VOLUNTARY DISCLOSURE AND THE OUTCOME OF SECURITIES LITIGATION * Joshua Cutler University of Oregon Lundquist College of Business

Transcription

1 VOLUNTARY DISCLOSURE AND THE OUTCOME OF SECURITIES LITIGATION * Joshua Cutler University of Oregon Lundquist College of Business jcutler@uoregon.edu Angela K. Davis University of Oregon Lundquist College of Business davisan@uoregon.edu Kyle Peterson University of Oregon Lundquist College of Business kylepete@uoregon.edu August 2013 Abstract We examine the impact of voluntary disclosure by defendant firms on the outcome of securities fraud class actions. We hypothesize that firms with more disclosure will experience more adverse litigation outcomes, given the strict pleading standards of the Private Securities Litigation Reform Act. We expect this effect to be driven by non-forward-looking disclosures since forward-looking statements enjoy safe harbor protection. Using a measure of disclosure collected directly from lawsuit complaints and a broader measure of disclosure from press releases, we find that more disclosure by the defendant firm does indeed lead to a higher likelihood of significant settlements. However, more forward-looking statements are less likely to result in a significant settlement. When we examine the nature of the forward-looking statements, we find that lawsuits against firms issuing more negative forecasts are generally more likely to be dismissed. Keywords: Voluntary disclosure, litigation, securities fraud * We thank Garet Loutzenhiser-King, Tiffany Phillips and Zhixuan Li for valuable research assistance and the University of Oregon Lundquist College of Business for financial support. We also thank workshop participants at the University of Oregon and 2013 AAA Annual Meeting. 1

2 1. Introduction The accounting literature recognizes the role of private litigation in firms disclosure choices, but important aspects of firm behavior remain unexplained. In particular, accounting studies have focused on how disclosure influences litigation risk, or the probability that shareholders will file a lawsuit against the firm. (See e.g. Francis, et al., 1994; Skinner, 1997.) This strand of literature generally agrees that voluntary disclosure can reduce the risk and cost of litigation. (Field, et al., 2005). Given this academic consensus, firms display a puzzling pattern of significantly decreasing their disclosure after experiencing litigation. (Rogers and Van Buskirk, 2009). We theorize that the impact of disclosure on the outcome, rather than incidence, of litigation may partly answer this puzzle. The Private Securities Litigation Reform Act of 1995 (PSLRA) created a stringent standard for proving securities fraud, but PSLRA also required plaintiffs to pass this threshold without access to the private firm documents and information usually obtained through the discovery process. In this environment, public disclosure may be the only means for plaintiffs to make their case and succeed in litigation. We theorize that when defendants issue more disclosure, the probability increases that plaintiffs will succeed and obtain a significant settlement. Managers perceive this relationship and therefore reduce their subsequent disclosure. However, the PSLRA also created specific protections for forward-looking statements, so this particular type of voluntary disclosure should be less useful to plaintiffs. Further, Skinner (1997) provides evidence that disclosing adverse earnings news early can even help defendants and lead to lower settlement values. Therefore, we theorize that issuing more forward-looking statements will not harm defendants, and may actually help them. 2

3 To test this theory, we conduct an in-depth, large-sample study of the effect of firms disclosure on the outcome of securities fraud class actions. We construct two alternative measures of disclosure. The first measure counts all disclosures cited in the relevant portion of each lawsuit complaint and the second counts all press releases issued by the firm during the class period, defined as the period of time during which the plaintiffs allege that a firm s stock traded at inflated prices due to management s false and misleading statements. Combining these disclosure measures with data on lawsuit outcomes, our tests show that more overall disclosure leads to more adverse outcomes for the sued firm. That is, when plaintiffs have more public statements to include in a complaint, the likelihood increases that the lawsuit will be settled for a significant sum, rather than dismissed or settled for nuisance value. We also construct measures of forward-looking disclosure using the lawsuit complaint and press releases issued during the class period. In addition to these word-based measures, we consider management forecasts issued during the class period as an alternative measure of forward-looking disclosure. In contrast to the results using measures of overall disclosure, we find that more forward-looking disclosure decreases the likelihood that a case is settled for a significant sum, consistent with the protections for forward-looking statements under the PSLRA working effectively. These results also suggest that the positive association we observe between our overall disclosure measures and the likelihood a lawsuit is settled are driven by non-forwardlooking disclosures. Next, we distinguish between optimistic and pessimistic forecasts issued by sued firms. Past studies have focused on management guidance as particularly relevant to increasing or decreasing the risk and cost of litigation, and, as noted, such disclosures are protected under the PSLRA. We find that more bad news guidance can increase the chance of dismissal, consistent 3

4 with the idea that firms who voluntarily release bad news are less likely to be withholding material information or misleading shareholders. Good news guidance does not significantly hurt nor help the sued firm. Given that other non-forward looking statements appear to hurt the firm, this result suggests that PSLRA safe harbors for forward-looking statements have been effective. This study has important implications for theory and practice. It is the first large scale study of disclosure s effect on litigation outcome, a critical issue for public firms. Skinner (1997) performed important preliminary work on this topic, examining whether early disclosure of adverse news lowered settlement values for a small sample of firms in the pre-pslra period. Johnson, et al. (2007) similarly examined whether the presence of forecasts affected the outcome of lawsuits for a small sample of firms, but found no significant results. To our knowledge, this is the first study to directly and extensively examine this issue for a large sample of firms. We highlight previously unexplored effects of the PSLRA on disclosure decisions, and we show how not only the presence of disclosure, but its overall frequency can impact the outcome of litigation. We also separately analyze forward-looking disclosure and show that it affects litigation differently than overall disclosure. Past studies have considered the legal impact of forward-looking guidance, but not the overall level of disclosure, including non-forward-looking disclosure. Our findings suggest that non-forward-looking disclosure can adversely impact firms in litigation, which has important consequences. First, it provides a more complete picture of firms disclosure incentives. Not only must managers balance conflicting litigation and non-litigation incentives, they must also consider whether disclosure that could decrease (increase) the risk of litigation could simultaneously increase (decrease) the risk of an adverse outcome should litigation occur. 4

5 Second and more important, the evidence shows that the U.S. system of securities regulation through private litigation may create an unintended chilling effect on disclosure. Litigation succeeds at punishing firms that make misleading statements, but is less effective at punishing firms that withhold material information. As a result, firms may rationally choose to withhold such information, contrary to the desirable goal of providing more information to the market. The remainder of this paper will proceed as follows. Section 2 reviews prior literature and develops our hypotheses. Section 3 describes our sample and research design, and Section 4 presents our results. Section 5 concludes. 2. Prior Literature and Hypothesis Development 2.1 Association between Overall Disclosure and Lawsuit Outcome The general conclusion from prior literature is that voluntary disclosure can reduce the probability of facing securities litigation. Skinner (1994) finds that managers are more likely to voluntarily preempt an earnings announcement when there is bad news. Since lawsuits often follow large negative earnings surprises (Francis et al. 1994), Skinner (1994) suggests that managers voluntarily release negative news early in order to reduce the chance of a lawsuit. However, Francis, et al. (1994) reach a conflicting conclusion, finding that more preemptive forecasts do not reduce litigation risk and may in fact relate to higher litigation risk. Skinner (1997) highlights the likely endogeneity problem inherent in the Francis, et al. (1994) study: firms that already anticipate litigation may use voluntary disclosure to mitigate the negative stock price response to bad news, thus limiting potential lawsuit damages. Indeed, Skinner (1997) presents evidence that early preemption of bad news is associated with lower settlement values, indicating that disclosure can at least reduce litigation cost, if not risk as well. Field, et 5

6 al. (2005) explicitly control for endogeneity using a simultaneous equations framework and find that 1) firms with higher litigation risk are more likely to disclose early to preempt lawsuits and 2) early disclosures do reduce litigation risk. Given this negative association between disclosure and litigation risk, the findings of Rogers and Van Buskirk (2009) create a puzzle. They find that sued firms subsequently reduce both the quantity and quality of disclosure. If increasing disclosure could reduce the risk of a future lawsuit, why would firms that have just experienced the high costs of litigation choose to reduce disclosure? Assuming that the link between disclosure and litigation risk is valid, the Rogers and Van Buskirk (2009) result may be attributable to something besides litigation risk. We propose that the explanation resides in the relationship between disclosure and the outcome of litigation, once the decision to sue has already been made. In making disclosure decisions, firms must consider not only the risk of litigation, but also how their disclosure will impact the course of the litigation. Simply facing a lawsuit is costly and undesirable, but actually paying a large settlement may be much worse. Knowing that they may face lawsuits in the future despite their best efforts, firms can rationally make disclosure decisions that enhance the likelihood of a dismissal or insignificant settlement. By achieving a favorable outcome, firms will still incur legal fees and face a negative stock price response when the suit is filed, but they avoid large settlements and the perception that they committed fraud. Isolating the litigation incentives of disclosures, firms should select a disclosure policy that jointly minimizes both the risk of a lawsuit and the likelihood of an adverse lawsuit outcome. Skinner (1997) showed that disclosure can help firms in litigation, suggesting that more disclosure is better for both litigation risk and litigation outcome. However, Skinner (1997) 6

7 examines only one type of disclosure, and there are reasons to believe that other types of disclosure can have the opposite effect. In fact, overall disclosure level may hurt more than help. The key to understanding the impact of disclosure lies in the litigation environment created by the PSLRA. Congress passed the PSLRA in 1995 to reduce the perceived problem of frivolous securities lawsuits. A key component of PSLRA was a significantly heightened pleading standard for this type of suit, intended to screen out meritless cases at an early stage. After PSLRA, plaintiffs in securities suits must state with particularity facts giving rise to a strong inference that the defendant acted with the required state of mind. (15 U.S.C. 78u- 4(b)(2)(A)). The required state of mind in this case is known as scienter. Plaintiffs must show that the defendants knew the statements they made were false, or that they would have to be reckless not to know. After a securities fraud suit is filed, defendants will generally file a motion to dismiss under Federal Rule of Civil Procedure 12(b)(6), arguing that even if everything in the complaint were true, the plaintiffs have not sufficiently alleged that the defendants committed securities fraud. If the plaintiffs have not met the heightened PSLRA pleading standards, the suit will be dismissed. Congress apparently based this strict threshold on a judge-made standard already applied in the Second Judicial Circuit (Sale 2002 at 908). However, the Second Circuit required this particularized showing of the defendants mental state only after the plaintiffs had gathered private information from the company through the formal discovery process. (Id.) But the PSLRA now prohibits any discovery until after the court decides on the defendant s motion to dismiss. That is, plaintiffs may not request any documents from the firm or conduct any depositions with firm insiders. They must cite particular facts to support each of their 7

8 allegations, including facts about the defendants mental state, but they cannot obtain private information from the defendant to establish such facts. In this environment, plaintiffs must rely on public information to state their case. Any public disclosure that management issues effectively becomes ammunition for the plaintiffs. Typically, disclosure issued during the class period defined as the period of time when the stock traded at a price inflated by management s false or misleading statements is most relevant to proving management s state of mind. The Supreme Court has held that in applying the PSLRA standard, judges are not to scrutinize each allegation in isolation but to assess all the allegations holistically and collectively (Tellabs, Inc. v. Makor Issues & Rights Ltd., 551 U.S. 308, 326 (2007)). Thus, the more public information that management issues, the more evidence the plaintiffs will have to make their case. The role of public disclosure in the post-pslra litigation environment leads to our first hypothesis: H1: The quantity of disclosure issued during the class period will be positively related to the likelihood of a costly settlement. We construct two alternative measures of disclosure during the class period. The first measure counts all disclosures cited in the relevant portion of each lawsuit complaint and the second counts all press releases issued by the firm during the class period. The lawsuit complaint provides a reasonable means to measure the defendant s disclosure during the class period for several reasons. First, as part of a Securities fraud class action under Rule 10b-5 and the PSLRA, plaintiffs must specify each statement alleged to have been misleading in the complaint. (15 U.S.C. 78u-4(b)(1)). Accordingly, each complaint typically contains a separate section listing each allegedly false and misleading statement. 8

9 Second, the plaintiffs have incentives to include as many statements as possible. There is no penalty for including superfluous statements, but once the case is dismissed or settled, any further claims regarding those facts are barred. For example, suppose a case is dismissed because the plaintiffs cannot prove scienter with regard to particular statements from the company s 10-K which they included in the complaint. Now suppose the plaintiffs believe they can prove scienter for another statement involving the same alleged fraud which a CFO made during a conference call. The plaintiffs will likely be barred from bringing this new claim. Before the final disposition of a case, a judge may grant leave to amend the complaint to include new statements, so we use the most recent amended complaint wherever possible. We expect that the suing plaintiffs will thoroughly search all public disclosure by the sued firm and will include in the complaint all of the relevant statements made during the class period. A review of selected lawsuit complaints indicates that plaintiffs indeed tend to be relatively comprehensive in citing disclosures made during the class period. Third, our theory assumes that management and others involved in disclosure decisions perceive the role that their disclosures play in the litigation process. In the complaint, the defendants directly see which statements brought about the litigation. As both sides thoroughly investigate and debate each statement, management clearly sees the impact their own disclosures have on the outcome of the lawsuit. Regardless of actual guilt, managers may perceive that they could avoid liability by never issuing a statement in the first place. Of course, managers cannot simply withhold all information. Section 10(b)-5 also prohibits the omission of material facts. However, there is only liability when omitting such facts causes other statements that were made to be misleading. For example, it is not uncommon for plaintiffs to point to a positive earnings announcement and allege that the defendants already 9

10 knew of negative news at the announcement date, thus making the positive release false or misleading. But the plaintiffs could likely not allege a material omission if the defendants never disclosed the positive news. In fact, the law rarely imposes an affirmative duty to speak that is not in connection with previous or concurrent statements. 1 The example of Concord Camera Corporation illustrates some of these dynamics. (See < The plaintiffs alleged that strong earnings disclosures misled investors because Concord failed to reveal that most of its revenues stemmed from accounts receivable pertaining to one customer, and Concord knew that this customer was unlikely to pay. Further, Concord failed to disclose that large portions of its inventory consisted of custom parts which would lose their value if the customer went out of business. The court found that the plaintiffs could not prove scienter for the initial disclosures because there was not strong evidence that Concord yet knew of the customer s struggles. However, as Concord continued to tout its strong revenues, the court found evidence of scienter for the later statements, as it became obvious with time that Concord knew the customer would not be able to pay its debts. In the case of the inventories, the court found that the plaintiffs only established a strong inference of scienter after the point when Concord completely ceased shipments of inventory to the customer. In this example, while all of the statements turned out to be materially misleading, only some of them were actionable. The statements only gave rise to a strong inference of scienter as they accumulated over time. Management might reasonably perceive that the lawsuit would have been dismissed had they been more tight-lipped. 1 Companies may have duties to disclose when there is a recognized relationship of trust and confidence, when the company itself is trading, or when the company knows insiders are trading on non-public information. See Palmiter (2011) at

11 2.2 Association between Forward-Looking Disclosure and Lawsuit Outcome Our general premise that more disclosure leads to a higher chance of a costly settlement may not hold for forward-looking statements for of the following reasons. First, the law treats such statements differently. Due to the belief that plaintiffs will simply plead fraud by hindsight whenever firms make forecasts that later proved inaccurate, the PSLRA included safe harbor protections for such disclosures. (Securities Act 27A; Exchange Act 21E). Specifically, plaintiffs must show: 1) the defendants had actual knowledge that the forwardlooking statement was false, 2) the statement was material, and 3) the statement was not accompanied by any meaningful cautionary statement. For this last requirement, the statement is protected as long as it is identified as a forward-looking statement and is accompanied by meaningful cautionary statements identifying important factors that could cause actual results to differ. If a statement fails even one of these three prongs, the court will dismiss allegations as to that statement. Here again, the example of Concord Camera Corporation is helpful. After the point in time when the court determined that Concord s non-forward-looking disclosures became actionable, Concord released statements containing much the same information, but phrased in a forward-looking manner. Because brief cautionary language accompanied these statements, the court held that PSLRA s safe harbors protected them and dismissed those statements from the case. Second, prior literature suggests that forward-looking disclosure can improve a firm s lawsuit outcome. As noted, Skinner (1997) examines the relationship between the preemption of bad earnings news and the outcome of litigation. He finds that earlier release of bad news, measured as the number of days before the mandatory earnings announcement, is significantly 11

12 related to smaller settlement amounts, after controlling for estimated damages. He includes dismissed lawsuits as zero dollar settlements, suggesting preemptive forecasts can affect the settlement versus dismissal decision by the court as well. This result comports with the idea that promptly releasing bad news is inconsistent with fraudulent behavior, and thus that such firms will achieve better outcomes in fraud cases. In contrast to our study, Skinner (1997) examines only the presence, but not frequency, of guidance in the quarter before the relevant earnings announcement, and he looks only at bad news guidance. Johnson, et al. (2007) directly measure whether the presence or absence of optimistic and pessimistic forecasts affect the dichotomous settlement or dismissal outcome, but find no significant relationship. However, they only look at a relatively small sample of computer firms, and also measure only the presence, but not frequency, of forecasts. Pritchard and Sale (2005) find that allegations of false forward-looking statements tend to make a lawsuit less likely to be dismissed, but this finding is strictly limited to 65 cases in the Second Circuit, considered the most lenient of the Circuits with respect to the PSLRA. Rogers, et al. (2011) find evidence indicating that optimistic tone in disclosure increases litigation risk, but do not examine the effect on litigation outcome. Overall, prior evidence is somewhat mixed and inconclusive. However, prior studies have been limited in their sample size and measures of disclosure. We capture a much broader sample of lawsuits and use more comprehensive measures of forward-looking disclosure, which allows us to provide more general evidence regarding the role of forward-looking disclosure on a lawsuit s outcome. Finally, it is possible that forward-looking disclosure could decrease the potential for large stock price drops, which could make loss causation more difficult to plead and prove, and 12

13 reduce damages. This possibility along with the legal treatment of forward-looking statements and prior literature leads to our second hypothesis: H2: Forward-looking disclosure issued during the class period will be negatively related to the likelihood of a costly settlement. We also categorize forward-looking guidance based on the nature of its news (i.e. optimistic or pessimistic) and the timing of its issuance relative to the class period and incorporate these measures into our analyses. 3. Research Design 3.1 Sample Selection and Description We start with data from Institutional Shareholder Services (ISS) Securities Class Action Service. ISS provides data for 7,595 lawsuits in the post PSLRA period and for the 43,137 individual securities involved in those lawsuits. 2 Combining these two datasets with each other and with Compustat identifier information results in 4,168 lawsuit observations, containing among other items the outcome (e.g. settled, dismissed, or active) and settlement amount, if any. We remove cases to which the relevant provisions of PSLRA, and thus our theory, may not apply, including state court cases, foreign court cases, SEC enforcement actions, cases with no Rule 10(b)-5 allegations, mutual fund cases, analyst cases, and cases which are part of the large IPO securities litigation master case of the early 2000s. This reduces the sample to 2,512 firmlawsuit observations. When separate defendants in the same lawsuit reach individual settlements with the plaintiffs, ISS reports each settlement separately. We aggregate these duplicate 2 A larger number of securities exist relative to lawsuits due to lawsuits filed against mutual fund families and other securitized assets, as well as firms with multiple classes of stockholders. 13

14 settlements, reducing the number of observations to 1,782. Merging with Compustat and CRSP data necessary to run our models lowers the sample size to 1,669. Our measures of disclosure come from three different sources. We first merge the sample with data from the Stanford Securities Class Action Clearinghouse (SCAC), where we obtain the primary lawsuit documents. This results in a sample of 1,214 observations. We then manually extract the false and misleading statements section from the most recent lawsuit complaint available in SCAC and save it as a separate file. To construct our first measure of overall disclosure, we run each file through a computer program which counts each time the phrase press release appears. Using the forward-looking word list from Bozanic, Roulstone, and Van Buskirk (2012), we also count forward-looking words, such as predict, forecast, or estimate. 3 We are able to obtain and analyze complaint data for 871 firm-lawsuit observations using the complaint-based disclosure measures. To create an alternative measure of disclosure we count the number of press releases issued by a sued firm during the class period using the Factiva database. Specifically, we search for all press releases issued on PR Newswire, Business Wire, or conference call transcripts available via CQ FD Disclosure for the 1,214 firms in our sample. We are able to obtain press release data for 1,079 observations. We use the number of press releases issued as our second measure of overall disclosure during the class period. For a random subset of this sample (622 observations), we download and save the press release files and conference call transcripts for the class period. For this subset of firm-lawsuit observations, we again use the word list from Bozanic et al. (2012) to count forward-looking words. 3 Please see the Appendix for the complete forward-looking word list from Bozanic et al. (2012). 14

15 We use the I/B/E/S Guidance database from Thomson Reuters, which combines forwardlooking guidance from First Call with analyst consensus data at the guidance announcement date from I/B/E/S. We condense all guidance issued during the class period for each firm into mean and aggregate measures and merge with the ISS lawsuit data. Firms with no guidance data receive measures of zero, so no observations are lost in this step. 3.2 Empirical Tests Our tests consist of probit regressions of lawsuit outcome regressed on measures of firm disclosure and control variables. The dependent variable is a binary variable equal to 0 if the case was dismissed or settled for nuisance value, which we define as less than 0.5% of the firm s market value following Johnson, et al. (2007), and 1 if the case settled for a higher amount. 4 Our measures of disclosure are either from the lawsuit complaint or counts of press releases from Factiva. Disc_comp counts the total number of times the phrase press release is found in the false or misleading statements section of the complaint and is intended to measure overall disclosure during the class period. Disc_class counts the number of press releases issued on PR Newswire, Business Wire or conference call transcripts available from CQ FD Disclosure during the class period. We truncate observations at the 1 and 99 percentiles for Disc_comp and Disc_class to remove the effect of outliers and observations where errors may exist in the data. We create three primary measures of a firm s forward-looking disclosure during the class period. FW_disc_comp counts the total number of forward-looking words in the false or 4 In untabulated tests, we also use three alternate dependent variables. First, we equate any settlement less than $1 million with dismissal, following Skinner (1997). Second, we simply classify any settlement as 1 and do not distinguish nuisance cases. Third, we regress the settlement amount as the dependent variable, with dismissals as $0, following Skinner (1997). The results using all alternate outcome variables are substantially the same as our reported tests. 15

16 misleading statements section of the complaint and is scaled by the number of words in this section of the complaint. FW_disc_class counts the total number of forward-looking words across all press releases issued by a firm during the related class period and is scaled by the total words in all the press releases issued during the same period. Fcst_freq is the total number of management forecasts issued during the class period divided by the number of days in the class period, multiplied by 365 following Rogers and Van Buskirk (2009). The management guidance data also allow us to analyze the nature and timing of forward-looking disclosures. Good_fcst_freq counts only forecasts that are greater than analysts consensus at the forecast date, while Bad_fcst_freq counts only forecasts that are below analysts consensus at the forecast date or that announce a charge to earnings. Bad_fcst_early_freq counts the frequency of bad news forecasts in the first two-thirds of the class period and Bad_fcst_late_freq counts the frequency during the final third of the class period. 3.3 Control Variables Past literature has identified several important determinants of lawsuit outcomes, which we include as controls. Johnson, et al. (2007) finds that restatements increase the likelihood of a settlement favoring the plaintiffs. This result makes sense, as courts may use a restatement as concrete evidence of knowledge by defendants that their statements were false. Accordingly, we include restated, a dummy variable equal to 1 if the defendant firm restated its earnings, and we expect a positive coefficient. Pritchard and Sale (2005) find that allegations of GAAP violations increase the probability of a settlement, so we include GAAP as a dummy variable equal to 1 if the complaint alleges violations of GAAP. GAAP violations are not dispositive to a case, but courts consider 16

17 them as evidence of scienter because managers are more likely to know their statements were false if they violated GAAP (Johnson, et al. 2007). Pritchard and Sale (2005) also find that in the Ninth Circuit, allegations of violations of the 1933 Securities Act, that is, allegations related to public offerings of securities, are less likely to be dismissed. To capture this, we include a dummy variable Sec_11 equal to 1 if the complaint alleges violations of Section 11 of the 1933 Act, one of the principle liability provisions of that statute. We also predict that Sec_11 will correlate with a lower probability of dismissal because the pleading standard for public offering fraud is much more lenient than that for 10(b)-5 violations (Palmiter, 2011 at 247). We also include an indicator variable Bankrupt equal to 1 if the defendant firm files for bankruptcy, since this could significantly influence the outcome of the litigation. On the one hand, a bankruptcy trustee may be more inclined to settle the litigation quickly, but a bankrupt firm may also have fewer assets available for a settlement. Therefore we make no prediction regarding this variable. We also include a variable to control for the estimated lawsuit damages because they may affect the incentives of both parties or otherwise influence the outcome of the litigation. Plaintiffs face similar fixed costs and risks for each case, so they may focus more of their time and resources on cases with higher expected payoffs. For defendant firms, high potential damages make it relatively more worthwhile to spend time and legal fees fighting in court rather than settling quickly. We do not predict a direction for this variable because it is unclear whether one side s incentives increase more rapidly than the other as estimated damages rise. The amount of potential damages in the lawsuit is generally a function of how many stock purchases occurred during the class period and how far the stock price dropped. 17

18 We calculate estimated damages following Skinner (1997) as: est_damages = -MVE*MRET*[1-(1-turnover) x ] where MVE equals shares outstanding times price 10 days before the end of the class period, MRET equals the cumulative return over the class period adjusted for the market return, Turnover equals the average ratio of volume to shares outstanding over the class period, and X equals the number of trading days over the class period. The PSLRA, like any statute, leaves much to the interpretation of courts, and each judicial circuit has implemented it differently. Therefore, we include indicator variables for each judicial circuit to control for any differences. Finally, we include the log of total assets (log_at) to control for firm size, we control for two-digit SIC industry code, and we also cluster by twodigit SIC industry. We winsorize all continuous variables at the 1% and 99% levels. We also use robust standard errors in all tests. Our primary tests are probit models of the form: Outcome it = β0 + β1 Disclosure it + β2 Restated it + β3 GAAP it + β4 Sec_11 it + β5 Bankrupt it + β6 Log_AT it + β8 Log_est_damages it + α Circuit_dummies it + γ Industry_dummies it + ε it where Disclosure it is one of our measures of overall disclosure or forward-looking disclosure. 4. Empirical Results Panel A of Table 1 summarizes the outcomes for the final sample and demonstrates the difficulty of surviving a motion to dismiss. The majority (54.6%) of all lawsuits are dismissed in this early stage of litigation or settled for nuisance amounts. If we do not apply the filter for nuisance settlements, approximately ten percent of the dismissal cases swing into the settled category. 18

19 Table 1 Panel B presents descriptive statistics for all variables in our sample. Firms in the sample generally do not issue forward-looking guidance during the class period, as the median frequency is zero. However, the disclosure measures from the complaint provide a more accurate picture of disclosure for all firms, with an average of over thirteen cites to press releases in the complaint, and a median of nine. Allegations of GAAP violations occur in most of our cases, but firms restated earnings only about a third of the time. Bankruptcies and Section 11 allegations are infrequent in our sample, with only eight percent of the sample filing for bankruptcy and fourteen percent accused of Section 11 violations. Table 2 shows the correlation matrix for all the variables. Our complaint-based measure of overall disclosure (disc_comp) and our press-release measure of disclosure are positively related to settlement outcomes, while our forward-looking disclosure measures are negatively related with significant settlements. These univariate results are consistent with our hypotheses. We note that our measures of overall disclosure are positively correlated (0.213) suggesting these measures both capture the disclosures issued by the firm during the class period. 4.1 Overall Disclosure Table 3 presents the results of probit regressions of lawsuit outcome on our overall disclosure measures. Consistent with our first hypothesis, Column 1 shows that the number of press releases cited in a complaint is positively and significantly related to the probability of a substantial settlement. That is, each additional press release that plaintiffs are able to cite increases the probability that the case will not be dismissed. Likewise, Column 2 shows that the number of press releases and conference calls during the class period is positively related to the 19

20 probability of a settlement. Column 3 shows that both measures remain positive and significant when included together in the model. Additionally, we calculate marginal effects as the change in probability of the dependent variable for a one unit change in each explanatory variable. The marginal effects of each disclosure variable are consistently highly statistically significant. For a greater sense of the impact of each variable on the probability of a significant settlement, we multiply the marginal effect times a one standard deviation change in the variable. For Disc_comp, a one standard deviation increase in the number of press releases plaintiffs cite results in an 8.15 percent increase in the likelihood of paying an adverse settlement. For Disc_class, the percentage increase is 6.2 percent. When regressing both disclosure measures together, a one standard deviation increase in Disc_comp continues to increase the probability of settlement by 7.5 percent, while a similar increase in Disc_class has an incremental effect of increasing settlement likelihood by 5.1 percent. Allegations of GAAP violations are consistently positively and significantly related to higher probability of settlement. The marginal increase in the likelihood of a significant settlement given the presence of an alleged GAAP violation is approximately 17 percent. This evidence indicates that judges are indeed relying on GAAP violations as tangible evidence of fraudulent intent. Restatements are also positively related to settlements, and marginally significant in some specifications. When GAAP is removed from the model, Restated becomes highly significant. These results are consistent with past literature and highlight the importance of accounting integrity in litigation. In the post-pslra world, accounting has become a key factor in both the cases that plaintiffs choose to bring to court and the outcome of those cases. 5 5 Johnson, et al. (2007) find that earnings restatements are more likely to result in lawsuits after the PSLRA, and that plaintiffs are more likely to allege accounting violations in these lawsuits. 20

21 Allegations of 1933 Act Section 11 violations strongly correlate with substantial settlements as expected because such violations do not generally require proof of scienter. Bankruptcy also significantly increases the probability of settlement. Due to space limitations, we do not tabulate coefficients on the individual judicial circuits. The Fourth Circuit is the only judicial circuit with a consistently significant effect on the litigation outcome, as cases in that Circuit are more likely to be dismissed. However, only 59 cases, less than 5% of our sample, occur in the Fourth Circuit. Columns 4 through 6 of Table 3 report results with the length of the class period included as a control variable. While length of class period has not been identified in the prior literature as a determinant of litigation outcome, it is possible that fraudulent intent or recklessness may become more apparent and easier to prove the longer a fraud endures, in which case class period length could have an effect on outcome. When we include class length, measured as the log of the number of days in the class period, the coefficients on both of our overall disclosure measures, Disc_comp and Disc_class, remain positive and significant. The coefficients on class length are also positive and significant in all specifications, suggesting that longer class lengths increase the probability of costly settlements. Overall, we interpret the results presented in Table 3 as support of our first hypothesis. When firms release more disclosure for plaintiffs to use in their lawsuit, those firms increase their chance of an adverse outcome to the litigation. This evidence is important and troubling. Congress intended PSLRA to allow firms to disclose information to the market without fear of vexatious lawsuits. While PSLRA may have reduced non-meritorious lawsuits (Johnson, et al. 2007), our results indicate that it has not necessarily encouraged more disclosure. In fact, the 21

22 PSLRA litigation environment may have a chilling effect whereby firms fare better when they disclose less. 4.2 Forward-looking Disclosure Table 4 presents the results of probit regressions examining the effect of forward-looking disclosures on lawsuit outcome. Panels A, B and C report results using three alternative measures of forward-looking disclosure, FW_disc_comp, FW_disc_class and ffcst_freq, respectively. For both the complaint-based forward-looking measure (FW_disc_comp) and management forecasts (Fcst_freq), we find a significant negative relation with the likelihood of settlement. We also find a negative relation using our forward-looking measure based on press releases issued during the class period (FW_disc_class), but it is less significant than the other two measures. Taken together, these results suggest that unlike overall disclosure, forwardlooking statements help the sued firm and increase the chance of dismissal. We find similar results when examining marginal effects of the forward-looking disclosure variables. A one standard deviation increase in the proportion of forward looking words in the complaint decreases the probability of paying a significant settlement by 5.3 percent, while a similar increase in the proportion of forward-looking words in press releases reduces the probability of an adverse settlement by 4.9 percent. For management guidance, a one standard deviation increase in the frequency of forecasts decreases the probability of a settlement by 5.4 percent. We also include our measures of overall disclosure together with the forward-looking disclosure variables to determine whether each type of disclosure variable adds information incremental to the other. Panel A, Column 2 shows that all key variables retain their signs and 22

23 significance when we include the number of press releases cited in the complaint. Specifically, the overall disclosure measure, disc_comp, is positive and significant whereas FW_disc_comp is negative and significant. This finding highlights the importance of distinguishing forwardlooking and non-forward-looking disclosures, and shows that each type has different implications for the outcome of litigation. Columns 3 and 4 of Panel A report results with class period length included as a control variable. The direction and significance of the disclosure variables and their marginal effects do not change in this specification, and Log_class_length is positive and significant. Similar to Panel A, Column 2 of Panel B shows both FW_disc_class and Disc_class retain their signs and significance levels when included together in the model. Columns 3 and 4 report results with class period length included. As reported in Column 4 of Panel B, disc_class loses its significance when Log_class_length is included in the model. Columns 2 and 3 of Panel C report results when measures of overall disclosure are included with management forecasts (Fcst_freq) as our measure of forward-looking disclosure. Because we are not able to construct an overall disclosure variable using management forecast data, we rely on the overall disclosure measures based on the number of press releases cited in the complaint (Disc_comp) and the press releases and conference calls issued during the class period (Disc_class). Consistent with the results in Panels A and B, both Disc_comp and Disc_class are positive and significant whereas Fcst_freq is negative and significant. Columns 5 and 6 report results with class period length included as a control variable. The direction and significance of the disclosure variables do not change when class length is included, and Log_class_length is positive and significant. 23

24 We interpret the results presented in Table 4 as support of our second hypothesis. In sum, they suggest that the PSLRA protections for forward-looking statements have largely been successful, as firms with more forward-looking disclosure are less likely to pay substantial settlements, all else equal. Because optimistic and pessimistic forecasts may influence litigation differently, we also include the frequency of each type of forecast as separate independent variables. We classify management forecasts as optimistic or pessimistic depending on whether they are greater than or below analysts consensus at the forecast date, and include these classifications in our probit model. Table 5 presents the results of these tests. Column 1 of Table 5 shows that while both types of forecast retain negative coefficients, bad news forecasts are significantly negative but good news forecasts are not significant. Similarly, the only significant marginal effects are those of the bad news forecasts, and the significant effects are concentrated among these forecasts at the end of the class period. A one standard deviation increase in the frequency of pessimistic forecasts late in the class period decreases the likelihood of a significant settlement by six percent. This result is consistent with Skinner (1995, 1997), which finds that bad news forecasts decrease the likelihood and severity of litigation. We next investigate whether the timing of bad news forecasts during the class period affects their relation with outcome. We separate the frequency of bad news forecasts issued during the first two thirds and last one third of the class period, and refer to these measures as bad_fcst_early_freq and bad_fcst_late_freq, respectively. Column 3 of Table 5 shows that only bad news guidance issued toward the end of the class period is significantly related to outcome. Bad news guidance issued earlier in the class period does not appear to be significantly related to litigation outcome. 24

25 One might expect negative news to matter more when it is released earlier, not later, as in Skinner (1997), which focuses on the timeliness of the disclosure of adverse earnings news and its effect on settlement amounts. The key difference between our research design and Skinner s is that we use the end of the class period as a benchmark for our measure of earliness, while Skinner uses the mandatory earnings announcement date. Because the end of the class period is defined as when the negative news is revealed to the public, it is likely that the bulk of all bad news forecasts occur near the end of the class period. Indeed, in our sample the mean frequency of bad news guidance in the last third of the class period is greater than the mean during the first two-thirds combined. Overall we believe these results generally support our second hypothesis, and more specifically suggest that voluntarily releasing negative forecasts increases the probability that a lawsuit will be dismissed, whereas issuing good news forecasts does not appear to have an effect on the probability a lawsuit will be dismissed. These results have two important implications. First, they show that the PSLRA protections for forward-looking statements have largely been successful, as firms with more guidance are less likely, or no more likely, to pay substantial settlements, all else equal. Second, if forward-looking statements tend to improve (or have no effect on) outcomes for the sued firm, then the result of Rogers and Van Buskirk (2009) that such statements decrease after litigation is still puzzling. In light of our earlier results that the overall number of disclosures does harm the firm in litigation, firms may simply perceive that all disclosure is harmful and reduce all voluntary disclosure in response, both non-forward-looking and forwardlooking. Another possibility is that even if firms correctly perceive that forward-looking disclosure overall is not likely to result in liability, they can still conclude that some individual 25

26 forecasts do result in liability, or at least in lawsuits. So while legal risk is less for forwardlooking statements, the risk still exists. Further, the largest litigation-related benefits come from bad news guidance, so firms may strategically use pessimistic forecasts while reducing all others. 5. Conclusion This study shows that disclosure can be an important determinant of securities fraud class action outcomes. It extends the important accounting literature on litigation risk to show the effect of disclosure on another facet of litigation: outcome. Skinner (1997), Johnson, et al. (2007), and Pritchard and Sale (2005) provided important preliminary work in this area. We place disclosure in the context of the PSLRA litigation environment and directly measure the effects of disclosure on litigation outcomes for a large sample of lawsuits and firms. We show for the first time that disclosure can lead to adverse litigation outcomes for the firm. When plaintiffs have more public disclosure available to include in their complaints, they have a higher chance of overcoming the PSLRA s high hurdles and avoiding dismissal. We also distinguish between non-forward-looking disclosure and forward-looking disclosure. Our evidence indicates that the PSLRA s protections for forward-looking statements appear to have been effective. Managers may generally provide guidance about the future in good faith without fearing that such statements will increase the chance of an adverse lawsuit outcome. In many cases, our tests show that voluntarily releasing negative guidance can lead to better outcomes. However, forward-looking statements can still be the basis for litigation and liability, so it is still possible firms may rationally reduce guidance to minimize legal exposure. Our results call for further study of how private enforcement of securities laws in the U.S. affects disclosure decisions by firms. While the possibility of private legal liability surely 26