in these firms until an offsetting match in the other extreme portfolio becomes available. Buy and hold returns are then computed on each individual hedge portfolio from that point forward.
To investigate this concern, the individual accrual strategy and the joint SUE and accrual strategy (based on quintiles) are replicated using the continuously balanced approach. Untabulated results show 120day abnormal returns of 7.84% for the individual accrual strategy, and 12.91% for the joint SUE and accrual strategy. These returns are of comparable magnitude to the returns accruing to the equally weighted hedge portfolio approach reported in Table 5. Thus it appears that equivalent abnormal returns could be earned using a much less costly approach.
A final robustness check examines the threeday returns surrounding the subsequent quarterly earnings announcement dates. Bernard, Thomas, and Whalen (1997) argue that abnormal returns clustered at future earnings announcement dates are less likely to be due to omitted risk factors and more likely to reveal a market anomaly with respect to accounting signals. To examine this issue, threeday returns beginning two days prior to the COMPUSTAT earnings announcement date are computed for the following two quarterly announcements. Results show that while the earnings announcement period constitutes 5% (6/120days) of the total accumulation period, 22.0% of the abnormal returns for the quarterly accrual strategy and 21.9% of the abnormal returns for the joint SUE and accrual strategy are earned in the threeday windows surrounding the next two quarters’ earnings announcements. While these percentages are lower than the 40% reported by Sloan (1996), the difference is likely due to the difference in sample periods and the rise in voluntary preearnings disclosures being issued by firms in recent years, particularly when earnings are expected to fall below analysts' expectations. For example, The Wall Street Journal (June 23, 1997) reports that earnings guidance rose from around 250 firms in 1994 to about 2000 firms in 1996 (and more than 700 firms through the first quarter of 1997). Given that our sample spans 19881997, earnings announcements during this time period are more likely to be preempted by preannouncements than in Sloan’s sample, which ends in 1991.
As a check on the validity of earnings preannouncements as a potential explanation, returns to the long and short portfolios are examined separately, as it is more likely that bad news is disclosed early (e.g. Skinner 1994; The Wall Street Journal 1997). Results show that on average, 46.3% of returns to the long position are earned in the threeday windows surrounding the future earnings announcement dates, while essentially zero abnormal returns are earned in the threeday windows for the short (bad news) position. Thus, it appears that voluntary predisclosure of information by management, particularly for "bad news" situations, may be affecting the returns earned at the earnings announcement date.
6. Summary and Conclusion
This paper provides evidence that the market systematically misestimates the future earnings implications of the accrual and cash flow components of current quarterly earnings. The market’s overestimation (underestimation) of the persistence of accruals (cash flows) leads to mispricing that can be exploited to generate significant positive abnormal returns over the two quarters after the portfolio formation date. The accruals/cash flow mispricing appears to be largely independent of the postearnings announcement drift phenomenon widely documented in previous literature. Thus, trading on the joint SUE and accruals (cash flow) signals earns greater abnormal returns than trading on any of the individual anomalies. Moreover, there appears to be little or no additional risk associated with trading on these joint signals.
Our findings have several important implications for extant research and suggest a number of areas for future research. Information content studies relating security returns to accruals and cash flows assume correct market pricing of earnings components. Yet evidence in Sloan (1996) and this paper suggests the market systematically overprices (underprices) the accrual (cash flow) component of earnings. Thus, studies demonstrating a stronger contemporaneous association between returns and accrual earnings relative to cash flows should be reexamined in light of this mispricing. Future information content studies should explicitly control for this mispricing by allowing for possible price reversals in the research design.
Much of the extant earnings forecasting literature relies on time series models that extrapolate past earnings. Our findings suggest that one may be able to exploit the different persistence properties of accrual and cash flow components of earnings to improve upon extant forecasting models that rely exclusively on past earnings.
Understanding sources of intertemporal differences in market anomalies is largely unexplored in the literature. Our results also show that postearnings announcement drift is exaggerated or mitigated based on the level of accruals embedded within the earnings surprise. This finding suggests that intertemporal differences in postearnings announcement drift are likely due, in part, to intertemporal differences in the magnitude of accruals contained within earnings surprises.
While our results suggest that accruals, in general, are mispriced, we do not address whether there is greater mispricing for certain types of accruals (e.g., current versus longterm accruals) which warrants further attention. Moreover, investigating whether the discretionary component of accruals suffers from greater mispricing than the nondiscretionary component is another interesting area for future research. Finally, understanding what causes the market to fail to fully impound the future earnings implications of current accrual (cash flow) signals is a perplexing question that deserves further investigation.
Bibliography
Ball, R. 1992. The EarningsPrice anomaly. Journal of Accounting and Economics 15 (2/3): 319346.
Ball, R. and E. Bartov. 1996. How Naive is the stock market’s use of earnings information? Journal of Accounting and Economics 21:319337.
Banz, R. 1981. The Relationship Between Return and Market Value of Common Stocks. Journal of Financial Economics 9: 318.
Basu, S. 1977. Investment Performance of Common Stocks in Relation to Their PriceEarnings Ratios. Journal of Finance 32: 663682
Bernard, V. and J. Thomas. 1989. Post Earnings Announcement Drift: Delayed Price Response or Risk Premium? Journal of Accounting Research 27(Supplement):
136.
Bernard, V. and J. Thomas. 1990. Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics 13: 305340.
Bernard, V., J. Thomas, and J. Wahlen. 1997. Accounting Based Stock Price Anomalies: Separating Market Inefficiencies form Research Design Flaws. Contemporary Accounting Research 14(1):89136.
Brown, L. and M. Rozeff. 1979. Univariate timeseries models of quarterly accounting earnings per share: A proposed model. Journal of Accounting Research 17(Spring):179203.
Collins, D. and P. Hribar. 1999. The Balance Sheet Approach to Estimating Accruals: Implications for Empirical Research: Working paper, University of Iowa.
Drtina, R. and J. Largay. 1985. Pitfalls in calculating cash flows from operations. The Accounting Review 60(2):314326.
Easton, P. and M. Zmijewski. 1993. SEC Form 10K/10Q Reports and Annual Reports to Shareholders: Reporting Lags and Squared Market Model Prediction Errors. Journal of Accounting Research 31(1):113129.
Fama, E. and J. MacBeth. 1973. Risk, Return, and Equilibrium: Empirical Tests. Journal of Political Economy May/June:607636.
Foster, G. 1977. Quarterly Accounting Data: TimeSeries Properties and PredictiveAbility Results. The Accounting Review 52(1):121.
Foster, G., C. Olsen, and T. Shevlin. 1984. Earnings releases, anomalies, and the behavior of security returns. The Accounting Review 59(Oct) :574603.
Freeman, R. and S. Tse. 1989. The Multiperiod Information Content of Accounting Earnings: Confirmations and Contradictions of Previous Earnings Reports. Journal of Accounting Research 27(Supplement):4979.
Greig, A. 1992. Fundamental Analysis and Subsequent Stock Returns. Journal of Accounting and Economics 15(2/3): 413442.
Jaffe, J., D. Keim, and R. Westerfield. 1989. Earnings yeilds, market values, and stock returns. Journal of Finance 44(1):135148.
Mishkin, F. 1983. A Rational Expectations Approach to Macroeconomics. The University of Chicago Press, Chicago, IL.
Ou, J. and S. Penman. 1989. Financial Statement Analysis and the Prediction of Stock Returns. Journal of Accounting and Economics 11(4):295330.
Rangan, S. and R. Sloan. 1998. Implications of the Integral Approach to Quarterly Reporting for the PostEarnings Announcement Drift. The Accounting Review 73(3):353371.
Reinganum, M. 1982. A Direct Test of Roll’s Conjecture on the Firm Size Effect. Journal of Finance 31: 2735.
Rendlemanm, Jones and Latane, 1987. Further insight into the standardized unexpected earnings anomaly: Size and serial correlation effects. The financial Review 22:131144.
Revsine, L., D. Collins, and B. Johnson. 1999. Financial Reporting and Analysis. Upper Saddle River, NJ: PrenticeHall.
Skinner, D. 1994. Why firms voluntarily disclose bad news. Journal of Accounting Research Spring:3860.
Sloan, R. 1996. Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? The Accounting Review 71(3):289316.
Stattman, D. 1980. Book value and Stock Returns. The Chicago MBA: A Journal of Selected Papers University of Chicago: 2545.
Weiss, I. 1999. Earnings Management in Response to an Exogenous Nonrecurring Item, Working Paper The University of Chicago.
Wilson, P. 1987. The Incremental Information Content of the Accrual and Funds Components of Earnings after Controlling for Earnings. The Accounting Review 62(2):293322.
Table 1a. Mean (median) values of selected characteristics for ten portfolios formed quarterly on the magnitude of accruals. Sample consists of 41,237 firm quarters over the years 19881997.
Quarterly Portfolio Accrual Ranking 

Income decreasing 
Income increasing 

1 
2 
3 
4 
5 
6 
7 
8 
9 
10 

Panel A: Components of Earnings 

Accruals

0.105 (0.073) 
0.041 (0.035) 
0.027 (0.025) 
0.019 (0.017) 
0.013 (0.012) 
0.007 (0.006) 
0.001 (0.001) 
0.007 (0.007) 
0.020 (0.020) 
0.070 (0.054) 

Cash Flows

0.082 (0.069) 
0.047 (0.045) 
0.037 (0.036) 
0.030 (0.029) 
0.024 (0.023) 
0.019 (0.018) 
0.014 (0.012) 
0.007 (0.006) 
0.005 (0.006) 
0.051 (0.039) 

Standardized Unexpected Earnings (SUE)

1.39 (0.066) 
0.259 (0.048) 
0.078 (0.066) 
0.048 (0.094) 
0.084 (0.083) 
0.171 (0.122) 
0.188 (0.132) 
0.260 (0.145) 
0.292 (0.147) 
0.547 (0.207) 

Panel B: Risk Proxies 

Beta

1.08 (1.08) 
1.05 (1.05) 
1.03 (1.02) 
1.02 (1.02) 
1.00 (1.01) 
1.01 (1.02) 
1.03 (1.04) 
1.07 (1.06) 
1.09 (1.09) 
1.10 (1.09) 

Mean Size Decile 
6.6 
7.4 
7.8 
8.0 
8.0 
7.9 
7.7 
7.6 
7.3 
6.6 

Median MarkettoBook 
1.70 
1.80 
1.86 
1.84 
1.78 
1.74 
1.74 
1.80 
1.83 
1.81 

Accruals = Earnings minus cash from operations, scaled by average total assets.
Cash Flows = Cash from operations, as reported on the statement of cash flows, scaled by total assets.
SUE = Seasonally differenced quarterly earnings divided by the standard deviation of the forecast error.
Beta = firm specific coefficients from a time series regression of the individual firm return over the risk free rate on the corresponding size decile return over the risk free rate. Regressions are based on the 60 months prior to the month of the earnings announcement.
Size Decile = average decile based on classifications of all NYSE/AMEX firms, where 1 contains the smallest firms and 10 contains the largest firms.
MkttoBook = Ratio of market Value of Common Equity to Book Value of Common Equity, measured at the end of the prior quarter.
Table 1b. Mean (median) values of selected characteristics for ten portfolios formed quarterly on the magnitude of operating cash flows. Sample consists of 41,237 firm quarters over the years 19881997
Quarterly Portfolio Cash Flow Ranking 

1 
2 
3 
4 
5 
6 
7 
8 
9 
10 

Panel A: Components of Earnings 

Accruals

0.053 (0.049) 
0.012 (0.016) 
0.002 (0.004) 
0.004 (0.003) 
0.010 (0.007) 
0.014 (0.012) 
0.019 (0.017) 
0.024 (0.022) 
0.035 (0.031) 
0.077 (0.057) 

Cash Flows

0.060 (0.046) 
0.011 (0.009) 
0.002 (0.003) 
0.010 (0.011) 
0.017 (0.017) 
0.024 (0.023) 
0.031 (0.030) 
0.039 (0.038) 
0.052 (0.049) 
0.102 (0.082) 

Standardized Unexpected Earnings (SUE) 
0.200 (0.039) 
0.158 (0.041) 
0.112 (0.075) 
0.047 (0.066) 
0.040 (0.069) 
0.040 (0.106) 
0.000 (0.106) 
0.101 (0.138) 
0.121 (0.171) 
0.245 (0.221) 

Panel B: Risk Proxies 

Beta

1.12 (1.11) 
1.09 (1.09) 
1.06 (1.06) 
1.03 (1.04) 
1.01 (1.02) 
1.03 (1.04) 
1.00 (1.01) 
1.03 (1.02) 
1.05 (1.05) 
1.08 (1.05) 

Mean Size Decile 
6.3 
6.9 
7.4 
7.6 
7.8 
8.0 
8.1 
8.1 
7.9 
7.3 

Median MarkettoBook 
1.61 
1.54 
1.52 
1.56 
1.66 
1.81 
1.94 
2.09 
2.17 
2.22 

Accruals = Earnings minus cash from operations, scaled by average total assets.
Cash Flows = Cash from operations, as reported on the statement of cash flows, scaled by total assets.
SUE = Seasonally differenced quarterly earnings divided by the standard deviation of the forecast error.
Beta = firm specific coefficients from a time series regression of the individual firm return over the risk free rate on the corresponding size decile return over the risk free rate. Regressions are based on the 60 months prior to the month of the earnings announcement.
Size Decile = average decile based on classifications of all NYSE/AMEX firms, where 1 contains the smallest firms and 10 contains the largest firms.
MkttoBook = ratio of market Value of Common Equity to Book Value of Common Equity, measured at the end of the prior quarter.
Table 2. Results from nonlinear generalized leastsquares estimation of the stock price reaction to information in current financial statement information. CSAR is the cumulative size and risk adjusted return following the release of financial statements; Q_{t+i} equals earnings for quarter i; Accruals equals earnings minus cash from operations; CashFlow equals cash from operations as reported on the statement of cash flows. Variables are scaled by average total assets to ensure crosssectional comparability.
Panel A: PostEarnings Announcement Drift Specification 

Q_{t+1} = Q_{t3} + a _{0} + a _{1}(Q_{t}  Q_{t4}) + v_{t+1} CSAR_{t+1} = b _{0} + b _{1}(Q_{t+1} – Q_{t3}  a _{0}  a _{1}^{*}(Q_{t}  Q_{t4})) + e _{t+1} 

Parameter 
Estimate 
Asymptotic Std. Error 
a _{1} 
0.306 
0.005 
a _{1}^{*} 
0.093 
0.033 
b _{1} 
1.819 
0.001 
Test of Market Efficiency 
a _{1} = a _{1}^{*} 

Likelihood ratio statistic 
43.55 

Marginal Significance level 
0.001 
Panel B: Decomposition of current earnings into accrual and cash flow components 

Q_{t+1} = Q_{t3} + g _{0} + g _{1}Accruals_{t} + g _{2}CashFlow_{t}  a _{1}Q_{t4} + v_{t+1} CSAR_{t+1} = b _{0} + b _{1}(Q_{t+1}  Q_{t3}  g _{0}  g _{1}^{*}Accruals_{t}  g _{2}^{*}CashFlow_{t }+ a _{1}^{*}Q_{t4}) + e _{t+1} 

Parameter 
Estimate 
Asymptotic Std. Error 
g _{1} 
0.234 
0.006 
g _{1}^{*} 
0.295 
0.037 
g _{2} 
0.247 
0.006 
g _{2}^{*} 
0.134 
0.037 
a _{1} 
0.301 
0.007 
a _{1}^{*} 
0.092 
0.042 
b _{1} 
1.776 
0.058 
Test of Market Efficiency 
g _{1 }= g _{1}^{*} and g _{2} = g _{2}^{*} 

Likelihood ratio statistic 
113.44 

Marginal Significance level 
0.001 
Table 3. Abnormal returns for a holding period of two quarters based upon quarterly rankings of unexpected earnings (SUE) and either accruals or cash flows. Quintiles 2 through 4 have been condensed into one cell. Mean values reported are computed as a mean across 36 quarters. The number of observations per cell is reported in parentheses and is based on the total number of firm quarters in a given cell spanning the years 19881997. Shaded cells represent observations that have congruent signals for future unexpected earnings.
Panel A: Unexpected Earnings (SUE) and Accrual classifications 

SUE Quintile 

SUE1 
SUE24 
SUE5 

ACC1 
0.29% (2612) 
1.70%* (3751) 
5.84%* (1705) 
2.16%* (8068) 

Accrual Quintile 
ACC24 
2.55%* (4029) 
0.09% (16764) 
3.08%* (4156) 
0.10% (24949) 

ACC5 
6.11%* (1302) 
1.57%* (4792) 
0.76% (2126) 
1.74%* (8220) 

2.40%* (7943) 
0.02% (25307) 
3.11%* (7987) 

Panel B: Unexpected Earnings (SUE) and Cash Flow classifications 

SUE Quintile 

SUE1 
SUE24 
SUE5 

CFO1 
4.43%* (2081) 
0.45% (4253) 
0.99% (1760) 
1.19%* (8094) 

Cash Flow Quintile 
CFO24 
1.89%* (4409) 
0.00% (16328) 
3.44%* (4133) 
0.23% (24870) 

CFO5 
1.10% (1453) 
0.61% (4726) 
4.10%* (2094) 
1.22%* (8273) 

2.40%* (7943) 
0.02% (25307) 
3.11%* (7987) 

* Significantly different from zero at a =0.01 level
Table 4. Regressionbased estimates of sizeadjusted abnormal returns over the two quarters following a quarterly earnings announcement. Portfolios formed using extreme SUE and Accrual quintiles, ranked on a quarterly basis. Return accumulation begins 18 days after the quarter t earnings announcement date and ends 17 days after the quarter t+2 earnings announcement. Parameter estimates are computed as the mean estimate across all 36 quarters, and tstatistics are computed using the sampling distribution of the parameter estimates. Significance levels are based on a tdistribution with 35 degrees of freedom.
Panel A: Joint SUE and Accruals Regression Parameter Estimates 

CSAR_{t, t+2} = a + b _{1t} SUE1 + b _{2t} SUE5 + b _{3t} ACC1 + b _{4t} ACC5 + b _{5t} SUE1*ACC1 + b _{6t} SUE1*ACC5 + b _{7t} SUE5*ACC1 + b _{8t} SUE5*ACC5 + e _{t} 

Parameter 
a 
b _{1} 
b _{2} 
b _{3} 
b _{4} 
b _{5} 
b _{6} 
b _{7} 
b _{8} 

Mean Estimate: 
.0009 
.0265 
.0298 
.0160 
.0166 
.0066 
.0189 
.0114 
.0065 

(tStatistic) 
(0.19) 
(3.99)** 
(4.20)** 
(2.95)** 
(3.32)** 
(0.55) 
(1.66)* 
(0.802) 
(0.60) 

Accrual and SUE hedge portfolio: 11.94% (t=7.13)** 

Panel B: Joint SUE and Cash Flow Regression Parameter Estimates 

CSAR_{t, t+2} = a + b _{1t} SUE1 + b _{2t} SUE5 + b _{3t} CFO1 + b _{4t} CFO5 + b _{5t} SUE1*CFO1 + b _{6t} SUE1*CFO5 + b _{7t} SUE5*CFO1 + b _{8t} SUE5*CFO5 + e _{t} 

Parameter 
a 
b _{1} 
b _{2} 
b _{3} 
b _{4} 
b _{5} 
b _{6} 
b _{7} 
b _{8} 

Mean Estimate: 
.0001 
.0187 
.0345 
.0044 
.0061 
.0210 
.0017 
.0200 
.0004 

(tStatistic) 
(0.01) 
(2.65)** 
(4.94)** 
(0.75) 
(1.23) 
(1.90)* 
(0.16) 
(1.67)* 
(0.05) 

Cash flow and SUE hedge portfolio: 8.53% (t=5.83) 

* Significant at a = 0.05, one tailed
** Significant at a = 0.01, one tailed
Results are based upon 41,237 observations.
Table 5. Cumulative size and risk adjusted returns accruing to hedge portfolios. Means are computed as a mean across 36 fiscal quarters over the years 19881997. Individual strategies are based on taking offsetting long and short positions in the appropriate extreme deciles. Joint strategies are computed using intersection of extreme deciles as well as extreme quintiles.
Hedge portfolio partition 
1Q – Abnormal Returns (Standard Deviation) 
2Q – Abnormal Returns (Standard Deviation) 
Individual strategies: 

SUE 
4.24%^{*} (4.63%) 
6.88%^{*} (6.03%) 
Accrual 
2.76%^{*} (2.98%) 
5.56%^{*} (4.63%) 
Cash Flows 
2.09%^{*} (3.48%) 
3.77%^{*} (5.06%) 
Joint Strategies: 

Accruals & SUE 

Deciles 
9.35%^{*} (12.94%) 
15.87%^{*} (17.48%) 
Quintiles 
7.05%^{*} (6.01%) 
11.94%^{*} (10.06%) 
Cash Flows & SUE 

Deciles 
5.83%^{*} (7.23%) 
12.73%^{*} (14.05%) 
Quintiles 
5.05%^{*} (5.54%) 
8.53%^{*} (8.77%) 
* Significantly different from zero at the 0.01 level