greater heterogeneity of expense surprises.^ Thus, when a revenue surprise is in the same direction as the earnings surprise, the change in earnings is.
FAJ
Financial Analysts Journal Voiume 62 . Number 2 ©2006, CFA Institute
Post-Earnings-Announcement Drift: The Roie of Revenue Surprises Narasimhan Jegadeesh and Joshua Livnat The study reported here consisted of estimating earnings and sales (or revenue) surprises either with historical time-series data or with analyst forecasts. Post-earnings-announcement drift was found to be stronger when the revenue surprise was in the same direction as the earnings surprise. This result proved to he robust to various controls, including the proportions of stock held by institutional investors, arbitrage risk, and turnover (prior 60-month average trading volume). This finding is consistent with prior evidence that earnings surprises have a more persistent effect on future earnings growth when they consist of higher revenue surprises than when they consist of lower expense surprises.
ne of the most puzzling market anomalies is post-earnings-announcement drift (henceforth, we also call it simply "drift"), in which stock prices continue to move in the direction of the earnings surprise up to a year after the earnings are publicly known.^ The academic literature offers three major explanations for this phenomenon: (1) It occurs because of shifts in the risks of companies with extreme surprises, which justify higher expected returns in equilibrium; (2) it is an artifact of methodological problems; (3) investors underreact to the information conveyed during earnings announcements, or they process the information only after a delay. Recent research into the drift has investigated factors that are associated with various drift levels according to prior intuition about the effects of these factors on investors' underreaction. For example, Bartov, Radhakrishnan, and Krinsky (2000) showed that the drift is lower for companies with higher proportions of institutional investors, who are considered to be more sophisticated than individual investors and less likely to vmderreact. Mikhail, Walther, and Willis (2003) provided evidence that the drift is smaller for companies followed by the more experienced analysts, who tend to incorporate earnings surprises more fully into their forecasts and to underreact less than the less experienced analysts. Mendenhall (2004) showed that the drift is stronger for companies subject to
O
Narasimhan Jegadeesh holds the Dean's Distinguished Chair in Finance at Goizueta Business School, Emory University, Atlanta. Joshua Livnat is professor ofaccounting at Stern School of Business, New York University. 22
www.cfapubs.org
higher arbitrage risks, which is consistent with a market equilibrium in which arbitrage costs prevent arbitragers from fully eliminating any pricing errors caused by underreaction. We investigated the role of sales (or revenue) surprises in predicting differential drift levels. The analysis was motivated by earlier findings that earnings surprises have a more persistent effect on future earnings growth when the surprises consist of higher revenues than when they consist of lower expenses. For example, Ertimur, Livnat, and Martikainen (2003) showed that for short return windows around earnings announcements, market reactions are stronger for revenue surprises than for expense surprises; the cause is the greater persistence of reactions to revenue surprises and the greater heterogeneity of expense surprises.^ Thus, when a revenue surprise is in the same direction as the earnings surprise, the change in earnings is more likely to persist in future periods and a greater drift in prices is expected. When the earnings surprise is driven primarily by an expense surprise (i.e., when the revenue surprise is small), the drift may not be as strong because future information is less likely to confirm the original earnings surprise.
Prior Research and Motivation Prior studies—beginning with Ball and Brown (1968) and Jones and Litzenberger (1970), followed by Latane and Jones (1979), Foster, Olsen, and Shevlin (1984), Bernard and Thomas (1989, 1990), and Chan, Jegadeesh, and Lakonishok (1996, 1999)— documented the existence of a post-earningsannouncement drift in stock returns. In particular. ©2006, CFA Institute
Post-Earnings-Announcement Drift
stock returns do not fully impound the surprise in announced quarterly earnings immediately upon the earnings disclosure; stock returns continue to drift in the same direction as the earnings surprise for up to a year after the announcement, although most of the drift occurs around subsequent earnings announcements.'^ In his review of the drift literature, Kothari (2001) argued that the drift provides a serious challenge to the efficient market hypothesis because it has survived rigorous testing for more than 30 years and cannot be fully explained by other documented anomalies. Recent studies of the drift convincingly demonstrate that its strength varies among companies in predictable and intuitively logical ways. For example, Bartov et al. showed that the drift is smaller for companies with greater proportions of institutional investors, perhaps because institutional investors tend to be more sophisticated and are less likely to rely on the too-simplistic seasonal random walk model of earnings. Similarly, Mikhail et al. showed that the drift is smaller for companies that are followed by experienced analysts, who tend to use more sophisticated prediction models for earnings than a simple seasonal random walk. Mendenhall (2004) showed that companies subject to lower arbitrage risks have smaller drifts. Brown and Han (2000) showed that for a sample of companies whose earnings-generating process can be described by a simple (autoregressive) ARl model, a smaller drift occurs for large companies than for small companies, which have a more opaque information environment (as measured by size, institutional holdings, and analyst following). Thus, recent research efforts have been directed toward understanding the factors that are associated with differential drift levels. These studies examined the relationship between drift and company characteristics (such as arbitrage risk) and/or a company's external environment (such as composition of stockholders). We examined how drift is related to the inherent persistence of earnings. For drift to exist after the initial earnings announcement, market participants must eventually realize that their immediate reaction to the earnings surprise was insufficient. This realization is likely to occur when, for example, subsequent new information confirms the prior earnings surprise, such as when a subsequent earnings release is made. Indeed, Shane and Brous (2001) showed that post-earnings-announcement drift is consistent with investors and analysts initially underreacting to the news in earnings and then correcting their underreactions as new information arrives. So, for the same level of earnings surprise. March/April 2006
one can expect a larger drift when the surprise has a more persistent effect on future earnings growth; in that case, subsequent signals are more likely to be confirmatory than conflicting. Our approach was to predict the strength of the drift by using company-specific characteristics to identify earnings surprises that were likely to have more persistent effects on future earnings. Specifically, we used contemporaneous revenue surprises to identify companies with different levels of earnings persistence and explored the interaction between revenue surprises and earnings surprises in post-earnings-announcement drift. Based on prior evidence about the differential properties of revenue and expenses, we chose contemporaneous revenue surprise as a measure of the degree to which earnings surprises would have more persistent effects on future earnings growth. Ertimur et al. showed that earnings surprises have more persistent effects on future earnings growth when they are the result of revenue surprises than when they are the result of expense surprises. An earnings surprise accompanied by a revenue surprise in the same direction and of the same magnitude probably indicates a greater probability that future information will confirm the current earnings surprise. We expected the more persistent effect of revenue surprises on earnings growth to be associated with stronger post-earningsannouncement drif t.^ In addition, Ertimur et al. showed that the heterogeneity of expenses may also negatively affect the interpretation of earnings surprises resulting from expense surprises. For example, an extreme earnings surprise may be caused by nonrecurring expenses, such as restructuring expenses or gains/losses on the sale of fixed assets, but the extreme expenses may also be caused by unusually high (low) levels of expenses that constitute investments, such as advertising expenses, which are likely to be associated with higher (lower) future earnings (i.e., resulting in a lower persistence of the effect of earnings surprise on future earnings). Thus, the persistence of the effect of an extreme earnings surprise is likely to be lower when the earnings surprise is driven by an expense surprise than when it is driven by a revenue surprise, and the subsequent drift in returns is likely also to be lower. This conjecture leads to the following hypothesis (posited in the alternative form): Hypothesis. The abnormal returns observed in the period after earnings (and revenue) announcements are positively and significantly associated with the surprise in revenue, even after the surprise in earnings has been controlled for. www.cfapubs.org
23
Financial Analysts Journal
A similar hypothesis was developed and tested by Jegadeesh and Livnat (forthcoming 2006). That study is different from the one reported here, however, in the testing methodology. Jegadeesh and Livnat did not use analyst forecasts of revenue, as we did in this study. That study also does not control for the level of institutional holdings, arbitrage risk, and turnover—variables that have been previously shown to be related to drift.
Methodology We describe in this section the methods we used to test the hypothesis—how we estimated earnings and revenue surprises; selected the sample; grouped companies; computed abnormal returns, institutional holdings, arbitrage risk, and turnover; and carried out the statistical tests. Estimation of Earnings and Revenue Surprises. We used two approaches to compute earnings and revenue surprises. The first assumed that quarterly earnings and revenue follow a seasonal random walk with a drift process; in this approach, one estimates the parameters with rolling windows of historical data. Specifically, we first estimated the following seasonal random walk models for each quarter: E
_ oe
p
e
f\\
and (2)
where Eji is quarterly earnings (Compustat database quarterly Item 8, income before extraordinary items) and R.f is quarterly revenue (Compustat quarterly Item 2) for company / in quarter t. The variables 5.- j and e, ^ are, respectively, the drift and random noise. The superscripts e and r denote, respectively, the earnings and revenue models. The standard deviation of the error term is denoted We used data for quarters t - 21 through t - 1 to estimate these models. We defined our measure of standardized unexpected earnings (SUE) based on time-series data as^ (3)
Similarly, we defined the standardized unexpected revenue surprise estimate (SURGE) based on historical time-series data as SURGE:, = -^
24
www.cfapubs.org
(4)
The main advantage of using historical data to compute SUE and SURGE was that we could compute them for any company in the Compustat database regardless of analyst following. The only requirement was that the company have sufficient prior earnings and revenue data to estimate Equations 1 and 2. This approach does, however, have a few problems. For companies that restate their earnings after their original earnings announcements (because of such events as mergers, acquisitions, divestitures, corrections of errors, etc.), the Compustat quarterly database reports orUy the restated accounting data. Therefore, when one uses these data to compute earnings and revenue surprises, one introduces a look-ahead bias to the extent that some input data were not available to investors on the earnings announcement date. An additional problem with computing SUE by using Compustat data is that special items may be included in the earnings data used to estimate Equation 1 but investors and analysts do not include special items when they form their earnings expectations. Inclusion of the special items in the estimation, therefore, will introduce noise in the estimates of unexpected earnings. Note that both of these data-related problems are likely to cause stronger biases in the extreme SUE deciles, where most of the drift occurs. To avoid these data issues, our second approach was to estimate earnings and revenue surprises by using analyst forecasts. We followed Mendenhall (2004) and estimated SUE as actual earnings minus the mean analyst forecast of earnings, scaled by the dispersion of analyst forecasts. Similar to Mendenhall, for each quarter t and company ;, all quarterly forecasts made by analysts during the 90-day period prior to the disclosure of actual earnings constituted the nonstale, relevant forecast group.^ We defined SUE as actual EPS from I/B/E/S minus the mean analyst forecast of EPS, scaled by the standard deviation of forecasts included in the group. Like Mendenhall, we deleted company-quarters with fewer than two forecasts in the group and set the standard deviation of EPS to 0.01 if it was equal to 0. We used actual earnings reported by I/B/E/S, not the Compustat restated earnings, to compute SUE based on analyst forecasts. Because I/B/E/S adjusts earnings for special items, I/B/E/S earnings correspond to the earnings that analysts actually attempt to predict. Note also that this approach does not require a long history of earnings; hence, it is suitable for young companies as well as seasoned companies. ©2006, CFA Institute
Post-Earnings-Announcement Drift
The main problem with this approach is that it is limited to comparues that are followed by analysts. This restriction is particularly severe when one is estimating revenue surprises because analysts' revenue forecasts have been collected by I/B/E/S only since 1997 (although a few revenue forecasts are available for 1996) and even then, only a subset of the analysts who report earnings forecasts provides revenue forecasts. Because analyst revenue forecasts in I/B/E/S are available for fewer companies than analyst earnings forecasts (and many company-quarters have only one available analyst forecast in the 90-day period prior to the announcement of earnings), we defined revenue surprise differently in this approach. We defined it as actual revenue from I/B/E/S minus the mean analyst forecast of revenue in the group, scaled by actual revenue from I/B/E/S (rather than by the standard deviation across forecasts). We computed SURGE based on analysts' revenue forecasts even if only one analyst's forecast was available in the I/B/E/S database. Sample Selection. The analysis in this study was carried out for three samples of companies, which differed by the definition of earnings and revenue surprises that we used. The first analysis required historical SUE and SURGE, and we needed only Compustat data to compute these surprises. The second sample required analyst forecasts to measure earnings surprise and historical data to measure revenue surprise. This sample is smaller than the first because it is restricted to companies that were followed by at least two analysts in the 90-day period prior to the disclosure of actual earnings. The third sample required analysts' forecasts of earnings and revenue to measure the respective surprises. This sample is the smallest of the three. The other selection criteria used in this study for each quarter t are as follows: • The earnings announcement date was reported in Compustat for both quarter t and quarter t + 1 (returns were cumulated through the next earnings armouncement date). • The number of shares outstanding and the price per share were available from Compustat as of the end of quarter f - 1 . We used this information to calculate market value of equity as of quarter t-1. • The book value of equity at the end of quarter t-1 was available from Compustat and was positive. • The company's shares were traded on the NYSE, Amex, or NASDAQ. March/April 2006
•
•
•
Daily returns were available from the CRSP dataset from one day after quarter t's earnings announcement through the announcement date of earnings for quarter t + 1. Data were available to assign the company to one of the six Fama-French (1992) portfolios based on size and book value to market value (BV/MV). Both SURGE and SUE could be calculated for the quarter. Assignment to SUE and SURGE Deciles.
Because SUE and SURGE have distributions with extreme observations at the tails, most drift studies use decile ranks in the regressions. We followed this practice and categorized companies into deciles based on their SUE ranks. We scaled the decile ranks to fall between 0 and 1 and labeled the scaled decile scores DSUE. Similarly, we assigned scaled decile scores based on SURGE, which we labeled DSURGE. When we used these scaled scores as explanatory variables in the regressions, the slope coefficients could be interpreted as the return on a hedge portfolio that held long the largest SUE or SURGE decile and shorted the smallest (most negative) SUE or SURGE decile. We first grouped companies with fiscal quarter endings within a particular calendar quarter into quarter cohorts. For example, the first calendar quarter of 1999 includes all company-quarters with a fiscal quarter-end of January through March 1999. For all cohorts in quarter t, we assigned decile ranks based on the cross-sectional distribution of SUE and SURGE in quarter t - 1. We used the distribution of these variables from quarter t -1 to assign decile ranks because the quarter t distributions would not be known until the last company armounced its earnings for that quarter. Therefore, assigning cutoffs based on the quarter t distribution would introduce a look-ahead bias.'^ Cumulative Abnormal Returns. Wecomputed daily abnormal return as the raw daily return minus the daily return on a portfolio of companies matched on the basis of size (market value of equity as of June) and BV/MV (as of December). The cutoff points to determine the size+BV/MV-matched portfolios were obtained from the data library of Kenneth French and were based on a classification of the population into six (two size by three BV/ MV) portfolios.^ The daily abnormal returns were surrmied over the period from one day after the earnings announcement date through the day of the following quarterly earnings announcement. Consistent with prior studies, we excluded companies in the top and bottom 0.5 percent of the cumulative abnormal returns (CARs) from the sample. www.cfapubs.org
25
Financial Analysts Journal
Control Variables. We used institutional holding, arbitrage risk, and turnover as control variables in the regressions because prior studies have shown that the drift is correlated with these variables. Li Institutional holdings. To compute institutional holdings, we first aggregated the number of shares of company / held by all institutions at the end of quarter t - 1, as reported on all 13F filings. We obtained these data from the Thomson Financial database maintained by Wharton Research Data Services. We divided the aggregate number of shares held by institutions by the number of shares outstanding for company /' at the end of quarter t -1 to obtain the proportion of outstanding shares held by institutional investors. Following Bartov et al., we ranked companies by the proportion of institutional holdings and assigned decile ranks, which is the variable we used in the regressions. We expected the drift to be smaller for companies with a larger proportion of institutional holdings (i.e., we expected a negative association between CAR and institutional holdings). il Arbitrage risk. We followed Mendenhall (2004) by estimating arbitrage risk as 1 minus the squared correlation between the monthly return on company /' and the monthly return on the S&P 500 Index, both obtained from CRSP. We estimated the correlation over the 60 months ending one month prior to the calendar quarter-end. The arbitrage risk is the percentage of stock return variance that cannot be attributed to (or hedged by) the S&P 500 return. Mendenhall documented that the drift is smaller when arbitrage risk is smaller; hence, we expected a positive association between CAR and arbitrage risk. : i Turnover. We computed turnover as the ratio of average monthly trading volume over the 60 months ending one month prior to the calendar quarter-end to total shares outstanding at the end of quarter t-1. We obtained these data from CRSP. Prior studies used turnover as a control variable in the examination of the association between CAR and SUE. Higher turnover should reduce the costs of arbitrage; therefore, we expected a negative association between turnover and CAR. Statistical Tests. Most prior drift studies relied primarily on regression analyses in which the dependent variable was CAR and the independent variables were DSUE and the control variables. Because most subsequent drift in returns occurs for extreme SUE surprises, the control variables typically interact with DSUE. This interaction is reasonable when the implicit assumption is that the higher the level of earnings surprise, the greater the effect of the control variables. For example, when the earnings surprise is large, one expects 26
www.cfapubs.org
smaller drift as the proportion of institutional holdings grows. For small earnings surprises, the proportion of institutional holdings is likely to have a minor effect (or no effect) on drift because the initial underreaction is likely to be insignificant. To examine the effect of revenue surprises, we used the product of DSURCE (the revenue surprise decile rank) and DSUE (the earnings surprise rank) as an independent variable in the regression. We used this interaction between DSURGE and DSUE because for small earnings surprises, the subsequent drift in prices is likely to be small regardless of the revenue surprise and investors are unlikely to seek additional information to help interpret the minor earnings surprise. In contrast, when the earnings surprise is large, investors are likely to seek additional information. For large surprises, when the revenue surprise is in the same direction as the earnings surprise, future news is likely to confirm the initial announcement, resulting in larger drift. Thus, the partial derivative of the drift with respect to the revenue surprise is an increasing function of the earnings surprise, which implies an interaction term in the regression. To assess the significance of regression coefficients, we used a pooled time-series cross-sectional regression as well as average coefficients from quarterly regressions in a methodology similar to that of Fama and MacBeth (1973). In addition to the regressions, we also examined the performance of hedge portfolios formed on the basis of earnings and revenue surprises. First, we constructed an earnings surprise hedge portfolio with long positions in all stocks in the top 30 percent of the SUE distribution (SUE decile ranks of 8 through 10) and short positions in the bottom 30 percent of the SUE distribution. We reconstituted the portfolio every quarter on the basis of the SUEs for that quarter. Next, we constructed a hedge portfolio that used both earnings and revenue surprises. This portfolio was a subset of the earnings surprise portfolio. The earnings+revenue surprise portfolio took a short position in stocks with both SUE and SURGE in the bottom 30 percent of the distribution and took a long position in stocks with earnings and revenue surprises in the top 30 percent of their distributions. We reconstituted this portfolio every quarter. The mean difference in returns between these two portfolios provides a measure of the extent to which revenue surprises have predictive power that adds to the power of earnings surprises. If the revenue surprises help identify stocks for which the effect of earnings surprises on future earnings is more persistent, the drift for companies with both earnings and revenue surprises in the same direction should be significantly larger than it is for companies with only earnings surprises. ©2006, CFA Institute
Post-Earnings-Announcement Drift
Sample Description
Table 1 provides summary statistics for these three samples. As can be seen in all three panels, the mean SUE is negative, although the median (the 50th percentile) is positive, which is consistent with a higher magnitude of negative earnings surprises, some of which were charges taken at management's discretion (restructuring charges). These statistics are similar to those reported in Bartov et al. In contrast. Table 1 reports the mean SURGE as positive for all three samples, as is the median when estimated from historical data (Panels A and B), which is consistent with sequential revenue increases for most companies. However, the median
The sample of companies with SUE and SURGE based on historical data begins in the second quarter of 1987 with 1,739 companies and ends in the third quarter of 2002 with a sample size of 3,557 companies. The sample of companies with SUE based on analyst forecasts of EPS but SURGE based on historical data begins in the second quarter of 1989 with 731 companies and reaches 1,526 companies in the third quarter of 2002. Finally, the sample for which analyst forecasts were used to compute both SUE and SURGE begins with 175 companies in the third quarter of 1998 and reaches 1,116 companies in the third quarter of 2002. Table 1. Summary Statistics
Percentili Variable
N
Mean
Std. Dev.
10th
25th
A. SUE and SURGE computed by using historical data, Q2 1987-Q3 2002 SUE 164,401 -0.185 17.607 SURGE
164,401
Proportion of institutional holdings Arbitrage risk
160,878
Average turnover (of outstanding shares) Market value of equity at f - 1 ($ millions) Book value of equity at f - 1 ($ millions) CAR (%)
50th
75th
90th
-1.771
-0.596
0.041
0.643
1.614
1.933
-1.672 0.048
0.551
2.006 0.704
0.665
0.798
0.068 0.336 0.907
0.970
0.243 0.135
-0.788 0.146
145,131
0.153 0.359 0.864
0.970
0.994
145,131
0.066
0.069
0.014
0.025
0.045
0.082
0.142
164,401
2,011.1
11,439.4
164,401
6.2
532.3
0.7
1.1
1.7
2.9
5.1
164,401
-0.4
22.7
-25.4
12.6
-1.1
10.5
24.8
13.8
41.1
170.4
810.1
B. SUE computed by using analysts' forecasts; SURGE computed by using historical data, Q2 1989-Q3 2002 SUE 65,377 -0.229 3.263 -2.006 -0.677 0.038 SURGE 65,377 0.179 2.088 -1.756 -0.825 0.069 Proportion of institutional holdings 64,956 0.519 0.204 0.234 0.368 0.532 Arbitrage risk 57,945 0.818 0.147 0.601 0.733 0.854 Average turnover (of outstanding shares) 57,945 0.080 0.072 0.022 0.035 0.059 Market value of equity at ( - 1 ($ millions) 65,377 4,482.9 17,602.7 117.7 291.5 864.4 Book value of equity at f - 1 ($ millions) 65,377 4.0 65.4 1.1 1.5 2.2 CAR (%) 65,377 -0.7 20.7 -24.4 -11.5 -0.4 C. SUE and SURGE computed by using analysts' forecasts, Q3 1998-Q3 2002 SUE 9,132 -0.273 3.669 -2.230 SURGE 9,132 0.009 1.892 -2.066 Proportion of institutional holdings 9,048 0.617 0.199 0.333 Arbitrage risk 7,860 0.858 0.106 0.710 Average turnover (of outstanding shares) 7,860 0.128 0.096 0.038 Market value of equity at f - 1 ($ millions) 9,132 9,727.1 31,391.2 201.5 Book value of equity at f - 1 ($ millions) 9,132 4.6 7.0 1.2 CAR (%) 9,132 0.8 26.3 -30.7
3,166.4
0.677
1.717
1.069 0.674 0.936
2.168 0.779 0.979
0.099
0.166
2,748.7
8,178.0
3.5 10.2
22.5
5.9
-0.817
0.031
0.708
-1.053
-0.114
0.948
0.490 0.795
0.645 0.879
0.769 0.940
0.851
0.061
0.101
0.166
0.258
1.763 2.124 0.977
489.1
1,451.4
5,351.1
18,403.6
1.8 -13.2
2.9
5.1 15.7
30.8
1.5
8.9
Note: N is number of company-quarters.
March/April 2006
www.cfapubs .org
27
Financial Analysts Journal
SURGE estimated from analysts' revenue forecasts (Panel C) is negative, possibly because analysts are overly optimistic in their revenue forecasts. The mean percentage of shares held by institutions for the sample based on historical data (36 percent. Panel A) is smaller than the 41 percent institutional holdings reported by Bartov et al. (Bartov et al. used only NYSE and Amex stocks, whereas we included NASDAQ stocks, which have less institutional ownership.) The institutional holdings are much higher for the samples reported in Panels B and C, which take analyst coverage into account, because analysts tend to follow stocks that are of interest to institutional investors. The mean arbitrage risk for the three samples in Table 1 is about 84 percent, which implies that the mean R^ of regressing the stock return on the S&P 500 would be about 16 percent, consistent with results reported in prior studies. The average monthly turnover is about 6.6 percent for the sample that used historical data but is slightly higher for the two samples that used analyst forecasts of earnings, probably because analysts tend to follow higher-turnover stocks. Table 1 also clearly shows that the historical data sample has a wider range of market capitalizations than the other two samples because it includes small companies that do not have any analyst following. The sample of stocks with orUy analysts' earnings forecasts (Panel B) has larger market values than the sample in Panel A, and the subset of stocks with analysts' forecasts of both earnings and revenue (Panel C) has the largest average market values. Finally, the mean CAR is small and negative in Panel A and Panel B but is a positive 0.8 percent in Panel C, possibly because of the specific period for this sample (1998-2002).
Results We obtained estimates from the following three regressions: jt = a + b(DSUEj,) + e^-,,,
(5)
j t = a + b(DSUEj;) + c[{DSURGEj ,)(DSUEj^,
(6)
and = a + KDSUEj^i) + c[{DSURGEj^,)(DSUEj,)] + d[iInst.holdingSjt){DSUEj,)] + e[(Arb.riskjt){DSUEjt)] +f [{Turnoverjt){DSUEj^,)] +Ey,(, (7)
where Inst.holdingsjf is institutional holdings, Arb.riskji is arbitrage risk, and Turnoverj^ is turnover 28
www.cfapubs.org
for company / at time t (with calculation of these variables as explained in the previous section). Regression Equation 5 examines the relationship between drift and earnings surprise. Regression Equation 6 examines the incremental effect of revenue surprises by measuring the interaction of SURGE and SUE, and regression Equation 7 includes the control variables. Consistent with prior studies, in regression Equation 7, the control variables interact with earnings surprise. Table 2 provides the estimates from the regressions. We fit the regressions within each of the three samples, but for brevity, we report only the results for the first sample (historical data, given in Panel A) and the third sample (analyst forecasts of earnings and revenue, given in Panel B). In addition to the standard pooled regression results. Table 2 also reports results of using the Fama-MacBeth (F-M) methodology. For the F-M approach, we fit the regressions separately for each quarterly cohort, and we report the time-series averages of the quarterly regression coefficients. Because of the shorter sample period for the third sample, the results in Panel B should be interpreted with caution. Note that the number of observations is smaller when we used the control variables because we excluded observations with missing data for any of the control variables. As can be seen in Panel A of Table 2, for the sample that used surprises calculated from Compustat data, the coefficient on DSUE has the predicted positive sign and is statistically different from zero at the 0.1 percent level. It is also similar in magnitude to that reported by Bartov et al. When we added DSURGE into the regression as an interactive term with earnings surprise (Equation 6), the slope coefficient on DSUE fell by 1.5 andl.8 percentage points for, respectively, the pooled regression results and the F-M results. The coefficient on the earnings+revenue surprise interaction term, DSURGE, is also positive, and it is significantly different from zero. The estimate of this coefficient is roughly 2.4 or 2.9 percent (depending on the estimation method), which implies that a trading strategy that shorts the smallest DSUE+DSURGE stocks and goes long the stocks with the largest DSUE+DSURGE can earn an incremental abnormal return (after earnings surprise is controlled for) of about 1.1 percentage points during the subsequent quarter, or about 4 percentage points annually. This increase in abnormal returns is an economically meaningful amount beyond the abnormal return that can be earned by using earnings surprise alone. The results for Equation 7 in Panel A indicate that DSURGE remained economically significant after we controlled for institutional holdings (negatively associated with drift, as expected), arbitrage ©2006, CFA Institute
Post-Earnings-Announcement Drift
Table 2.
Regression Results: CAR for Earnings and Revenue Surprises
Equation
N
Intercept
Expected sign:
DSUE
DSURGE
Institutional Holdings
+
+
_
Arbitrage Risk Turnover +
A. DSUE and DSURGE computed by using historical data, Q2 1987-Q3 2002 Equation 5: CARj, = a + b(DSUEj,) + e^, Earnings only 164,400 -2.933 5.640 Sigrvificance 0.001 0.001 Earnings only (F-M) -3.058 5.422 Significance (F-M) 0.001 0.001 Equation 6: CARjj = a + b{DSUEjt) + c[{DSURGEj,){DSUEj,)] Earnings and revenue 164,400 -2.833 4.177 Significance 0.001 0.001 Earnings and revenue (F-M) -2.938 3.643 Significance (F-M) 0.001 0.001
+ EJJ 2.386 0.001 2.918 0.001
Equation 7: CARjt = a + b(DSUE,,) + c[{DSURGEj ,)(DSL7E-,)] + d[{Inst.holdings: ()(DS!iE,-,)] +f [{Turnoverj,,){DSUEij)] + ejj '' '' Earnings, revenue, and 142,420 -2.887 7.642 controls 2.965 -0.725 Significance 0.001 0.001 0.001 0.001 Earnings, revenue, and controls (F-M) -3.104 7.855 3.128 -0.728 Significance (F-M) 0.001 0.001 0.001 0.001
R^
significance
0.005
0.001
0.005
0.001
_
+ e[{Arb.risk: ,)(DSUE,-,)] '' '' 0.056 0.006
-0.893 0.006
0.067 0.232
-1.714 0.016
0.008
B. DSUE and DSURGE computed by using analysts' forecasts, Q3 1998-Q3 2002 Equation 5: GARj, = a + b{DSUEj,) + E,- ( Earnings only 9,131 -1.569 5.019 0.003 Significance 0.004 0.001 Earnings only (F-M) -0.131 4.633 Significance (F-M) 0.936 0.026 Equation6: CARj, = a + b{DSUEjj) + c[{DSURGEj,)(DSUEj,)] + Ey_, Earnings and revenue 9,131 -1.399 2.513 4.483 0.003 Significance 0.011 0.092 0.024 Earnings and revenue (F-M) 0.205 -0.053 8.316 Significance (F-M) 0.898 0.983 0.044 Equation7: CARj, = a + b{DSUEjj) + c[(DSURGEj,)(DSUEjj)] + d[{Inst.holdingSj,){DSUEj,)] + e[{Arb.riskj,){DSUEj,)] Earnings, revenue, and controls Significance Earnings, revenue, and controls (F-M) Significance (F-M)
7,788
-0.897 0.122
-1.722 0.579
3.913 0.057
0.104 0.947
-5.250 0.370
7.631 0.073
'
0.445 0.207
0.126 0.518
0.025 0.982
0.684 0.239
0.449 0.190
-3.621 0.341
0.003
0.001
0.001
0.001
0.001
Notes: N is number of company-quarters. The entries for DSURGE represent the interaction between DSUE and DSURGE. The table reports results of a pooled company-quarter regression as well as quarter-by-quarter regressions summarized according to the methodology of Fama and MacBeth, denoted F-M in the table. Significance is based on f-statistics for regression coefficients and F-tests for significance of the R^ values. The R^ values are for tests that the entire regression model has significant explanatory power. Entries in boldface are statistically different from zero at the 10 percent level or better.
risk (positively associated with drift, as expected), and turnover (negatively associated with drift, as expected). Therefore, we conclude that using revenue surprises in addition to earnings surprises can significantly improve profits relative to profits from a trading strategy that uses earnings surprises alone as a signal. Panel B provides the results of these regressions for the sample of companies with analyst March/April 2006
forecasts of both earnings and revenues. The earnings surprise (DSUE) has a coefficient of magnitude similar to that of the DSUE computed with Compustat data (see Panel A) when DSUE is the only independent variable in the regression (Equation 5). Because the sample in Panel B contains only observations from 1998 forward, the evidence in this panel indicates that drift exists in recent periods as well as earlier periods. www.cfapubs.org
29
Financial Analysts Journal
Panel B also shows that wheri interaction with the revenue surprise (DSURGE) is added into the regression (Equation 6), the earnings surprise (DSUE) coefficient loses significance but the interaction with the revenue surprise (DSURGE) is larger in magnitude than DSUE and is significantly different from zero. The significance of the interaction with the revenue surprise continues to hold even after institutional holdings, arbitrage risk, and turnover are controlled for (Equation 7), all of which have now become insignificantly different from zero.^ Thus, the results in Table 2 are consistent with
our studied hypothesis and support the role of revenue surprise in extreme earnings surprises. Tabie 3 provides the mean CAR, the associated significance level of the CAR, and the number of observations for a 3 x 3 table of companies classified according to DSUE and DSURGE, in which the classifications to DSUE and DSURGE were independent of each other. We assigned companies to three groups based on their earnings and revenue surprises, with the bottom 30 percent in one group, the second 40 percent in another, and the top 30 percent in the third group. We expected the bottom
Table 3. Distribution of Abnormal Returns: SUE+SURGE Groups DSURGE DSUE
Bottom 30%
Middle 40%
Total
Top 30%
A. SUE and SURGE computed by using historical data, Q2 1987-Q3 200. Bottom 30% CAR (%) Significance N Middle 40% CAR (%)
-2.38 0.001
-2.32 0.001
-2.01 0.001
25,254
16,796
7,892
49,942
-1.13
-0.41
-0.05
-0.49
\ I
-2.30 0.001
i j
0.001 16,232
30,361
0.766 17,886
0.001 64,479
CAR (%) Significance
0.57
1.52
2.07
0.029
0.001
0.001
1.64 0.001
j
N
7,903
17,824
24,253
49,980
!
-0.37
0.67
0.001 64,981
-0.39 0.001
!
0.001 50,031
164,401
Significance N
0.001
Top 30%
Total CAR (%) Significance N
-1.50 0.001 49,389
B. SUE and SURGE conip uted by using analysts' forecasts, Q3 1998-Q3 2002 Bottom 30% CAR (%)
-0.25
0.15 0.867 897
2.95
0.062 1,050
0.063 355
0.669 2,302
-0.03 0.967
-0.57 0.312
0.20
1,114
1,972
0.805 880
-0.25 0.531 3,966
2.41
Significance
2.10 0.067
0.001
4.48 0.001
3.11 0.001
N
496
1,326
1,042
2,864
Significance N Middle 40% CAR (%) Significance N Top 30% CAR (%)
[Total CAR (%) Significance N
-1.67
1 -0.28 0.587 2,660
0.53 0.187 4,195
2.59 0.001 2,277
0.80 0.003 9,132
Notes: N = number of companies in a particular cell. Significance is based on f-statistics. Entries in boldface are statistically different from zero at the 10 percent level or better.
30
www.cfapubs.org
©2006, CFA Institute
Post-Earnings-Announcement Drift
30 percent group to have negative drift (CAR), the top 30 percent to have positive drift, and the middle 40 percent to have insignificant drift. The shaded column on the right of Panel A in Table 3 reports that through the next earnings announcement, the bottom 30 percent SUE portfolio had a mean drift (CAR) of -2.30 percent, the middle 40 percent SUE portfolio had a mean drift of -0.49 percent, and the top 30 percent SUE portfolio had a mean drift of 1.64 percent. In contrast, the bottom 30 percent SURGE portfolio (see the shaded rows in Panel A) had a mean drift of -1.50 percent; the middle 40 percent, a mean drift of-0.37 percent; and the top 30 percent, a mean drift of 0.67 percent. Thus, the spread in the earnings surprisebased drift is larger than in the revenue surprisebased drift for this sample. Further examination of Panel A in Table 3 shows that more companies are in groups along the diagonal than outside the diagonal. This result is to be expected because SUE and SURGE were sorted independently and companies with extreme earnings surprises were also more likely to have extreme revenue surprises in the same direction. We found that revenue surprise can help identify future return drift. For example, for the bottom 30 percent of the earnings surprise stocks in Panel A, the mean drift is -2.38 percent, but it is only -2.01 percent for those (7,892) stocks in the top 30 percent of the revenue surprise group. Similarly, although the average drift is 1.64 percent for the top 30 percent of earnings surprise stocks, it is only 0.57 percent for the (7,903) observations in the bottom 30
percent of the revenue surprise stocks, but it is 2.07 percent for those stocks in the top 30 percent of revenue surprises as well as earnings surprises. A similar picture, albeit more extreme, can be seen in Panel B of Table 3. In this sample, the mean drift of the group with high positive earnings+ revenue surprises based on analysts' forecasts is 4.48 percent, compared with a mean drift of -1.67 percent for the group with earnings+revenue surprises in the bottom 30 percent. This result is comforting because, in practice, obtaining returns on long positions is much easier than obtaining returns on short positions (if most of the drift were derived from short positions, practitioners would have more difficulty implementing the investment strategy).^^ The results in Table 3 are also economically significant. The returns shown are mean quarterly returns, so the annual returns to the combined SUE+SURGE hedge portfolio could be significant— 2.07 percent plus 2.38 percent per quarter in Panel A, or about 19 percent annually, and 4.48 percent plus 1.67 percent per quarter in Panel B, or about 27 percent annually. Given the risk of the strategy, actual application of the hedge portfolio should consider the variability of the returns among quarters. To address this issue. Table 4 provides information about the average quarterly returns and standard deviations of the returns for the hedge portfolio based on earnings surprises (Column 1) and for the hedge portfolio based on eamings-i-revenue surprises (Column 2). The earnings (earnings+revenue) surprises strategy is based on holding short positions in the bottom
Table 4. Hedge Portfolio Returns
Statistic
Earnings-Based Hedge Portfolio (1)
Eamings+ Revenue-Based Hedge Portfolio (2)
A. SUE and SURGE computed by using historical data, Q2 1987-Q3 2002 CAR (%) 1.944 2.210 Standard deviation of CAR (%) t-Statistic Significance level Average number of companies
1.159
1.833
13.31 0.001
9.57 0.001
1,586
786
B. SUE and SURGE computed by using analysts' forecasts, Q3 1998-Q3 2002 CAR (%) 2.199 3.620 Standard deviation of CAR (%) 3.331 4.611 t-Statistic 2.72 3.24 Significance level 0.015 0.005 Average number of companies 302 122
Difference (2-1) 0.266 0.917 2.30 0.025
1.421 2.620 2.24 0.040
Notes: The hedge portfolio assumed long positions in the top 30 percent and short positions in the bottom 30 percent of companies sorted according to SUE (Column 1) and SUE+SURGE (Column 2). Significance is based on f-statistics. Entries in boldface are statistically different from zero at the 5 percent level or better.
March/April 2006
www.cfapubs.org
31
Financial Analysts Journal
30 percent of SUE (SUE+SURGE) and long positions in the top 30 percent of SUE (SUE+SURGE), where positions were closed each quarter. As can be seen in Panel A of Table 4, based on the 63 observed quarters, the mean historical earnings-based portfolio yielded an average quarterly profit of 1.9 percent, with an associated significance level of 0.1 percent. The earnings+ revenue hedge portfolio, however, earned a higher return—2.2 percent quarterly—that is also significantly different from zero. The difference between this portfolio return and the return on the earnings-based hedge portfolio is about 0.27 percentage points per quarter, significantly different from zero at the 2.5 percent level. Panel B of Table 4, which presents the results for SUE and SURGE based on analysts' forecasts and reflects only 17 quarters, shows an even larger difference in returns. Note, however, that the significant improvement in the performance of the earnings+revenue hedge portfolio in this sample comes at a cost—namely, the SUE+SURGE portfo-
lio has substantially fewer companies, on average, than the earnings-based hedge portfolio. The improvement in the performance of the SUE+SURGE hedge portfolio comes from the elimination of companies with conflicting earnings and revenue signals, which eliminates stocks that are likely to have smaller drift.^^ The portfolio of extreme revenue surprises and earnings surprises nevertheless contains a sufficient number of companies for adequate portfolio diversification . Table 5 reports regression results similar to those in Table 2 for three subperiods: 1987-1992, 1993-1997, and 1998-2002. The sample observations are all based on SUE estimated from Compustat by using the seasonal random walk model for both earnings and revenue. As can be easily seen, the earnings drift is present and significant in all three subperiods; Table 5 provides no indication that the effect has decreased in the most recent period from that of prior periods. Notice that this 1998-2002 period includes times of steep market increases and decreases. The incremental revenue effect is present
Table 5. Regression Results: CAR for Earnings and Revenue Surprises—Seasonal Random Walk Model, Various Subperiods N
Observations
DSUE Intercept Earnings +
Expected sign:
DSURGE :Institutional Holdings Revenue +
-
Arbitrage Turnover Risk +
R^
Significance
0.009
0.001
0.010
0.001
0.014
0.001
0.006
0.001
0.006
0.001
0.008
0.001
0.004
0.001
0.004
0.001
0.005
0.001
-
A. Prior to 1993 -3.448
6.546
0.001
0.001
-3.319
4.575
0.001
0.001
-3.417 0.001
9.445 0.001
50,670
-3.442 0.001
4.859 0.001
50,670
-3.319 0.001
3.159 0.001
2.714 0.001
Earnings, revenue, and controls 44,392 Significance
-3.507 0.001
5.749 0.001
3.371 0.001
-2.248
5.681
0.001
0.001
69,207
-2.176 0.001
4.596 0.001
1.790 0.010
Earnings, revenue, and controls 58,301
-2.164 0.001
8.025 0.001
2.467
Earnings only
44,521
Significance Earnings and revenue
44,521
Significance Earnings, revenue, and controls 39,725 Significance B.1993-1997 Earnings only Significance Earnings and revenue Significance
3.233 0.001 3.367
-0.914
0.003
-0.107
0.001
0.001
0.970
0.882
-0.512 0.001
0.034
-1.644
0.555
0.001
C. 1998-2002 Earnings only
69,207
Significance Earnings and revenue Significance
Significance
0.004
-0.750 0.001
0.034
-0.524
. 0.318
0.328
Notes: Pooled company-quarter regressions. Significance is based on f-statistics for regression coefficients and F-tests for R^ values. The R^ values are for tests that the entire regression model has significant explanatory power. Entries in boldface are statistically different from zero at the 5 percent level or better.
32
www.cfapubs.org
©2006, CFA Institute
Post-Earnings-Announcement Drift
and statistically different from zero in all three subperiods. It seems to contribute slightly less after 1998, only about 1.8 percentage points per quarter beyond the earnings effect. Nevertheless, it is an economically significant additional return. The results for the earnings+revenue surprises are present in all three subperiods after we controlled for institutional holdings, arbitrage risk, and turnover. Thus, the results of this study are not driven by any subperiod and are consistently observed in different market conditions. They are also observed for earnings and revenue surprises based on historical estimates and on analyst forecasts.
Sensitivity Analysis To assess the effect of removing small companies with poor information and trading environments, we repeated the regression tests in Panel A of Table 2 for companies with market values in excess of $100 million at the prior quarter's end. The incremental drift based on SURGE was still significantly positive. The institutional holdings variable became positive and significant, but the arbitrage risk variable became insignificant. In addition, to assess the effect of removing NASDAQ stocks, we repeated the regression tests reported in Panel A of Table 2 for only companies traded on the NYSE and Amex. The incremental drift from SURGE was still positive and significantly different from zero. The institutional holdings variable became positive and significant. We found the results of the study to be qualitatively similar for a subsample of companies with more than one analyst forecast of revenue, although significance levels declined somewhat.
Summary and Conclusions We found that the magnitude of the observed drift in security returns after the announcement of earnings depends on the contemporaneous magnitude of the revenue (or sales) surprise. When the two signals confirm each other, the magnitude of the drift is larger, quite likely because revenue surprises identify companies for which earnings surprises should have a more persistent effect on future earnings growth. Investors and other market participants can use this evidence to improve trading strategies. Practitioners who base their portfolio decisions partly on earnings surprises should take into account revenue surprises as well. Fundamental security analysis in academe and practice should incorporate detailed analysis of the persistence of the effect of prior revenue surprises on the company's earnings growth and should assess the effects on security prices. Research efforts to understand and investigate post-earrungs-announcement drift, its causes and its effects, should take into account the extent to which changes in earnings are persistent. We used revenue surprises to identify earnings persistence, but future research could focus on identifying other variables. We gratefully acknowledge the contribution of Thomson Financial for providing forecast data through I/B/E/S. These data were provided as part of a broad academic program to encourage earnings expectations research. We also thank Shai Levi, Suresh Radhakrishnan, Stephen Ryan, and Dan Segal for their comments on an earlier version of this article. This article qualifies for 1 PD credit.
Notes 1,
A reversal typically occurs when earnings are announced for the same quarter in the following year, 2, In Ertimur et al,, as in our study, the term "expenses" includes all items listed on the income statements between "sales" and "net income before extraordinary items"; thus, expenses consist of operating expenses, financial expenses, gains or losses on disposition of long-term assets, and nonrecurring items, 3, For other drift-related studies, see Bartov (1992), Ball and Bartov (1996), and Bartov et al. For the relationship of drift to analysts' forecasts, see Abarbanell and Bernard (1992), Evidence that analysts may not fully incorporate past information in their forecasts is presented in Lys and Sohn (1990), Klein (1990), Abarbanell (1991), and Mendenhall (1991). 4, One could argue that a stronger market reaction at the time of the preliminary earnings announcement to earnings surprises that are accompanied by sales surprises of the same
March/April 2006
5, 6, 7,
sign should be associated with lower subsequent drift because the initial market underreaction was smaller. As Livnat and Mendenhall (2004) pointed out, this argument assumes that the total (immediate and drift) market reaction is fixed, but they presented evidence consistent with a stronger initial market reaction and drift for earnings surprises calculated by using analyst forecasts than for reactions and drift calculated by using historical time-series forecasts. This result implies that the total market reaction is not fixed and that stronger initial market reaction may be associated with stronger subsequent drift, Earlier studies by Bartov (1992) and Chan et al, (1996) also used this definition of SUE. This group includes only the most recent forecast made by a specific analyst within this period, In assigning companies to deciles, most researchers rely on Bernard and Thomas (1990), who reported that the drift is www,cfapubs,org
33
Financial Analysts Journal
8. 9,
insensitive to the assignment of companies into a SUE decile on the basis of the current quarter's SUE values instead of using the SUE cutoffs from quarter f - 1, The library is at mba.tuck,dartmouth,edu/pages/faculty/ ken,french/data_library,html, Results when the sample was for earnings based on analyst forecasts and revenues based on historical data (not detailed here) indicate that the interactive DSURGE term had a positive and significant coefficient in the pooled regression even after the effects of institutional holdings, arbitrage risk, and turnover were controlled for. This coefficient was positive but not significant in the Fama and MacBeth regressions, however, possibly because of the added noise in the individual quarterly regressions.
10, A similar picture emerged for the sample based on analysts' forecasts of earnings but revenue based on historical data. The group of companies with top 30 percent SUE+top 30 percent SURGE had a mean quarterly return of 2,53 percent, as compared with 1,95 percent for all companies in the top 30 percent SUE group. Companies in the bottom 30 percent SUE+SURGE group, however, did not have a higher mean return than those in the bottom 30 percent SUE group, 11, For the subsample with SUE based on analysts' forecasts but SURGE based on historical data, the hedge portfolio based on SUE+SURGE earned an average quarterly return of about 0,25 percentage points more than the return on the hedge portfolio that was based only on SUE. Because of a greater variance of these differences among quarters, however, the significance level of this difference was only 18,7 percent.
References Abarbanell, J, 1991, "Do Analysts' Earnings Forecasts Incorporate Information in Prior Stock Price Changes?" Journal of Accounting and Economics, vol, 14, no, 2 0une);147-165, Abarbanell, J,, and V, Bernard, 1992, "Tests of Analysts' Overreaction/Underreaction to Earnings Information as an Explanation for Anomalous Stock Price Behavior," Journal of Finance, vol, 47, no, 3 (fuly):1181-1207. Ball, R,, and E, Bartov, 1996, "How Naive Is the Stock Market's Use of Earnings Information?" Journal of Accounting and Economics, vol, 21, no, 3 0une):319-337, Ball, R,, and P, Brown, 1968, "An Empirical Evaluation of Accounting Income Numbers," Journal of Accounting Research, vol, 6, no, 3 (Autumn):159-177, Bartov, E, 1992, "Patterns in Unexpected Earnings as an Explanation for Post-Announcement Drift," Accounting Review, vol, 67, no, 3 auly):610-622. Bartov, E,, S, Radhakrishnan, and I, Krinsky, 2000, "Investor Sophistication and Patterns in Stock Returns after Earnings Announcements," Accounting Review, vol, 75, no, 1 Qanuary): 43-63, Bernard, V,, and J, Thomas, 1989, "Post-Earnings-Announcement Drift: Delayed Price Response or Risk Premium?" Journal of Accounting Research, vol, 27 (Supplement):l-48, , 1990, "Evidence That Stock Prices Do Not Fully Reflect the Implications of Current Earnings for Future Earnings," Journal of Accounting and Economics, vol, 13, no, 4 (December): 305-340, Brown, L,, and J, Han, 2000, "Do Stock Prices Fully Reflect the Implications of Current Earnings for Future Earnings for ARl Firms?" Journal of Accounting Research, vol, 38, no, 1 (Spring):149-164, Chan, L,K,C,, N, Jegadeesh, and J, Lakonishok, 1996, "Momentum Strategies," Journal of Finance, vol, 51, no, 5 (December):1681-1713, , 1999, "The Profitability of Momentum Strategies," Financial Analysts Journal, vol, 55, no, 6 (November/December): 88-90, Ertimur, Y,, J, Livnat, and M, Martikainen, 2003, "Differential Market Reactions to Revenue and Expense Surprises," Review of Accounting Studies, vol, 8, nos, 2-3 (June):185-211, Fama, E,F,, and K, French, 1992, "The Cross-Section of Expected Stock Returns." Journal of Finance, vol, 47, no, 2 (June):427-465,
34
www,cfapubs,org
Fama, E,F,, and James D, MacBeth, 1973, "Risk, Return, and Equilibrium: Empirical Tests," Journal of Political Economy, vol, 81, no, 3 (May/June):607-636, Foster, G,, C, Olsen, and T, Shevlin, 1984, "Earnings Releases, Anomalies and the Behavior of Security Returns," Accounting Review, vol, 59, no, 4 (October):574-603, Jegadeesh, Narasimhan, and Joshua Livnat, Forthcoming 2006, "Revenue Surprises and Stock Returns," Journal of Accounting and Economics.
Jones, CP,, and R,H, Litzenberger, 1970, "Quarterly Earnings Reports and Intermediate Stock Price Trends." Journal of Finance, vol, 25, no, 1 (March):143-148, Klein, A, 1990, "A Direct Test of the Cognitive Bias Theory of Share Price Reversals," Journal of Accounting and Economics, vol, 13, no, 2 (July):155-166, Kothari, S,P, 2001, "Capital Markets Research in Accounting," Journal of Accounting and Economics, vol, 31, nos, 1-3 (September):105-231, Latane, H,A,, and CP, Jones, 1979, "Standardized Unexpected Earnings—1971-1977," Journal of Finance, vol, 34, no, 3 (June):717-724, Livnat, Joshua, and Richard Mendenhall, 2004, "Why Is the PostEarnings-Announcement Drift Larger for Surprises Calculated from Analyst Forecasts?" Working paper. New York University, Lys, T,, and S, Sohn. 1990, "The Association between Revisions of Financial Analysts' Earnings Forecasts and Security Price Changes," Journal of Accounting and Economics, vol, 13, no, 4 (December):341-363, Mendenhall, R, 1991, "Evidence on the Possible Underweighting of Earnings-Related Information," Journal of Accounting Research, vol, 29, no, 1 (Spring):170-179, , 2004, "Arbitrage Risk and Post-Earnings-Announcement Drift," Journal of Business, vol. 77, no, 4 (October):875-894, Mikhail, Michael B,, Beverly R, Walther, and Richard H, Willis, 2003, "The Effects of Experience on Security Analyst Underreaction," Journal ofAccounting and Economics, vol, 35, no,l Shane, Philip, and Peter Brous, 2001, "Investor and (Value Line) Analyst Underreaction to Information about Future Earnings: The Corrective Role of Non-Earnings Surprise Information," Journal ofAccounting Research, vol, 39, no, 2 (September):387-404,
©2006, CFA Institute
