The Timeliness of Earnings News and Litigation Risk

Transcription

1 The Timeliness of Earnings News and Litigation Risk Dain C. Donelson McCombs School of Business, University of Texas at Austin 1 University Station, B6500, Austin, TX dain.donelson@mccombs.utexas.edu Phone: (512) Fax: (512) John McInnis McCombs School of Business, University of Texas at Austin 1 University Station, B6400, Austin, TX john.mcinnis@mccombs.utexas.edu Phone: (512) Fax: (512) Richard D. Mergenthaler Henry B. Tippie College of Business, University of Iowa Iowa City, IA rick-mergenthaler@uiowa.edu Phone: (319) Fax: (319) Yong Yu McCombs School of Business, University of Texas at Austin 1 University Station, B6400, Austin, TX yong.yu@mccombs.utexas.edu Phone: (512) Fax: (512) December 2010 Keywords: Securities litigation;, disclosure; analyst forecast; earnings news. Data Availability: Data are available from public sources identified in the paper. JEL Classification: K22, K41, M41 We thank Dan Collins, Ross Jennings, Robert Prentice and workshop participants at the University of Iowa, the University of Pennsylvania (Wharton), and the University of Texas for their helpful comments.

2 The Timeliness of Earnings News and Litigation Risk ABSTRACT This study investigates whether the timely revelation of bad earnings news is associated with a lower incidence of litigation. The timeliness of earnings news is captured by a new measure based on the evolution of the consensus analyst earnings forecast. Holding total earnings news constant, we find that earlier revelation of bad earnings news lowers the likelihood of litigation. Further, we examine analysts research reports to provide evidence on the information source of the bad news revelation. We find that the vast majority of bad news is revealed via management disclosure. More importantly, while sued firms are more likely to issue earnings warnings via press releases, non-sued firms provide earlier disclosures and utilize alternative disclosure channels such as conference calls or management presentations. Unlike prior studies, we find robust evidence that earlier revelation of bad earnings news deters litigation. Keywords: Securities litigation;, disclosure; analyst forecast; earnings news. Data Availability: Data are available from public sources identified in the paper. JEL Classification: K22, K41, M41 1

3 I. INTRODUCTION This study examines the relation between the timeliness of earnings news and the incidence of securities litigation. Skinner (1994) proposes that the earlier revelation of bad news reduces the threat of litigation because earlier revelation diminishes the perception that management hid the truth and minimizes damages by shortening the class period (the litigation reduction hypothesis ). Despite the logic of this prediction, results from prior studies are mixed and do not provide strong support for the litigation reduction hypothesis. Prior studies test the litigation reduction hypothesis by using management earnings warnings or pre-announcements via press releases to measure early disclosure. Francis et al. (1994) and Skinner (1997) find that quarters triggering litigation are actually more likely to contain warnings. However, Skinner (1997) suggests this positive relation may be driven by endogeneity because managers with material bad news have strong incentives to disclose early, but also face higher litigation risk. Using a simultaneous equation approach to control for endogeneity, Field et al. (2005) find no significant relation between early disclosure and litigation incidence using all suits, and some evidence supporting the litigation reduction hypothesis for settled suits. Overall, the validity of the litigation reduction hypothesis remains uncertain. Given recent evidence in Rogers and van Buskirk (2009) that managers appear to emerge from litigation with a belief that disclosure triggers litigation, rather than deterring litigation, more research is needed in this area. Our innovation is that we adopt and test a broader view of the litigation reduction hypothesis by focusing on the timeliness of total earnings news revealed to the market. There are two motivations for our focus on total earnings news. First, press releases are only one way managers can reveal bad earnings news to capital markets. Alternative disclosure channels 2

4 include analyst conference calls, presentations, webcasts, and private communications (at least before Regulation FD). Second, in cases involving bad earnings news, plaintiffs must demonstrate that management made misleading statements (misstatements) or withheld bad news they had a duty to disclose (omissions). However, liability under U.S. securities laws does not depend on any particular form or source for the corrective revelation. Thus, the litigation reduction hypothesis should hold as long as total bad earnings news is revealed on a more timely basis regardless of the method of revelation. In addition, the presence and materiality of a misstatement or an omission is determined in the context of the total information available to the market, including that from non-management sources (Francis et al. 1994). We measure total earnings news using analysts' consensus earnings forecasts. The consensus forecast is the best available proxy for the market earnings expectation and reflects both information disclosed by managers through all channels (e.g., press release, conference call, analyst presentations) and information revealed by non-management sources (e.g., analyst research or press articles). Our sued sample consists of 423 securities class action lawsuits from For each sued firm-quarter (i.e., the quarter triggering the lawsuit), we select a nonsued firm from the same quarter with the closest total earnings news (i.e., the ultimate earnings revelation less the beginning-of-quarter consensus forecast, scaled by beginning stock price). After matching on total earnings news, we measure earnings news timeliness by computing the proportion of total earnings news revealed up to each day in the quarter and then taking the average of the daily proportions. For example, assume the ultimate earnings revelation in a quarter is $0.10 per share, the beginning forecast is $0.50 per share, and the forecast on a given day is $0.30. The proportion of the total news revealed up to that day is 50% (i.e., ( )/ ( )). We calculate this proportion for each day and then compute the average 3

5 across all days in the quarter. Thus, our timeliness measure captures the average daily proportion of total earnings news revealed to the market during the quarter triggering litigation. Our univariate and multivariate tests indicate that more timely revelation of bad earnings news is associated with a lower threat of being sued. This negative relation is robust to controlling for a variety of firm characteristics related to litigation risk. Furthermore, the litigation reduction effect is economically meaningful: moving up one quintile in our timeliness measure is associated with a 7.3 percentage point decrease in the probability of litigation. We conduct robustness tests to address alternative explanations. First, our results are not driven by sued firms having larger stock price drops over the litigation quarter or over a short window (e.g., a single day). Second, our results are not driven by innate differences between sued and non-sued firms. We find similar patterns when we benchmark sued firms against themselves by examining the speed of news revelation in non-litigation quarters. Third, our findings are not attributable to differences in the nature of the bad earnings news (i.e., idiosyncratic news vs. industry news) between sued and non-sued firms. In addition to the alternative explanations we test, the endogeneity concern from prior studies cannot explain our findings. The endogenous link between disclosure and litigation risk induces a spurious positive relation between timely bad news revelation and litigation incidence (Skinner 1997; Field et al. 2005). We find just the opposite relation with our design. To reconcile the difference between our findings and prior studies, we examine the underlying sources driving analysts' forecast revisions in our sample. We hand-collect analyst research reports for a random sub-sample of firms with large negative total earnings news. These reports contain analysts explanations for their forecast revisions. We also collect press releases from Factiva and Lexis-Nexis. We then classify each revision into three categories based on the 4

6 source of the information that triggered the revision: a) traditional managerial warnings through press releases (i.e., those studied by prior research), b) non-traditional managerial disclosures (e.g., conference calls, webcasts, presentations), and c) non-management news (e.g., analysts research with respect to industry news or polling of customers or suppliers). This analysis yields several important results. First, the overwhelming majority of analyst revisions are driven by managerial disclosures, traditional or non-traditional. Second, consistent with prior studies, sued firms are more likely to issue traditional earnings warnings. However, when non-sued firms do issue traditional earnings warnings, the warnings are more timely. More importantly, non-sued firms are more likely than sued firms to issue non-traditional managerial disclosure, and these disclosures are also more timely. Finally, controlling for the revelation channel has little effect on the negative association between timeliness and litigation risk, suggesting our main findings are driven by news timeliness rather than the form of disclosure. This study makes two primary contributions to the academic literature. First, we provide the first strong evidence supporting Skinner's (1994) litigation reduction hypothesis. Second, our results demonstrate the importance of considering multiple revelation channels when studying managerial disclosure. Our findings indicate that managers frequently utilize channels other than explicit press releases to disclose bad earnings information. Therefore, examining only press releases yields an incomplete picture of managerial disclosure and can lead to erroneous inferences. In addition to academic contributions, our study has important practical implications for managers who wish to understand how disclosure timeliness affects litigation incidence. Rogers and van Buskirk (2009) find that firms reduce their level of disclosure for both good news and bad news after being sued. Reducing the disclosure of bad news appears inconsistent with the 5

7 belief that preemptive bad news disclosures help prevent lawsuits (Lowry 2009; Rogers and van Buskirk 2009). Our findings indicate that early disclosure deters litigation and that managers should consider revising their disclosure strategies accordingly. Section II discusses the institutional background and related literature. Section III describes the research design and reports results. Section IV concludes. II. INSTITUTIONAL BACKGROUND AND RELATED LITERATURE Securities litigation involving bad earnings news alleges violations of Rule 10b-5. The elements of a Rule 10b-5 violation are (1) a misstatement or omission of (2) a material fact (3) made with intent (4) that the plaintiff justifiably relied on (5) causing injury in connection with the purchase or sale of securities (Skinner 1994). Suits involving bad earnings news, therefore, allege either an omission (i.e., the failure to disclose earnings news on a timely basis) and/or a misstatement (i.e., an incorrect earlier forecast or statement of fact), either of which may satisfy element (1). A significant negative stock price reaction to the news demonstrates both materiality (element (2)) and causation (element (5)). Element (4) is generally presumed under the fraud on the market theory (see Francis et al. 1994). In order to survive a motion to dismiss under the Private Securities Litigation Reform Act of 1995 (PSLRA), the plaintiffs complaint must allege facts that lead to a strong inference of fraudulent intent. Thus, earnings disclosure cases often turn on whether the alleged omission or misstatement was intentional. Under the high pleading standard of the PSLRA, a mere stock price drop accompanying the release of bad news is insufficient to withstand a motion to dismiss. Rather, the circumstances surrounding the alleged omission or misstatement are critical and the appearance that managers are trying to hide information is beneficial to plaintiffs. Likely because 6

8 they must allege facts that give a strong inference of intent, plaintiffs generally present as many facts as possible to demonstrate that it was implausible that managers were ignorant of the bad news prior to its revelation and commonly allege both omissions and misstatements. 1 The first study to formally propose that the timing of earnings news could affect the likelihood of litigation is Skinner (1994). Skinner (1994) and several later studies (e.g., Skinner 1997; Field et al. 2005; Healy and Palepu 2001) outline three reasons underpinning the litigation reduction hypothesis. First, timely disclosure weakens the claim that managers withheld bad earnings news. Second, timely disclosure shortens the nondisclosure period, reducing the size of the plaintiff class and potential damages in a lawsuit. Third, timely disclosure may mitigate large negative stock price reactions, which are often viewed as a trigger for litigation. Disclosing part of the bad news earlier spreads bad news out over time and helps avoid a large, one-time price drop. In addition, timely disclosure may enhance managers perceived competence or credibility, engendering a less severe negative stock price reaction. Several studies examine the litigation reduction hypothesis using managerial earnings warnings or preannouncements through press releases as a proxy for timeliness. These studies use the presence of a warning prior to the earnings announcement (Field et al. 2005), whether the warning is issued before or after the quarter-end (Francis et al. 1994; Skinner 1997), or the number of days between the press release date and a benchmark point (i.e., quarter-end, quarterbeginning or earnings announcements) (Francis et al. 1994; Skinner 1997) as their measures of 1 Brody et al. vs Zix Corporation is a representative case. On May 4, 2004, Zix reported quarterly earnings eight cents below the I/B/E/S consensus forecast. Zix s stock price declined by over 50% in the days following the earnings announcement. The class action complaint alleged that Zix s managers knew of the operational weaknesses responsible for the poor earnings performance, but a) continued to make false representations and projections (misstatements) and b) failed to disclose material operational and financial information (omissions). The case was settled in 2008 for $5.6 million, slightly more than 3% of Zix s market capitalization at that time. 7

9 timeliness. 2 Francis et al. (1994) find that sued firms issue voluntary earnings disclosures prior to the earnings announcement at a significantly higher rate than non-sued firms. This suggests that earnings warnings may trigger more lawsuits, opposite to the litigation reduction hypothesis. Skinner (1997) generally confirms the main findings of Francis et al. (1994) with a different design. He compares press release disclosure patterns of sued firms during quarters that triggered litigation to their disclosure practices in non-litigation quarters and finds that early disclosure of bad news via press releases is much more common in litigation quarters. 3 However, Skinner (1997) notes that endogeneity could drive the positive relation between disclosure timeliness and litigation risk because firms facing higher litigation risk may disclose bad news early to mitigate litigation costs. Skinner (1997) explicitly ties the endogeneity problem to the magnitude of the bad earnings news. Using a simultaneous equation approach to address the endogeneity problem noted by Skinner, Field et al. (2005) find no significant relation between earnings warnings and litigation risk when examining all suits, inconsistent with the litigation reduction hypothesis. After excluding dismissed suits, they find that earnings warnings are negatively related to litigation incidence, consistent with the litigation reduction hypothesis. To our knowledge, Field et al. (2005) is the only prior study that finds any evidence supporting the litigation reduction hypothesis with respect to the incidence of litigation. 4 Overall, results from prior research with respect to the litigation reduction hypothesis are 2 One exception is Francis et al. (1994), who also examine earnings disclosures through analyst reports, and find that managers sometimes disclose news to analysts who then disseminate the information in the pre-regulation FD environment. Francis et al. (1994) also examine analyst forecasts for a small sample of firms to test whether analysts anticipate negative earnings news to a larger extent for non-sued firms than for sued firms. They find no such evidence. However, they do not examine the timing of the forecast revisions during the litigation quarter. 3 A related line of research examines how the risk of shareholder litigation affects managers' disclosure choices (e.g., Skinner 1994; Kasznik and Lev 1995; Rogers and van Buskirk 2009). 4 In this study, we discuss the litigation reduction hypothesis in terms of the incidence of litigation. However, we recognize that after controlling for estimated shareholder damages, Skinner (1997) finds some evidence suggesting that earlier disclosure is associated with lower settlements. 8

10 quite mixed. While the recent work of Field et al. (2005) helps explain the seemingly anomalous positive relation between litigation incidence and explicit earnings warnings in prior studies, there is no consistent evidence supporting the litigation reduction hypothesis. In addition, as noted above, the results in Field et al. (2005) supporting the litigation reduction hypothesis are limited to settled suits, where a strong inference of fraud exists. Thus, there is no evidence to date that early disclosure of bad earnings news helps prevent non-meritorious litigation. 5 One potential reason for the general lack of support for the litigation reduction hypothesis in prior work is the use of only explicit warnings to measure the timeliness of earnings news. This focus is understandable given Skinner s (1994) interest in managerial disclosure of bad news. However, it is critical to note that the bad news revelation is not the source of legal liability under U.S. securities laws. Rather, it is the prior failure to disclose when a duty to disclose exists (an omission) or a prior misstatement (e.g., a misleading or knowingly false earnings projection) that actually exposes firms to litigation. 6 Thus, there is no requirement that the revelation that leads to litigation comes from any particular source (e.g., management) or takes any particular form (e.g., explicit warning). Rather, a wide array of disclosure sources and forms may reveal the alleged fraud and result in litigation. 7 The importance of total earnings news is also supported by the materiality requirement of Rule 10b-5, which takes into account the total information mix available to the market (see Basic, Inc. v. Levinson, 485 U.S. 224 (1988) and Francis et al. 1994). Whether any managerial 5 The Skinner (1997) sample predates the reforms of the PSLRA, but the failure to find results related to the dismissal decision also limits inferences regarding non-meritorious litigation. 6 There is no overarching duty to disclose all material information immediately (e.g., In re Parametric Technology Corp. Securities Litigation, 300 F. Supp. 206 (D. Mass. 2001); Rogers and van Buskirk 2009). However, if a firm decides to disclose information, it must disclose the whole truth such that the disclosure is not misleading (Rogers and van Buskirk 2009). In addition, if a firm makes an initial disclosure, many courts have extended the whole truth doctrine to provide a duty to update if the information is still relevant to market participants (e.g., Langevoort and Gulati 2004). Finally, a duty to disclose arises when corporate insiders trade based on material, nonpublic information (e.g., San Leandro Emergency Med. Plan v. Philip Morris, 75 F. 3d 801 (2d Cir. 1996)). 7 For example, Dyck et al. (2010) detail numerous cases where external whistleblowers revealed the fraud. 9

11 statement (or failure to disclose) is misleading is determined in the context of the total information available to the market, including information from non-management sources. Nontimely bad news revelation of any form (e.g., a warning, a conference call, a press article) simply leads to the appearance that management had something to hide, which may make plaintiffs allegations of intentional omissions or prior misstatements by management seem more credible. Thus, the timeliness of total bad earnings news available to the market is likely of the highest importance with respect to the litigation reduction hypothesis. While prior studies utilize press releases to measure managerial disclosure, managers can reveal negative earnings news through public disclosure channels such as conference calls (Tasker 1998; Frankel et al. 1999), presentations at conferences (Bushee et al. 2009), and company websites (Deller et al. 1999). Further, anecdotal evidence and academic research indicate that, at least prior to Regulation FD, managers often engaged in private communication with institutional investors and financial analysts through closed conference calls, one-on-one meetings, or phone conversations. 8 Analysts also conduct research through industry analysis and the polling of customers and suppliers, which can lead to the revelation of earnings news. Finally, whistle blowing by a former employee or a press article may reveal news to the market. Unfortunately, it is difficult to gather data regarding many of these revelation channels, and some are simply unobservable. 9 Fortunately, analysts' forecasts are a widely-used proxy for market expectations and provide a measure of all earnings news that reaches market participants (e.g., Bartov et al. 2002). Analysts forecasts incorporate information developed through independent analyst research as well as information released in all disclosures, including those 8 Managers continue to conduct one-on-one meetings with analysts and institutional investors post-regulation FD. For example, in a survey by Rivel Research (NIRI 2001), 79% of the 577 National Investor Relations Institute member companies claim that they hold one-on-one meetings to the same or greater extent post-regulation FD. 9 For example, conference call and conference presentation transcripts became available only recently, and managers' private communication with financial analysts is largely unobservable. 10

12 otherwise unavailable to researchers (e.g., private communication with managers). Thus, this proxy is a broad measure that captures the timeliness of all earnings news during the litigation quarter and complements prior research that examines official press releases. 10 In our empirical analysis below, we examine two related issues. First, we examine whether the timely revelation of total bad earnings news is linked to lower litigation risk. Second, we examine the source of the information that drives analyst forecast revisions (our measure to capture the timeliness of total earnings news). III. RESEARCH DESIGN AND EMPIRICAL RESULTS Sample We use the Securities Class Action Services (SCAS) database from RiskMetrics to identify all class action securities lawsuits filed in Federal court from Similar to Field et al. (2005), we exclude cases that involve restatements of prior financial statements (i.e., accounting irregularities) because it is less likely that timely disclosure of, say, fictitious revenue will meaningfully reduce litigation risk. Our sample thus contains bad news lawsuits that allege omissions, misstatements, or both. We require firms to have unfavorable (i.e., negative) earnings news in the fiscal quarter that triggered the lawsuit. We define earnings news as the earnings revelation at the end of the class period (the period over which fraud is alleged) minus the first consensus forecast of the quarter, scaled by stock price at the beginning of the quarter. We measure the first consensus 10 A potential, broader alternative than analyst forecasts would be to examine stock returns. However, utilizing returns has drawbacks accompanying the potential benefit. First, returns capture information beyond short-term earnings news. Second, stock returns are less controllable by firm management and the implications of our findings would be less clear with respect to disclosure policy. Thus, we examine earnings news to provide evidence with respect to managerial disclosure choices and to be consistent with the focus of prior research on earnings news. In untabulated tests, we find similar, but slightly weaker, results when we use stock returns to measure news timeliness. 11

13 forecast of the quarter at least one day after the announcement of prior quarter earnings. 11 For cases where the end of the class period coincides with the official earnings announcement date (about 17% of cases), we define the earnings revelation as actual earnings in I/B/E/S. For all other cases, the earnings revelation is defined as the consensus forecast at the end of the class period. 12 In these cases, the bad news revelation comes most commonly from a management warning or pre-announcement, and less commonly from a popular press article or analyst report. We refer to the time from the beginning consensus forecast to the end of the class period as the quarterly revelation window. For cases where the class period ends on the earnings announcement date, the median length of the quarterly revelation window is 91 days. For all other cases, the median length of this window is 61 days. Overall, our final sued sample contains 423 sued firm-quarters in which unfavorable earnings news triggered litigation. Since few firms in our sample (15 out of 408 unique firms) have more than one suit, we refer to firm-quarters as firms throughout the remainder of the study for brevity. For each sued firm, we select a non-sued firm from the same quarter with the closest total earnings news. For lawsuits where the class period ends on the earnings announcement day, total earnings news for matched firms is measured as actual earnings minus the beginning consensus forecast, scaled by beginning stock price. For all other cases, total earnings news for the matched firm is measured as the ending consensus forecast less the beginning consensus forecast, scaled by beginning stock price. The ending consensus forecast for non-sued firms is measured the same number of days after the prior quarter earnings announcement date as the ending consensus forecast for the corresponding sued firm. This procedure helps ensure the length of 11 We do this to help ensure forecasts have been updated for the prior quarter s earnings announcement. Results are similar if we measure the first consensus forecast at least two weeks after the prior quarter announcement date. 12 When the class period does not end at the earnings announcement date, we measure the consensus up to five days after the end of the class period to allow for a lag in the incorporation of bad news into analysts I/B/E/S forecasts. 12

14 quarterly revelation window is roughly equal for sued and non-sued firms. Our final non-sued sample contains 423 matching non-sued firms with unfavorable earnings news of a similar magnitude to sued firms. Summary of Main Findings Our interest lies in whether the timeliness of total earnings news differs significantly between sued and non-sued firms, holding the magnitude of total earnings news constant. Before presenting the full results of our empirical analysis, we offer a simple picture to summarize our main finding. For each firm in our sample, we measure the difference between the consensus forecast each day and the ultimate earnings revelation, scaled by stock price. This difference, or forecast error, measures the extent to which the consensus forecast at any given time reflects the total earnings news of the period. For each firm we then divide the quarterly revelation window into 20 equal intervals, and calculate the median forecast error across firms by interval. The results are plotted in Figure 1. The dashed line represents firms that are sued whereas the solid line represents firms that were not sued. By construction, total earnings news (as measured by the decline in the consensus) is nearly identical for sued and non-sued firms. However, Figure 1 reveals a stark difference in the news timing. Bad earnings news is revealed much earlier for firms that are not sued. For example, at the halfway point of the revelation window, over 55% of the bad news is revealed on average for non-sued firms, compared to less than 25% for sued firms (untabulated). Figure 1 crystallizes our main finding, and all of our remaining tests reinforce this basic result. 13

15 Research Design Our timeliness measure is defined as the average proportion of total news revealed up to a given day during the quarterly revelation window. To illustrate, assume that quarterly revelation window is only four days, the beginning consensus forecast is $0.50 per share, and the ultimate earnings revelation is $0.10 per share, implying total (unscaled) news of $0.40 per share during the quarterly revelation window. If the consensus forecast was revised to $0.30 on the second day of the quarterly window, then 50% (($ $0.30)/ ($ $0.10)) of the total news had been revealed at that point. If the consensus forecast is revised to $0.20 on day three of the quarterly window, then 75% (($0.50 $0.20)/ ($ $0.10)) of the total news was revealed up to that day. Finally, if the consensus forecast is revised to $0.10 on the final day of the revelation period, 100% of the news was revealed at that point. Averaging these figures for each day yields our timeliness measure, which is 56% ((0% + 50% + 75% + 100%)/4) in this example. In contrast, if the consensus forecast revision is less timely and there is no news disclosed until day three, then the value of our timeliness measure drops to 44% ((0% + 0% + 75% +100%). Hence, our timeliness measure captures the timeliness of the revelation of total news. We refer to this measure as TIMELINESS for each firm. Higher (lower) values of this variable imply more (less) timely revelation of bad earnings news. To examine whether TIMELINESS differs significantly across our sued and non-sued samples, we estimate the following logistic regression: Prob(SUED = 1) = F(α + β 1 TIMELINESS + β (CONTROLS)) (1) where F is the CDF of the logistic distribution, SUED is an indicator variable for sued firms, TIMELINESS is defined above, and CONTROLS is a vector of control variables. The coefficient on TIMELINESS, β 1, measures the extent to which timely revelation of bad news impacts the 14

16 probability of litigation. If earlier revelation of bad news reduces litigation risk, we expect β 1 to be negative. Control variables include total earnings news (NEWS), market value of equity (SIZE), the book-to-market ratio (BM), the raw stock return over the quarter (RET), an indicator variable for high industry litigation risk (HIGHLIT), share turnover (SHARE_TURN), share price volatility (VOLATILITY), and analyst following (FOLLOW). Inclusion of these variables in equation (1) helps control for differences between our sued and matched samples. We discuss each control variable below. Detailed variable descriptions are provided in Appendix A. We include total earnings news (NEWS) to neutralize any remaining difference in earnings news between sued and non-sued firms not accounted for by our matching procedure. We expect no significant difference in earnings news across samples, so we have no directional prediction for this variable. We expect stock returns (RET) to be negatively related to the probability of a lawsuit. More negative returns imply greater stockholder damages, which should increase litigation risk. We note, however, that one of the hypothesized mechanisms by which timely news revelation could reduce litigation risk is through attenuating the negative reaction of stock prices to bad earnings news (discussed earlier). To the extent investors react less severely to bad news released early in the quarter versus late in the quarter, including stock returns in equation (1) may control for some of the effect of interest. Nevertheless, we include stock returns because they are a strong predictor of litigation (e.g., Palmrose and Scholz 2004, Francis et al. 1994). Firms named in lawsuits also tend be growth firms (Shu 2000), implying a negative relation between BM and the probability of a lawsuit. Shu (2000) also shows that market value of equity (SIZE), share turnover (SHARE_TURN), and share price volatility (VOLATILITY) are positively related to litigation risk. Lawsuit incidence is higher in industries such as 15

17 technology and pharmaceuticals, so we include an indicator variable (HIGHLIT) for firms in industries identified in Francis et al. (1994). 13 Finally, since sued firms tend to be larger growth firms, we expect them to have higher analyst following (FOLLOW) than non-sued firms. We include this variable to control for any effect analyst following might have on the timeliness with which news is incorporated into the consensus forecast. Main Results Univariate Results Panel A of Table 1 contains descriptive statistics for our sued and non-sued samples. Consistent with Figure 1, mean and median values of TIMELINESS are significantly higher for non-sued firms relative to sued firms. For the median non-sued firm, a daily average of roughly 45% of total earnings news has been revealed throughout the quarter. For the median sued firm, the daily average of revealed news is about 12%. This difference is significant (p<0.01). <Insert Table 1 about here> As expected, there is no significant difference in total earnings news between the two samples. The mean level of bad earnings news for sued (non-sued) firms involves an earnings revelation that is 1.35% (1.13%) of market equity lower than the beginning consensus forecast. Sued firms have more negative quarterly mean stock returns relative to non-sued firms (-38.88% versus -9.69%). 14 This difference is large and, given the importance of stock price drops in triggering litigation, reinforces the need to control for differences in returns across sued and 13 Alternatively, we include indicator variables for two-digit SIC codes in lieu of HIGHLIT to allow more fully for differences in industry composition across sample and results are very similar. 14 Despite less negative returns, the poor stock performance of non-sued firms is economically significant. The mean quarterly return annualizes to roughly -45% per year, the lower quartile of the distribution of quarterly returns is -22%, and the average loss in market capitalization over the quarter is roughly $200 million (untabulated). 16

18 matched firms. We address this concern in two ways. First, we include stock returns in equation (1). Second, in additional results reported below, we re-estimate equation (1) using a subsample where differences in stock returns are neutralized across sued and non-sued firms. Inferences are unchanged, indicating differences in returns do not substantially affect our conclusions. The descriptive statistics in Panel A of Table 1 also reveal other key univariate differences between sued and non-sued firms in our sample. Sued firms are larger on average, have lower book-to-market ratios, are more likely to be in a high litigation risk industry, and have higher share turnover, stock price volatility, and analyst following. These results are consistent with prior findings (e.g., Francis et al. 1994; Shu 2001). Panel B of Table 1 provides evidence on the strength of the univariate relation between TIMELINESS and the probability of a lawsuit. We sort the sample observations into quintiles of the TIMELINESS variable and calculate the proportion of firms that were sued. This sorting procedure yields a dramatic (and nearly monotonic) spread in the incidence of litigation across quintiles. The incidence of lawsuits among the lowest quintile of TIMELINESS is 71.0%, while lawsuit incidence falls dramatically to 27.4% among the highest quintile of TIMELINESS. Overall, this simple analysis suggests that the timeliness of earnings news is linked in an economically significant way to the incidence of litigation. Table 2 presents correlations among the variables in equation (1). TIMELINESS is strongly negatively correlated with SUED (corr. = -0.30, p<0.01) and strongly positively correlated with RET (corr. = 0.22, p<0.01), implying earlier earnings news is associated with less negative stock returns. Other correlations are consistent with the patterns in Table 1. <Insert Table 2 about here> 17

19 Multivariate Results Table 3 contains parameter estimates from equation (1). To facilitate straightforward interpretation, we utilize the quintile rank of the TIMELINESS variable to estimate equation (1) instead of the raw TIMELINESS measure (results are similar if we use the raw measure). There is a negative relation between TIMELINESS and the probability of a lawsuit (p<0.01), even after controlling for stock returns and other variables from equation (1). To assess the economic significance of our findings, we include marginal effects for each parameter estimate, which are akin to slope coefficients in an OLS regression. Table 3 indicates that a one unit increase in ranked TIMELINESS (i.e., moving up from one quintile to the next in our TIMELINESS measure) is associated with a decrease in the probability of a lawsuit of 7.3 percentage points. 15 <Insert Table 3 about here> Parameter estimates for the control variables are generally significant in the predicted direction. Exceptions include the high litigation risk industry variable and stock price volatility, which are insignificant at conventional levels. The pseudo R-Square is 39.0%, indicating the covariates help explain the incidence of lawsuits reasonably well. Table 3 also provides the results from estimating equation (1) separately for dismissed and settled lawsuits (and their respective matching non-sued firms). This analysis addresses two issues. First, there may be something fundamentally different about the timeliness of information flow when firms actually commit fraud. It seems unlikely that many would object to firms being sued when they actually commit fraud. Second, on a related note, dismissed lawsuits are those 15 The marginal effects in Table 3 are calculated as the mean of the individual marginal effects evaluated at each sample observation (see Greene 2003, p. 668 for more discussion). An alternative way to calculate marginal effects for ranked or dichotomous variables is to calculate the predicted probability of a lawsuit when the variable takes on its highest value less the predicted probability of a lawsuit when the variable is at its lowest value, holding all other variables constant at their mean. When we employ this approach for the ranked TIMELINESS variable in (1), we find that moving from the lowest to the highest quintile of TIMELINESS is associated with a decrease in the probability of a lawsuit of roughly 45 percentage points (untabulatated). 18

20 that courts deem to be non-meritorious (i.e., the factual allegations do not convincingly lead to an inference of fraud). Exposure to non-meritorious litigation is arguably the very type of litigation that early revelation of bad news may best preempt. This subset of non-meritorious litigation is particularly important when examining the implications of the litigation reduction hypothesis for corporate disclosure policies. The results with respect to TIMELINESS are statistically significant and economically meaningful when we run our analysis separately for dismissed and settled lawsuits. Interestingly, the results are slightly stronger for dismissed lawsuits (both the coefficient and marginal effect). Thus, our main findings are not sensitive to the eventual case outcome. 16 Overall, the results from Figure 1 and Tables 1-3 consistently indicate that more timely revelation of bad earnings news is associated with a substantially lower incidence of litigation. This evidence provides strong support for the litigation reduction hypothesis. Alternative Explanations Fundamental Differences between Sued and Non-Sued Firms One potential concern with our main findings is that sued firms information environments may be fundamentally different from those of non-sued firms. Perhaps news tends to be less timely, or analysts react in a less timely manner to news, for sued firms versus nonsued firms due to differences in industry membership, business models, or other factors. Thus, 16 We perform two additional sub-sample tests. First, we assess the sensitivity of our findings to the magnitude of total earnings news as large drops in earnings news likely yield the strongest inference of fraud. Our results strengthen when we focus on cases with a larger magnitude of bad earnings news. Second, we examine the periods before and after the implementation of Regulation FD separately due to changes in the analyst environment at that time (see Gintschel and Markov 2004). The coefficient on TIMELINESS in the post-fd period drops by about onehalf from its value in the pre-fd period, though it is still economically and statistically significant. This evidence that the difference in news timeliness between sued and non-sued firm shrinks in the Post-FD period is consistent with non-sued firms engaging in more private communication with analysts in the Pre-FD period. However, because other major economic and regulatory events occurred in the 2000s (e.g., Sarbanes-Oxley), we are leery of drawing any definitive causal inferences from this finding. 19

21 our main analysis may simply be picking up innate differences in information flow or other firm characteristics between sued and non-sued firms. We address this concern in two ways. First, we benchmark sued firms against themselves and estimate equation (1) utilizing sued firms that also had negative earnings news in the quarter prior to the quarter that triggered litigation. We measure total earnings news in this quarter as the ending consensus forecast less the beginning consensus, scaled by beginning stock price. The results from this regression are in Table 4. Again, we find that negative earnings news is much less timely in the litigation quarter, even when sued firms are compared to themselves. Thus, the lack of timeliness of earnings news for sued firms is unique to the litigation quarter. <Insert Table 4 about here> Second, we assess whether difference in timeliness between sued and non-sued firms are unique to the litigation quarter by examining the first fiscal quarter that begins after the filing of the lawsuit. We require total news in this quarter to be negative for sued firms and match these firms to non-sued firms in the same quarter with the closest total news. Figure 2 plots the timeliness of earnings news using the same methodology used to construct Figure 1. The striking difference in timeliness between sued and non-sued firms evident in Figure 1 disappears. In fact, news revelation appears slightly more timely for sued firms, but a t-test reveals no significant difference in TIMELINESS across samples (t = -0.29, untabulated). 17 Combined with the above result (utilizing sued firms as their own control), this suggests that our main finding is not driven by industry or firm differences between sued and non-sued firms. 17 Inferences are similar if we examine the patterns in Figure 2 using a multivariate framework as in Table 3. In addition, we also compare sued and matched non-sued firms in the fiscal quarter prior to the lawsuit quarter and find patterns virtually identical to those presented in Figure 2 (no significant difference in news timeliness). 20

22 Differences in Total Stock Returns between Sued and Non-Sued Firms Another potential concern regarding our main results is the large average return difference between sued and non-sued firms (see Table 1). Although we match firms on the magnitude of total earnings news, we do not effectively match firms on investors reaction to this news. This difference in returns may be related to differences in the appearance of fraud and the amount of potential damages, both of which are related to the risk of litigation. Utilizing returns as a control variable in equation (1) may not completely alleviate this concern. To further address this issue, we form a sub-sample of our primary sample where firms are more closely matched on returns by eliminating sued firms and their corresponding match firm where the difference in returns is greater than 15 percentage points (results are very similar if we utilize a 10 percentage point cutoff). This yields a sub-sample of 230 firms (115 sued, 115 matches). <Insert Table 5 about here> Descriptive statistics and results of the logistic regression analysis for this sub-sample are presented in Table 5. Panel A shows that the substantial difference in earnings news timeliness between the sued and non-sued samples remains despite no significant difference in returns in this sub-sample. The mean (median) timeliness measure for sued firms is 0.22 (0.10), compared with 0.41 (0.38) for non-sued firms. Both differences are significant at the 1% level. The mean (median) returns for sued firms are % (-18.94%), compared with % (-14.90%) for non-sued firms. Neither difference is significant at conventional levels. Panel B of Table 5 shows the results of estimating the logistic regression from equation (1) for this sub-sample of firms. The primary finding with respect to timeliness still holds, with a coefficient of 0.585, similar to the coefficient from the primary sample and significant at the 1% level. The marginal effect of 10.9% is economically significant and somewhat higher than that of 21

23 the primary sample. Thus, differences in average stock returns across our sued and non-sued samples do not contribute significantly to our findings. Differences in One-Day Stock Returns between Sued and Non-Sued Firms Anecdotal evidence suggests that large, one-day stock returns may trigger suits. To the extent that plaintiffs lawyers utilize large, one-day stock returns to screen potential suits, this could affect the inferences of our study. To address this, we estimate equation (1) after controlling for the largest, one-day negative return during the quarter using both the full sample and the sub-sample of firms with similar returns from Table 5. In both cases, TIMELINESS remains highly significant. Nature of Earnings News Another potential explanation for our findings is that the nature of the earnings news itself might differ between sued and non-sued firms. For instance, the negative earnings news of sued firms may be idiosyncratic (and thus potentially indicative of fraud) whereas earnings news of non-sued firms may be driven by industry or macro factors. Furthermore, firms may release industry or macro news in a more timely fashion than idiosyncratic news. Hence, the significant relation between timeliness and the likelihood of being sued may be driven by inherent differences in the nature of the news. We deal with this concern in two ways. First, we utilize a within-industry matched pair design where sued firms are matched to non-sued firms in the same industry with the closest earnings news and rerun our main analysis using this within-industry matched-pair design. Second, we include an additional control variable that measures industry earnings news (using 22

24 both equal-weighted and value-weighted measures of earnings news) revealed during the litigation quarter and rerun our main analysis while including this variable. In both of these untabulated analyses, the statistical significance of our results does not change and economic significance is similar to that reported in Table Timeliness Measurement The final potential concern pertains to our TIMELINESS variable. The measurement period ends at the class period end date, which is likely a bad news event for sued firms. Hence, sued firms almost certainly reveal some portion of the total bad news at the end of the timeliness measurement period. However, non-sued firms do not necessarily have a bad news event at this point. To address this issue, we examine the subset of sued firms whose class periods end before the earnings announcement date. We then measure both the total earnings news and its timeliness through the earnings announcement date (i.e., over the entire quarter) for this subset of sued firms and their matched non-sued firms. In this specification, neither the sued firms nor the nonsued firms necessarily have an incremental bad news revelation at the end of the timeliness measurement period. The results (untabulated) of this analysis are very similar to those obtained in our main analyses. What Drives Revisions to the Consensus Analyst Forecast? The main advantage of using the evolution of the consensus analyst forecast is that we can precisely measure the timeliness of total earnings news. The downside is that we cannot directly observe the information source of revisions. We do not know, for example, whether 18 We also examine a measure of bad news smoothness, the extent to which the consensus forecast drops in a smooth, linear fashion over the quarter. When included in equation (1) along with TIMELINESS, this variable is insignificant. Thus, the timeliness of bad earnings news, not its smoothness, appears to influence litigation risk. 23