The Post Earnings Announcement Drift, Market Reactions to SEC Filings and the Information Environment

Transcription

1 The Post Earnings Announcement Drift, Market Reactions to SEC Filings and the Information Environment Joshua Livnat Professor of Accounting Stern School of Business Administration New York University 311 Tisch Hall 40 W. 4 th St. New York City, NY (212) jlivnat@stern.nyu.edu Daqing Qi Professor of Accounting Cheung Kong Graduate School of Business Room 332, Tower E3, Oriental Plaza 1 East Chang An Street Beijing, China (8610) dqi@ckgsb.edu.cn Woody Wu Professor in Accounting School of Accounting Chinese University of Hong Kong 205 K.K. Leung Building Shatin, Hong Kong (852) woody@baf.msmail.cuhk.edu.hk First Draft: January 2005 Current Draft: April 6, 2005 The authors gratefully acknowledge the preliminary and original Compustat quarterly data provided by Charter Oak Investment Systems, Inc. The authors are also grateful for the contribution of Thomson Financial for providing forecast data available through the Institutional Brokers Estimate System. These data have been provided as part of a broad academic program to encourage earnings expectations research. The authors also gratefully acknowledge the SEC filing dates data provided by Compustat.

2 The Post Earnings Announcement Drift, Market Reactions to SEC Filings and the Information Environment Abstract This study examines whether the well-documented tendency of returns to move in the direction of the earnings surprise following the preliminary announcement can be partially explained by investors who react to confirming information in the immediately subsequent SEC filings. We show that market reactions around SEC filings are positively and significantly associated with the preliminary earnings surprise, i.e. that information in SEC filings confirms, on average, the preliminary earnings surprise. We further show that the average hedge portfolio daily return on extreme earnings surprises is largest during the SEC filing window than during any other window between the two preliminary announcements, and larger than the return during a same length window just prior to the SEC filing. Finally, we show that analysts revise their earnings forecasts for the subsequent quarter based on information in SEC filings that confirm the preliminary earnings surprises, and that their forecast accuracy improves after the SEC filings. The above tests also examine the effects of the information environment on market reactions to confirming information in SEC filings. Overall, our evidence is consistent with a drift caused by an initial underreaction to earnings surprises, which is corrected later on when the market reacts to new confirming information.

3 The Post Earnings Announcement Drift, Market Reactions to SEC Filings and the Information Environment One of the most resilient market anomalies is the Post Earnings Announcement Drift (drift or Standardized Unexpected Earnings, SUE), which is the tendency of stock prices to move in the direction of extreme earnings surprise after earnings are announced in preliminary earnings press releases. The drift has been documented to persist through the subsequent three quarters, reverse in the fourth quarter, and to concentrate around subsequent earnings announcements (Bernard and Thomas, 1989, 1990). One of the explanations for the drift is investors initial underreaction, which is subsequently corrected when new information tends to confirm the initial earnings surprise. However, the drift literature has not studied market reactions around the Securities and Exchange Commission (SEC) filings filed immediately after the preliminary earnings announcements, which presumably may contain new information that can help investors assess the implications of the initial earnings surprises about future earnings. Whether SEC filings are associated with significant market reactions is far from clear. While prior level and long-window association studies indicate that corporate disclosures in SEC filings on such items as cash flows, pensions and postretirement benefits provide valuation relevant information (e.g., Livnat and Zarowin, 1990, Barth, 1991 and Amir, 1993), research on short-window market reactions to the SEC filings remains inconclusive (Hollie et al., 2005) 1. The purpose of this study is to investigate whether the drift is associated with significant market reactions around SEC filings and whether this relationship depends on the information environment. The information environment may 1 The next section surveys this literature in detail.

4 be important for our investigation because prior studies show that the drift is inversely related to the level of investors sophistication and firm size (Bartov et al., 2000; Brown and Han, 2000). Thus, we study whether the relationship between the drift and market reactions to the immediately subsequent SEC filings is attenuated for firms with better information environment. To address these research questions, we first investigate whether returns around the SEC filing dates are positively associated with the preliminary earnings surprise (SUE), which becomes available in the immediately preceding preliminary earnings announcement. A positive and statistically significant association provides evidence that the market reacts to additional information in these filings that confirms the previously disclosed SUE. We then construct a hedge portfolio based on SUE and examine whether the average daily arbitrage return in the short window centered on the SEC filing date is greater than the average daily return between the preliminary earnings announcement and the SEC filing date, and between the SEC filing date and the preliminary earnings announcement for the subsequent quarter. This sheds light on whether more confirmatory information is disclosed in the subsequent SEC filing than on any other day in the window between two consecutive quarterly earnings announcements. We also test whether the direction and magnitude of analyst revisions of the next quarter s earnings forecast following the SEC filings is positively correlated with the previous SUE, and the effects of such revisions on the bias and error of the forecasts. Finally, our tests examine the sensitivity of the primary analyses to the information environment of the firm and, specifically, to such factors as size, proportion of institutional investors, analyst coverage, book/market ratio, availability of balance sheet and cash flow information in preliminary 2

5 earnings, and the period after Regulation FD (REGFD) as compared to the pre-regfd period. This study provides evidence that the cumulative abnormal returns (CAR) in the three-day window centered on the SEC filing dates are significantly and positively associated with the previous SUE, after controlling for all other information that market participants obtain between the preliminary earnings announcement and the SEC filing. This significant positive association is stronger after REGFD and slightly lower for firms covered by financial analysts, but is almost unaffected by other measures of the information environment. Moreover, in a reverse regression of SUE on returns, the association between SUE and the SEC filing CAR is stronger than the association between SUE and CAR in windows either before or after the SEC filings bounded by preliminary earnings announcements. We also show that the average daily return on the SUE-based hedge portfolio is greater in the three-day SEC filing window than in any other day before or after that window, bounded by the previous or subsequent preliminary earnings announcements. Finally, revisions of analyst forecasts of the next quarter s earnings following SEC filings are significantly and positively associated with SUE, and the degree of this association is related to the firm s information environment. Further analysis indicates reductions in both forecast bias and error after SEC filings, with more significant reductions for smaller firms or firms with lower institutional holdings. This study contributes to three strands of literature. It provides evidence that a portion of the market underreaction to the initial SUE is corrected when additional information becomes available in SEC filings, consistent with an underreaction by investors who seek additional information to help interpret the initial extreme earnings 3

6 surprise. The study also provides evidence that the market indeed reacts to SEC filings, when the SEC filings contain information that can allow investors to interpret the previously issued earnings announcement. Finally, the study helps academics assess the effects of the information environment on investors underreaction and subsequent corrections due to information in SEC filings. The study has practical implications for investors too. As most firms file their SEC reports on the last days of the statutory period, the SEC filing dates are easy to forecast and trading strategies based on the prior SUE are feasible. Such trading strategies are more likely to be profitable for smaller firms, for firms with less institutional holdings and analyst following, and for firms that do not provide balance sheet or cash flow information in their preliminary earnings announcements. The rest of this paper is organized as follows. The next section provides background and motivation. Section 3 discusses the hypotheses and research design. Section 4 describes the sample selection and provides the empirical results. The last section concludes. II. Background and Motivation 2.1 The Post Earnings Announcement Drift The post-earnings-announcement drift (SUE) is one of the best-documented and most-resilient capital markets anomalies. Brennan (1991) calls it a most severe challenge to financial theorists (p. 70) and Fama (1998) refers to it as the granddaddy of all underreaction events (p. 286). The SUE effect was first discovered by Ball and Brown (1968) using a sample extending back to the 1940 s. Over the following decade, 4

7 several papers using different samples and methods confirmed the apparent drift (e.g., Jones and Litzenberger 1970; Latane, Joy, and Jones 1970; Joy, Litzenberger, and McEnally 1977). As these studies appeared, increased interest and skepticism led to the careful, large-scale studies of Rendelman, Jones, and Latane (1982), Foster, Olsen, and Shevlin (1984), and Bernard and Thomas (1989) all of whom continued to find the drift. Even very recent research (e.g., Francis et al. 2003; Mendenhall 2003; Narayanamoorthy 2003; Livnat 2003a,b) continues to document the existence of the SUE effect. Following its discovery, possible explanations for the drift quickly fell into two broad categories. The first potential explanation is that SUE researchers have systematically underestimated the risk of positive earnings-surprise firms, while overestimating the risk of negative earnings-surprise firms. If true, this would give rise to the observed positive (negative) abnormal returns for positive (negative) surprise stocks. The second explanation is that the drift represents a delayed reaction to the news in earnings announcements. While efficient-markets proponents clearly prefer the misestimation-of-risk argument, researchers have been unable to find meaningful support for the claim. Ball, Kothari, and Watts (1993) and Bernard and Thomas (1989) document a positive association between earnings surprises and changes in stock betas. But Bernard and Thomas show that the beta shifts are far too small to explain observed differences in abnormal returns across surprise categories. Bernard and Thomas also point out that if differences or changes in systematic risk were to explain the drift, then high-sue stocks should perform poorly in down markets. But they show that positive-surprise firms outperform negative-surprise firms in each of the 13 years of their sample, including 5

8 1974 when the S&P 500 fell by more than 26%. Further, Bernard and Thomas subject the drift to a diverse battery of robustness tests, such as controlling for common risk factors used in tests of the Arbitrage-Pricing Theory (APT), and conclude that much of their evidence cannot plausibly be reconciled with arguments built on risk mismeasurement but is consistent with a delayed price response (p. 34). Given that Bernard and Thomas s conclusion about the drift as a delayed response to the information in earnings announcements is the predominant belief among researchers, it seems important to examine the various significant occasions when market participants systematically correct their initial underreactions. For a drift to exist, market participants must eventually realize that their immediate reaction to the preliminary earnings surprise was insufficient. This is likely to occur, for example, if subsequent new information confirms the prior earnings surprise, such as a subsequent earnings release or additional information in SEC filings. Indeed, Shane and Brous (2001) show that the post earnings announcement drift is consistent with investors and analysts initially underreacting to the news in earnings and eventually correcting their underreactions, as new information arrives. However, they use Value Line forecasts and preliminary earnings announcements to test their assertions, as compared to the immediately subsequent SEC filings we use in this study. Indeed, SEC filings have not been available for research in machine-readable form, possibly explaining why drift researchers have not examined SEC filings for new information that explains how investors correct their initial underreaction. 6

9 2.2 Market Reactions to SEC Filings Most firms disclose their preliminary earnings for the quarter or year through a press release, following it with an SEC filing several weeks later. Easton and Zmijewski (1993) report a median lag between the balance sheet date and the preliminary earnings announcement of 28 days and a median SEC filing lag of 45 days. Our sample shows a similar pattern with a median preliminary earnings lag of 25 days, and a median SEC filing lag of 45 days. While some firms issue a press release to discuss earnings after their SEC filings (Stice, 1991), others do not issue any press release at all and rely on the information available in the SEC filings alone. Under the current financial reporting system in the United States, SEC reports represent the formal public release of firms detailed financial statements. Moreover, they contain disclosures mandated by the Financial Accounting Standards Board (FASB) and/or the SEC, and other voluntary disclosures. Such disclosures include, but are not limited to, management discussions of past and expected firm performance, segmental data, accounting policy choices and changes, business conditions, and the auditor s opinion in Form 10-K. Therefore, even though disclosures in the SEC and annual reports are not released to the public until several weeks after the announcements of preliminary earnings, they may still contain incremental information that is useful for market participants to revise their expectations about future earnings. Therefore, unless all the information in SEC filings has been preceded by information from alternative sources, the market should react to the release of these reports. However, most of the prior research on market responses to SEC filings provides little evidence of incremental information content in SEC filings beyond the preliminary 7

10 earnings announcements. Foster and Vickery (1978), as well as Wilson (1987), document that 10-Ks have information content beyond earnings announcements. In contrast, subsequent studies suggest that the market fails to react to earnings information contained in SEC filings (Foster et al. 1983; Foster et al. 1986; Cready and Mynatt 1991; Stice 1991; Easton and Zmijewski 1993; and Chung, et al. 2003). Easton and Zmijewski (1993) examine whether the SEC filing dates are associated with abnormal returns, using squared market model prediction errors to avoid any predictions about the direction of expected returns around the SEC filing dates. Their results show significantly different from zero abnormal market returns around preliminary earnings announcements but no significant differences for market reactions to SEC filings, except in those cases where only the 10-Q dates are known but no preliminary earnings announcement dates are available on the Quarterly Compustat File. These results seem to imply that SEC filings contain no incremental information beyond preliminary earnings announcements. Stice (1991) examines whether the information content of an earnings announcement can be affected by the method in which earnings are announced, concentrating on firms that file their SEC forms several days before the earnings announcement. Stice (1991) finds that SEC filings are not fully reflected in prices until subsequent earnings announcements are made. 2 Chung et al. (2003) corroborate Stice s (1991) findings and show that some of the firms in their sample behave as if they manage earnings. Qi et al. (2000) suggest that prior research s inability to detect little, if any, 2 Stice (1991) conducts this study at a time when SEC filings were not as readily available (e.g., on-line and other media) as they are today. Chung et al. (2003) examine the same issue when filings were available on EDGAR, but use only a handful of quarters from the beginning of the EDGAR database. Their findings seem to suggest that Stice s results hold true even with the availability of the SEC EDGAR database. 8

11 information content around the SEC filing date may be due to the SEC paper filing system in place at the time of prior studies. Their study compares SEC paper filings with SEC electronic filings to test whether the information content of 10-Ks has changed as a result of electronically available SEC filings. In contrast to most of the prior research, Qi et al. (2000) provide evidence that 10-K filings through the EDGAR system provide incremental information content that did not exist for the paper filings. However, they study the years , in which the EDGAR system was still voluntary. 3 In addition, their study is limited to firms with available AIMR analyst rankings. In a recent study covering the period , Griffin (2003) finds SEC filings to have significantly different from zero abnormal market returns, where the abnormal returns are the absolute value of excess returns around the filing date compared to the excess returns in a prior period. He finds greater market reactions to SEC filings of smaller firms, to firms with lower proportions of institutional holders, to firms that report on days with many filings by other firms, and to firms with delayed filings. Additionally, in multivariate results, Griffin (2003) finds evidence of stronger market reactions to filings made in recent periods and to delayed SEC filings. Balsam et al. (2002) investigate whether investors in firms that are suspected of engaging in earnings management are able to more rapidly incorporate the information about accruals available in 10-Q filings, and whether institutional investors seem to respond even before the SEC filing dates. They find evidence consistent with no investors reactions to the managed accruals around the preliminary earnings announcement (with event windows up to 9 days later), with market reactions to discretionary accruals by firms with at least 40% institutional investors during the 3 The EDGAR system became mandatory in May

12 window spanning 10 days after the preliminary earnings announcement through two days before the SEC filing date, and with market reactions to discretionary accruals in the window from a day prior to the SEC filing date through 15 days afterwards for firms with fewer institutional investors. Their interpretation is that institutional investors seem to find the information necessary to reverse accruals faster than other investors and prior to the SEC filing dates. Asthana et al. (2004) show that small trades increase in the five-day period around the 10-K filings after EDGAR as compared to the pre-edgar period, but not large trades, implying that small investors are better able to use the information in SEC filings in the post-edgar period. They also show that small investors seem to better incorporate the information content of the 10-Ks (as measured by returns around the filing) in the post-edgar period than before, and provide evidence consistent with an erosion of the information advantage that larger traders have as compared to small traders in the post-edgar period. Hollie et al. (2005) show that when preliminary earnings are materially different from the SEC filed earnings, the market reacts to the new earnings surprise in the SEC filings, but that financial analysts do not revise their forecasts according to this new information. The above studies indicate that the literature is inconclusive about whether SEC filings provide information to investors beyond that available in preliminary earnings releases. Earlier studies tend to document no information content in SEC filings, whereas more recent studies tend to show that SEC filings may have information content in certain settings where further information can be useful. Most prior studies explored 10

13 unsigned market reactions around the SEC filing dates, because the expected direction of the additional information in SEC filings is unknown. When the expected direction of the information is known, Stice (1991) finds no market reactions, Balsam et al. (2002) find market reactions only for firms with low institutional ownership and only for long windows after the filings, and Hollie et al. (2005) find market reactions when the SEC filings contain materially new earnings surprise. Thus, we still do not know whether SEC filings can be used by investors to help them assess the implications of an extreme initial earnings surprise about future earnings, causing at least a portion of the observed drift in prices after the preliminary earnings announcement. 2.3 The Firm s Information Environment Market reactions to earnings announcements and the level of the drift in subsequent returns have been previously shown to depend on the firm s information environment. For example, the Earnings Response Coefficient (ERC) has been shown to depend on the firm s size, growth prospects, whether the firm is actively followed by financial analysts, the extent of institutional investors and the composition of institutional investors (Kothari, 2001). Typically, larger firms or firms that have analyst following or a larger proportion of institutional (non-transient) investors are likely to have lower ERC because some of the information may have been privately obtained by select investors earlier. In contrast, smaller, neglected firms are likely to have a higher ERC because earnings is one of the primary sources of information about these firms for investors. Bartov et al. (2000) show that the level of the drift is also negatively related to the proportion of institutional investors, implying that there is less mispricing when the firm has a sufficient number of professional investors. Mikhail et al. (2003) find that the drift 11

14 is smaller for firms that are followed by experienced analysts, who tend to employ more sophisticated prediction models for earnings than just a seasonal random walk. Mendenhall (2003) shows that firms subject to lower arbitrage risks have smaller drifts, because arbitragers can exploit the arbitrage opportunities at lower arbitrage costs. Brown and Han (2000) find that for a selected sample of firms whose earnings generating process can be described by a simple AR1 model, there is a smaller drift for large firms than for small firms with a poorer information environment (measured by size, institutional holdings, and number of analysts following the firm). Thus, any attempt to specifically study the relationship between the drift and new information released through SEC filings should take into account the firm s information environment. In this study, we specifically examine the sensitivity of our analyses to firm s size, book/market ratio (a measure of growth prospects and information asymmetry), the proportion of institutional investors, analyst coverage, and whether the observation is preor post-regfd, when private communication between the firm and select investors is prohibited. We also examine the influence of balance sheet and cash flow disclosure in the preliminary earnings release. Such disclosure may affect the new information contained in SEC filings, and firms may choose to provide this additional disclosure in their preliminary earnings release because they have a poorer information environment (Chen et al., 2002 and Levi, 2004). 3.1 Timeline III. Hypotheses and Research Design To better explain our hypotheses and test procedures, the following graph shows the 12

15 sequence of events from preliminary earnings announcement for quarter t to the preliminary earnings announcement for quarter t+1: bf p ap f af Preliminary earnings SEC filing for Preliminary earnings announcement for quarter t announcement for quarter t quarter t+1 In the rest of the paper, we use subscript p for the period from one day before to one day after the preliminary earnings announcements for quarter t, subscript f for the period from one day before to one day after the SEC filing for quarter t, subscript ap (after preliminary) for the period from two days after the preliminary earnings announcement for quarter t through two days before the SEC filing for quarter t, and subscript af (after filing) for the period from two days after the SEC filings for quarter t through two days before the preliminary earnings announcement for quarter t+1. One of our tests also uses the 3-day window that begins four days before the SEC filing for quarter t through two days before the SEC filing. 3.2 SEC Filing Date Return and SUE We employ the following OLS regression to examine whether market reactions to SEC filings are related to the drift (with subscripts for firm i and quarter t omitted): CAR f = α + β 1 RSUE + γcar ap + ε (1) where CAR f is the three-day cumulative abnormal returns centered around the SEC filing date, RSUE is the adjusted rank of SUE (earnings surprise) for quarter t, and CAR ap is the cumulative abnormal returns from two days after the preliminary earnings announcement 13

16 for quarter t through two days before the SEC filing. The daily abnormal return is calculated as the raw daily return from CRSP minus the daily return on the portfolio of firms with the same size (market value of equity as of June) and book-to-market (B/M) ratio (as of December). The daily returns (and cut-off points) on the size and B/M portfolios are obtained from Professor Kenneth French s data library, based on classification of the population into six (two size and three B/M) portfolios. 4 The daily abnormal returns are then summed over the relevant period. For example, for the window of SEC filings, CAR f is cumulated over the three-day period centered on the SEC filing date. CARs for other windows are similarly estimated during the windows described in the previous sub-section. Consistent with prior studies, observations in the top and bottom 0.5% of CARs are deleted from the sample to ensure that our results are not driven by outlying returns. To estimate RSUE, we first derive a time-series forecast based on a rolling seasonal random walk model with a drift estimated over the previous 8 quarters to proxy for expected earnings, as do most prior drift studies 5. The SUE is defined as actual quarterly earnings (before extraordinary items and discontinued operations), X t, minus predicted earnings, scaled by the standard deviation of these prediction errors over the previous 8 quarters. To obtain RSUE, we sort the sample according to SUE each quarter, assign firms into deciles from 0 to 9, divide the firm s decile rank by nine to obtain a scaled rank between zero and one, and subtract We estimate SUE as: Xt Xt 4 δ SUE =, where X t is earnings for quarter t, δ is the average of X t - X t-4 over the VAR( Xt Xt 4) prior eight quarters, and the variance of X t - X t-4 is also estimated over the prior eight quarters. 6 This allows us to obtain, in equation (1), an intercept that is equal to the sample s average CAR, and a 14

17 If the SEC filings contain additional information that confirms investors assessments about the implications of the preliminary earnings surprise for future earnings, we expect a positive association between CAR f and the rank of the earnings surprise, RSUE; hence a positive and statistically different from zero slope coefficient β 1 in (1). Note that we do not have any predictions about the sign of CAR ap in Equation (1). It is included only to control for any news that affect market prices between the preliminary earnings announcement and the SEC filing date. To further examine whether the positive association between CAR f and RSUE is sensitive to the information environment, we estimate the following regression: CAR f = α + β 1 RSUE + β 2 DUMMY*RSUE + γcar ap + ε (2) where CAR f, RSUE and CAR ap are as defined above, and DUMMY is one of the dummy variables defined as follows: IH equals 0 (1) if the number of shares held by institutional investors as of the end of quarter t divided by the number of shares outstanding is below (above) the median firm in that quarter. SIZE equals 0 (1) if the firm s market value of equity at the end of quarter t-1 is below (above) the median size in that quarter. BM is equal to 0 (1) if the ratio of book to market value of equity as of the end of quarter t-1 is below (above) the median of that quarter. FAF equals 0 (1) if the firm has no (at least one) analyst forecasts of earnings in quarter t. REGFD equals one if the quarter end is after October 2000, and zero before. B/C is equal to 1 if the firm reports information about accounts receivable, accounts payable, and inventories in its preliminary earnings announcement or if the firm discloses net operating cash flows and capital expenditures in the preliminary earnings announcement, otherwise it is set to 0. slope coefficient that measures the return difference on the two extreme SUE deciles. 15

18 While β 1 is expected to remain positive, we expect β 2 to be negative when DUMMY is IH, SIZE and FAF, because institutional investors are likely to obtain some of the confirming information privately prior to the SEC filings. We do not make a directional prediction for β 2 when DUMMY is REGFD, as it is unclear whether, on average, the regulation prompts firms to disclose more information to all investors (in which case β 2 would be negative) or simply stop disclosing private information to select investors (in which case β 2 would be positive). We make no directional prediction about BM because it may proxy for both growth opportunities and information asymmetry. For comparison with intuition and prior studies, we exclude CAR ap from Equations (1) and (2) and re-estimate them with CAR p as the dependent variable, which is the three-day cumulative abnormal return centered on the preliminary earnings announcement date. The possibility exists that the positive correlation between CAR f and RSUE in equations (1) and (2) is trivial, i.e., there may exist a positive correlation between cumulative abnormal returns in any window after the preliminary earnings announcement and SUE due to the well-documented post earnings announcement drift. To address this concern, we estimate the following reverse regression: RSUE = α + β 1 CAR p +β 2 CAR ap +β 3 CAR f +β 4 CAR af + ε, (3) where RSUE, CAR p, CAR f and CAR ap are as previously defined, and CAR af is the cumulative abnormal return from the second day after the SEC filing through two days before the preliminary earnings announcement of quarter t+1. A positive correlation between CAR f and RSUE reflects market reactions to the preliminary earnings surprise that is confirmed by new information in the SEC filing. To the extent that most of the 16

19 new information about the firm is disseminated through the subsequent SEC filings, we expect β 3 to be greater than β 2 and β Tests Based on Daily Returns of a Hedge Portfolio To further provide evidence that the market reaction to SEC filings is related to the SUE, we construct an equally-weighted hedge portfolio each quarter by taking a long (short) position in firms falling into the highest (lowest) SUE quintile. We then normalize the portfolio return in each window by the number of days in the window and obtain the following average daily returns: DR f DR ap DR bf DR af = the average daily return in the three-day window centered on the SEC filing date, = the average daily return in the window from the second day after the preliminary earnings announcement of quarter t through two days before the SEC filing date, = the average daily return in the window [-4, -2] just prior to the SEC filing date, where day 0 is the SEC filing date, = the average daily return in the window from the second day after the SEC filing date through two days prior to the preliminary earnings announcement of quarter t+1. If the market reacts to the SEC filings because they contain new information that confirms the implications of the preliminary earnings surprise for future earnings, and if, on average, there is relatively less other information with implications for future earnings released until the preliminary earnings announcement of quarter t+1, then we expect DR f to be greater than either DR ap, DR bf, and DR af. We test for the daily return differences using standard t-tests. 17

20 3.4 Analyst Revisions Following SEC Filings If the SEC filings contain relevant information about future earnings, financial analysts should be able to detect and analyze such information and are expected to revise their earnings forecast for quarter t+1 in line with this information. We therefore expect that the analyst forecast revisions should be positively correlated with SUE after the SEC filings, if the information in the SEC filings confirms the preliminary earnings surprise. We use the following regression to examine this hypothesis and expect β 1 to be significantly positive: Where REV f = α + β 1 ISUE + ε (4) REV f = I/B/E/S revision of earnings per share for quarter t+1 after the SEC filing, defined as the average I/B/E/S forecast in the 20 days after the SEC filing minus the average I/B/E/S forecast in the period between the preliminary announcement and the SEC filing, scaled by share price at the end of quarter t, and ISUE = I/B/E/S earnings surprise, defined as the I/B/E/S actual earnings per share for quarter t minus the average I/B/E/S forecast for earnings per share for quarter t in the 90-day period ending one day before the preliminary announcement of earnings for quarter t, scaled by the standard deviation of analyst forecasts in that period. To further examine whether the positive association between REV f and the ISUE is affected by the firm s information environment, we estimate the following regression: REV f = α + δdummy + β 1 ΙSUE + β 2 ISUE*DUMMY + ε (5) where REV f and ISUE are as above and DUMMY is one of the variables used to assess the information environment that are defined in equation (2). FAF is excluded from DUMMY as all firms used to estimate (5) are followed by analysts. As in equation (2), we expect β 2 to be negative when DUMMY is B/C, IH, or SIZE, and make no directional 18

21 prediction when DUMMY is REGFD or BM. For comparison, we also examine the association between analyst forecast revisions for earnings per share of quarter t+1 after the preliminary announcement and ISUE, by replacing REV f in equations (4) and (5) with REV p, where REV p = I/B/E/S revision of earnings per share for quarter t+1 after the announcement of preliminary earnings for quarter t, defined as the average I/B/E/S forecast in the period between the preliminary announcement and the SEC filing minus the average I/B/E/S forecast in the 90-day period ending one day prior to the preliminary announcement of quarter t, scaled by share price at quarter end. In addition to the above tests that focus on the relation between analyst revisions after SEC filing and SUE, we also examine the effects of the SEC filings on the bias and absolute value of the error in analyst forecasts for quarter t+1earnings. We expect both to decrease after the SEC filing if they contain relevant information about future earnings. IV. Sample and Results 4.1 Sample Selection and Description Our initial sample contains 204,824 firm-quarters extracted from Compustat from the fourth quarter of 1990 (our SEC filing dates are for calendar years ) through the first quarter of 2003 (I/B/E/S data limits us to the first quarter of 2003), on which we impose the following selection criteria: 1. The earnings announcement date is reported in Compustat for both quarter t and quarter t+1, and daily returns are available in CRSP from one day before quarter t s earnings announcement through two days before the announcement of earnings for quarter t+1. 19

22 2. Earnings for quarter t and previous twelve consecutive quarters are available on Compustat. 3. The firm has a positive net book value. 4. The price per share is available from Compustat as of the end of quarter t and is greater than $1. This is likely to reduce noise caused by using penny stocks and also a small number to scale the earnings surprise. 5. The intervals between the preliminary earnings announcement for quarter t and the subsequent SEC filing, and between SEC filing and preliminary earnings announcement for quarter t+1, must be at least seven days. 6. Data for control variables are available. 7. CAR p, CAR f, CAR bf, CAR ap and CAR af are not among the top or bottom 0.5% for each quarter. In addition to the above criteria, we require data about variables used to measure the information environment. Institutional holdings are from Thompson Financial. Analyst forecast data are from I/B/E/S. Data about balance sheet or cash flow information in preliminary earnings releases is from the Charter Oak database, as explained in the Appendix. SEC filing dates are provided to the authors by Compustat. Panel A of table 1 summarizes the sample selection process and its effect on the size of our final sample that contains 93,884 firm-quarters. As panel B of table 1 shows, while the fourth quarter of 2001 has the largest number of firms (2,891), the number of firms generally increases each quarter over the sample period, from 1,203 in the fourth quarter of 1990 to 2,178 in the first quarter of Slightly less than a quarter of the observations are after the fourth quarter of 2000, when the SEC regulation on fair 20

23 disclosure became effective. (Insert Table 1 about here) Table 2 provides descriptive statistics. The mean of SUE is , although its median is positive at The distribution of SUE is skewed to the left and characterized by extreme values, hence justifying the need to use its rank in regression analysis. It is also evident from the table that the sample has a wide distribution in terms of size (market value of equity at the end of quarter t-1), which is skewed to the right. The mean and median of institutional holding are 40.3% and 38.7%, respectively, similar to those in Bartov et al. (2000). The mean and median abnormal returns for all windows are negative, except for those centered on the preliminary earnings announcement. The mean CAR af, from two days after the SEC filing through two days prior to the next quarter s preliminary earnings announcement, is rather large 0.88%, for which we have no explanation. The mean (median) ratio of book to market value of equity is (0.533), which is similar to that reported in prior studies. Finally, the last two lines of table 2 provide descriptive statistics about the earnings surprise calculated using I/B/E/S data (ISUE), and analyst revision of earnings for quarter t+1 following the SEC filings (REV f ). Because additional data for I/B/E/S forecasts are needed, the sample size is smaller in these two rows. The mean (median) of ISUE is -0.08% (0.02%). The mean and median of REV f are -0.06% and -0.01% respectively, indicating that, for our sub-sample with analyst forecast data, analysts, on average, revise earnings for quarter t+1 downward following the SEC filings for quarter t. (Insert Table 2 about here) Panel B of table 2 reports the Pearson correlations among the variables of interest. 21

24 The relationship between RSUE (the rank of SUE) and CAR around the earnings announcement date, as well as CAR during windows between the preliminary earnings announcement date of quarter t through two days before the preliminary earnings announcement of the subsequent quarter, are all positive and statistically different from zero. The positive associations are consistent with the prior drift literature, which documents positive associations between the preliminary earnings surprise and subsequent returns. Correlations among other variables are generally similar to those reported in other studies, and are sufficiently low to cause no multicollinearity concerns. 4.2 Filing Date Return and SUE To form a base of comparison, we first conduct a simple regression of CAR p on RSUE and report the results in panel A of Table 3. The coefficient on RSUE is and significantly positive (t = 44.62). Panel B of Table 3 reports the OLS parameter estimates and their t values based on equation (1), in which CAR f, the three-day cumulative abnormal return centered on the SEC filing date, is the dependent variable. Information released to the market between the preliminary earnings announcement and the SEC filing is controlled for through CAR ap. The coefficient on RSUE is , significantly positive at a level below This indicates that the return on the SEC filing date is positively associated with SUE, supporting our hypothesis that the market responds to SEC filings because they contain new information that confirms the implications of the preliminary earnings surprise for future earnings. The coefficient for CAR ap is and significant at below a 0.01 level, indicating that market participants obtain some significant information between the two dates, which tends to negate the effects of the preliminary earnings surprise on future prices. 22

25 Panels A and B of Table 3 also report the results of regression analyses based on Equation (2), where control variables are added one at a time to assess the importance of the information environment on the information content of SEC filings. We find that both the magnitude and the significance level of the coefficient for RSUE remain essentially the same, indicating that the association between the return on the SEC filing date and SUE is insensitive to the inclusion of these control variables. However, we can see that, consistent with prior studies, when the firm has a significant proportion of institutional investors, is larger and is followed by analysts, its ERC is significantly smaller (Panel A), but the information relevance of the SEC filings are the same as for other firms. We can also see that having a high B/M ratio is positively associated with ERC but insignificantly associated with returns on the SEC filing date, and that the ERC is lower post-regfd than before, but that the SEC filings contain more information post-regfd, possibly because there is less private communication with select investors. Finally, whereas the disclosure of balance sheet or cash flow information in the preliminary earnings announcement tends to reduce the ERC, possibly because investors react more to non-earnings information in the announcement (see also Levi, 2004), there is no noticeable difference on the SEC filing date. Thus, Table 3 confirms that market participants generally obtain supporting information on the SEC filing date to the preliminary earnings surprise, and their reactions to this information are significantly different from zero regardless of various measures of the information environment, except the post-regfd period, which shows significantly stronger market reactions than the pre-regfd period. (Insert Table 3 about here) 23

26 Panel A of Table 4 reports the OLS parameter estimates and their t values for analysis based on Equation (3), which is essentially a reverse regression of RSUE on abnormal returns cumulated over non-overlapping windows bounded by the preliminary announcements of earnings for quarter t and quarter t+1. The adjusted R 2 is and the coefficients for CAR p, CAR ap, CAR f, and CAR af are all significantly positive at the 0.01 level or below. Not surprisingly, the coefficient for CAR p is the largest among the four, indicating that the bulk of the preliminary earnings surprise is impounded in returns on the preliminary earnings announcement date. More importantly, consistent with our hypothesis, the coefficient for CAR f is , which is larger than that of for CAR ap and for CAR af. An F-test on whether the three latter coefficients impound the same proportion of the preliminary earnings surprise indicates that these coefficients are significantly different from each other at a level of significance below This is evidence that the positive correlation between returns on the SEC filing date and RSUE, reported in Table 3, reflect market reactions to new information in SEC filings that confirms the original earnings surprise on the preliminary announcement, and that, on the average, new information released during any other time between preliminary earnings announcements is less relevant for investors than in the SEC filings. (Insert Table 4 about here) Additional analysis reported in Panel B of Table 4 provides evidence that these results remain essentially the same for samples formed on the basis of proxies for information environment, except for large firms, firms with high proportion of institutional investors, firms not followed by analysts, during the post-regfd period, 7 Untabulated tests show that the coefficient on CAR f, the abnormal return centered on the SEC filing date, is significantly larger than either CAR af or CAR ap the returns before and after filing. 24

27 and for firms that released balance sheet or cash flow information in their preliminary earnings release. Note that, however, for all sub-samples, the coefficient on the return around the SEC filing date (CAR f ) is larger than either the coefficient on return before (CAR ap ) or after (CAR af ) the SEC filing, although the difference may not be statistically different in some sub-samples. Thus, the reverse regression results indicate that, in the period between the preliminary earnings announcements of quarter t and t+1, a larger proportion of the lingering effect of the SUE of quarter t is realized in the SEC filing window than in any other window. 4.3 Tests Based on Daily Returns Table 5 reports results of tests based on average daily returns on a SUE hedge portfolio during the SEC filing date, before it and after it through the day before the subsequent preliminary earnings announcement. The hedge portfolio is constructed each quarter by taking a long (short) position in firms falling into the highest (lowest) SUE quintile. The mean daily return in the window centered on the SEC filing date, DR f, is 0.053%, significantly larger than the mean daily return in the period prior to the SEC filing, DR ap, of 0.012% and the period that follows the SEC filing, DR af, of 0.025%. The table also shows that the average daily return in the SEC filing window is significantly larger than the average daily return in the same length window [-2,-4] that precedes the SEC filing window, DR bf. We test for statistical differences in average daily returns of SEC filing dates and in other windows using not only pooled data from all firms and quarters, but also using a Fama and MacBeth (1973) approach based on the 50 quarters of our sample period with very similar results. (Insert Table 5 about here) 25

28 Panel C of Table 5 reports the average daily returns in various sub-samples constructed according to variables related to the information environment. For all sub-samples, we find that the average daily return on the hedge portfolio during the SEC filing window is significantly larger than that during the same length period just prior to the SEC filing. In sum, for a hedge portfolio formed on extreme SUE during the period from two days after the preliminary announcement for quarter t through two days before the preliminary announcement for quarter t+1, the average daily return is the largest in the SEC filing date window, and larger than the average daily return in the same length window just before it. 4.4 Analyst Forecast Revisions Following SEC Filings If SEC filings contain new information that confirms the preliminary earnings surprise and analysts can access and analyze such information, they should revise their earnings forecasts for quarter t+1 according to this new information. Panel A of Table 6 reports results of regressions of analyst revisions in quarter t+1 forecasts right after the preliminary announcement but before the SEC filing on the ranked I/B/E/S earnings surprise (ISUE). Not surprisingly, the coefficient on ISUE is and significantly positive (t=54.08), indicating that analysts revise their subsequent quarter s earnings forecast according to the preliminary earnings surprise. The panel also includes the sensitivity of these revisions to the information environment of the firm with results that are consistent with prior studies and intuition. Panel B of Table 6 presents results based on equation (4), in which the analyst revisions following the SEC filings are regressed on ISUE. The coefficient for ISUE is and significantly positive at a level below This indicates that analyst revisions following SEC filings are positively and significantly 26