Litigation Risk and Corporate Voluntary Disclosure: Evidence from Two Quasi-Natural Experiments

Litigation Risk and Corporate Voluntary Disclosure: Evidence from Two Quasi-Natural Experiments Hui Dong School of Accountancy Institute of Accounting and Finance Shanghai University of Finance and Economics Huai Zhang Nanyang Business School Nanyang Technological University Abstract We examine the effect of litigation risk on corporate voluntary disclosure using two quasi-natural experiments, which have substantial and opposing impacts on the litigation risk of firms headquartered in the Ninth Circuit. We find that firms in the Ninth Circuit decrease (increase) the quantity and quality of their voluntary disclosure, relative to control firms, when their litigation risk is lowered (elevated). The pre-treatment test shows an indistinguishable trend between treatment and control firms. A battery of robustness checks indicates that our results are not driven by alternative explanations. We hypothesize and find that the impact of litigation risk is more pronounced when firms have bad news and that firms are more likely to preempt bad news through voluntary disclosures when litigation risk is elevated. Overall, results from both experiments suggest that litigation risk causally increases corporate voluntary disclosure. Keywords: Litigation Risk, Voluntary Disclosure, the Ninth Circuit Court. JEL Classification: G3; G38; M41. Corresponding author. Tel: (+86)-21-6590-4833, Email: dong.hui@mail.shufe.edu.cn. Postal address: No. 777 Guo Ding Road, School of Accountancy, Shanghai University of Finance and Economics, Shanghai, China. Tel: (+65)-6790-4097, Email: huaizhang@ntu.edu.sg. We acknowledge helpful comments from Zhaoyang Gu, Allen Huang, Bin Ke, Clive Lennox, Chen Lin, and seminar participants at Nanyang Technological University and Shanghai University of Finance and Economics. Dong acknowledges the financial support from the National Natural Science Foundation of China (No.71401095), and the MOE project of Key Research Institute of Humanities and Social Science in University (No.18JJD790011). Zhang acknowledges the financial support from Singapore Ministry of Education for providing research funding (RG75/16, and RG163/17). All errors are our own.

Litigation Risk and Corporate Voluntary Disclosure: Evidence from Two Quasi-Natural Experiments Current version: November 2018 First version: March 2015 Abstract We examine the effect of litigation risk on corporate voluntary disclosure using two quasi-natural experiments, which have substantial and opposing impacts on the litigation risk of firms headquartered in the Ninth Circuit. We find that firms in the Ninth Circuit decrease (increase) the quantity and quality of their voluntary disclosure, relative to control firms, when their litigation risk is lowered (elevated). The pre-treatment test shows an indistinguishable trend between treatment and control firms. A battery of robustness checks indicates that our results are not driven by alternative explanations. We hypothesize and find that the impact of litigation risk is more pronounced when firms have bad news and that firms are more likely to preempt bad news through voluntary disclosures when litigation risk is elevated. Overall, results from both experiments suggest that litigation risk causally increases corporate voluntary disclosure. Keywords: Litigation Risk, Voluntary Disclosure, the Ninth Circuit Court. JEL Classification: G3; G38; M41. 1

1. Introduction How does litigation risk causally affect firms voluntary disclosure? The answer to this important question is far from obvious. On one hand, high litigation risk may lead firms to provide more disclosures, because full disclosure invalidates the plaintiff s claim of firms withholding information intentionally and helps to avoid the stock price crash, a potential trigger of investor lawsuits (Skinner 1994; Skinner 1997; Field et al. 2005). On the other hand, high litigation risk may deter firms voluntary disclosure, especially forward-looking disclosures, because such disclosures may be inaccurate ex post and thereby provoke lawsuits (Johnson et al., 2001; Baginski et al., 2002; Rogers and Van Burskirk 2009). Given the conceptual ambiguity and academic interest, more evidence on this topic seems warranted. Inquiries into this causal effect however are challenging, because litigation risk and voluntary disclosure are likely to be mutually dependent. We address this endogeneity issue by taking advantage of two natural experiments where the variation in the litigation risk is exogenous to firms voluntary disclosure. 1 Both experiments are related to the interpretation of the pleading standards of the Private Securities Litigation Reform Act of 1995 (PSLRA). These pleading standards are conditions that must be satisfied for shareholders to legally form a class. 2 The first natural experiment is the ruling by the U.S. Ninth Circuit Court of Appeals on July 2, 1999. In the ruling Re: Silicon Graphics Inc. Securities Litigation (SGI), the Ninth Circuit Court issued the most stringent interpretation of the pleading standards, requiring that plaintiffs offer proof that the defendants acted with deliberate recklessness. This compares to other circuits, where mere recklessness was sufficient. This ruling makes it much more difficult for 1 Our use of the U.S. data may limit the generalizability of our results. We however believe that the following points help to alleviate the concern and justify our choice of the U.S. setting. First, the research question how litigation risk influences corporate disclosure is likely relevant and interesting globally. Second, the U.S. has the largest economy and the largest capital market in the world. Therefore, results based on the U.S. data are likely to be economically important worldwide. Third, the U.S. capital market has a wealth of information, allowing us to easily obtain information on managerial voluntary disclosures. 2 Please refer to the online appendix for a detailed discussion about the institutional setting. 2

investors to file class action litigations against firms headquartered in the Ninth Circuit. Crane and Koch (2016) document that after the ruling, the number of class action litigations drops by 43% in the Ninth Circuit, while it increases by 14% in all other circuits. Given the types of claims in private securities class action litigation and the fraud on the market presumption (Halliburton Co. v. Erica P. John Fund, Inc., 134 S. Ct. 2398 (2014); Basic Inc. v. Levinson, 485 U.S. 224 (1988)), almost all witnesses and evidence are likely located at the defendant firm s headquarter. As a result, although private securities class action litigations can be brought in any of the federal circuit court, the defendant corporation can move to relocate the lawsuit to the district where the firm is headquartered, on the grounds that most witnesses and documents reside in this district (Cox et al., 2009; Jhong 2016). Cox et al. (2009) show that 85% of class action cases are consolidated to the home circuit of the defendant corporation. The second natural experiment is the ruling Tellabs v. Makor Issues and Rights, Ltd. (Tellabs) by the Supreme Court on June 21, 2007. The Tellabs decision represents the Supreme Court s interpretation of the pleading standards of PSLRA. It effectively relaxed the stringent requirement by the Ninth Circuit and made it much easier to file class action lawsuits against firms headquartered in the circuit. Choi and Pritchard (2012) show that the class action litigations are less likely to be dismissed and the likelihood of low-value settlements increases in the Ninth Circuit, relative to other circuits, after the Tellabs ruling. There are several reasons why we choose the two natural experiments for our analyses. First, the SGI ruling is a judicial decision made by three judges randomly selected from a pool of Ninth Circuit judges, while the Tellabs ruling is a legal decision made by the Supreme Court. Since the rulings are unlikely affected by firms voluntary disclosure, they represent exogenous shocks to firms litigation risk. Second, given that most securities lawsuits are eventually 3

litigated in the state where the firm is headquartered, the two ruling altered litigation risk for firms headquartered in the Ninth Circuit relative to firms headquartered elsewhere (Cox et al. 2009; Choi and Pritchard, 2012), allowing us to conduct a difference-in-differences analysis. Third, and the most important reason for us to choose two instead of one ruling lies in the strikingly similar macroeconomic situations surrounding the two rulings. Both were preceded by a stock market boom and succeeded by a stock market crash. 3 We document that firms alter their voluntary disclosure in opposite directions after the two rulings. Since the macroeconomic circumstances around the two rulings are similar, it is difficult to attribute our findings to macroeconomic factors concurrent with the rulings. Using a sample of 12,495 firm-year observations from the period 1996-2006 for the SGI ruling and a sample of 6,382 firm-year observations from the period of 2006-2010 for the Tellabs ruling, we empirically test the causal effect of litigation risk on firms voluntary disclosure. We employ the difference-in-differences research design, and identify firms under the jurisdiction of the Ninth Circuit Court as treatment firms and other firms as control firms. We compare between treatment firms and control firms the changes in firms voluntary disclosure after the SGI and Tellabs rulings, respectively. This research design helps to control for both non-time-varying firm characteristics and changes in macro-level factors that have similar influences on treatment firms and control firms. We adopt voluntary disclosure measures in Rogers and Van Buskirk (2009) in our analyses. Our results show that, relative to control firms, treatment firms reduce voluntary disclosure after the SGI ruling. Specifically, treatment firms become less likely to issue management forecasts. When they issue management forecasts, their forecasts have a shorter 3 The SGI ruling was followed by the burst of the internet bubble while Tellabs ruling the financial crisis of 2008. 4

horizon and become less precise and less specific. Our results are economically significant. The relative drop in forecast frequency experienced by treatment firms represents a 51.4% change, using the sample mean as the basis, while the decreases in forecast horizon, forecast precision, and forecast specificity are about 23.8%, 9.4%, and 86.9% of the sample mean, respectively. Results based on indirect measures of voluntary disclosure present a similar picture. We document an increase in both absolute earnings surprises and absolute abnormal returns around earnings announcement dates for treatment firms, relative to control firms. The increases are substantial, representing about 10.7% and 15.1% of the sample mean value, respectively. The bigger earnings surprises and higher magnitude of announcement returns are consistent with a decrease in voluntary disclosure. Turning to the Tellabs ruling, which increases the litigation risk for treatment firms, our results show that, after the ruling, relative to control firms, treatment firms increase the quantity and quality of voluntary disclosure. After the ruling, treatment firms are more likely to issue management forecasts, their forecasts are more precise, more specific and of longer horizon, and their earnings surprises are of lower magnitude. Our results are economically significant. The increases in forecast frequency, forecast horizon, forecast precision, and forecast specificity of treatment firms relative to control firms amount to 27.6%, 12.5%, 3.2%, and 50% of the sample mean value, respectively. In sum, our results from both experiments indicate that litigation risk causally increases firms voluntary disclosure. A central assumption in drawing causality inference from the difference-in-differences estimation is that, absent the two court rulings, treatment firms voluntary disclosure would have evolved in the same way as that of control firms. This assumption is inherently untestable, because we don t get to observe voluntary disclosures of treatment firms when there were no 5

issuance of the Rulings. We can however provide some peripheral evidence by examining the pre-treatment trends. Our test, reported in Table A1 of the online appendix, shows that the pre-treatment trends are indistinguishable between treatment and control firms for both rulings. Under the assumption that the treatment effect does not alter the trends, such evidence lends credibility to our difference-in-differences approach. Moreover, the reduction (increase) in treatment firms voluntary disclosures takes place after the SGI (Tellabs) ruling, suggesting a causal effect. In addition, we have to ensure that the treatment effect is uncorrelated with covariates that affect firms voluntary disclosure, in order to draw causal inferences from our results. We are mainly concerned that state or geographic factors may explain our findings, because the jurisdiction of the Ninth Circuit Court extends to firms headquartered in certain states located mainly in the west of the U.S. We conduct a series of robustness checks to address this and other concerns. First, we use the propensity score matching method. For each treatment firm, we choose the control firm with the nearest propensity matching score. We obtain similar inferences from this approach, suggesting that different firm characteristics of treatment and control firms do not explain our findings. Second, we separately analyze treatment firms with a high and low proportion of out-of-state sales. If our results are driven by state-specific conditions, we predict that the treatment effect is weaker for firms with a higher proportion of out-of-state sales, since the fortune of these firms is less dependent on the local state. Our empirical results are reported in Table A2 of the online appendix and they do not support this prediction. Third, since many of the internet-related businesses and high-tech firms are located in the 6

Ninth Circuit, our results may be driven by these firms. We test this alternative explanation by removing these firms from our sample. We find that our results continue to hold, evidence inconsistent with the notion that our results are due to characteristics of firms located in the Ninth Circuit. Our results are reported in Table A3 of the online appendix. Fourth, we examine how the treatment effect varies with ex ante litigation risk. Since the SGI ruling lowers while the Tellabs ruling elevates the litigation risk for the treatment firms, if the effect on treatment firms goes through the litigation risk channel, we expect it to be more pronounced for firms with a higher (lower) ex ante litigation risk in the SGI ruling (the Tellabs ruling). Our empirical results confirm that the treatment effect is indeed through the litigation risk channel. Next, Francis et al. (1994) show that litigation risk mainly affects firms with bad news. We thus hypothesize that the impacts of the two court rulings are more pronounced for firms with bad news. We use both the average quarterly earnings surprises and earnings announcement returns to gauge the nature of the news. Results from both measures support our prediction. It is difficult to see why the treatment effect is more pronounced for bad news firms if our main finding is driven by other factors. Furthermore, several studies show that firms preempt bad news through voluntary disclosures and highlight the role of disclosure in reducing litigation risk (Skinner 1994; Kasznik and Lev, 1995; Field et al., 2005; Billing and Cedergren., 2015). We therefore hypothesize that firms are more likely to preempt bad news when litigation risk increases. Consistent with this hypothesis, we show that firms are more likely to issue bad news forecasts, and these bad news forecasts are timelier, when litigation risk is elevated. Finally, our un-tabulated results show that our conclusions continue to hold, when we 7

restrict our sample period to one year before and one year after the two rulings. This result not only addresses the concern that the pre-ruling and post-ruling periods are not of equal length, but also helps to attribute the changes in firms disclosures to the two rulings. Our primary contribution to the literature lies in that we deepen our understanding of the causal effect of the litigation risk on voluntary disclosure. This effect has received academic attention, dating back to Skinner (1994). We take advantage of the natural experiments offered by court rulings to address the endogenous relationship between litigation risk and voluntary disclosure. We find that firms respond to the lowered (heightened) litigation risk by reducing (increasing) their voluntary disclosure. A battery of robustness checks indicates that our findings are unlikely due to alternative explanations based on state- or geographic factors. We therefore bring useful new evidence, reasonably free from biases induced by the endogeneity issue, to a topic that has intrigued accounting researchers for decades. Two contemporaneous papers study the SGI ruling. Cazier et al. (2017) show that firms increase non-gaap reporting in response the SGI lawsuit, while Houston et al. (2015) find that firms issue fewer earnings forecasts and in a less timely fashion in response to the SGI case. We differ from Cazier et al. (2017) in that we analyze the management s disclosure of earnings forecasts, essentially forward-looking information, while they focus on the management s disclosure of non-gaap earnings, essentially historical information. In addition, while we show a reduction in management forecasts after the SGI ruling, they find an increase in non-gaap earnings disclosures; that is, the two papers reach distinct conclusions. As discussed in Zhang and Zheng (2011) and other studies, pro forma earnings do not have a fixed or official definition. Given such, investors can rarely call foul on pro forma earnings disclosures. In contrast, management forecasts, if they deviate from actual earnings, can easily trigger a lawsuit. 8

When litigation risk is lowered, firms tendency to preempt bad news is reduced, resulting in lower likelihood of management forecasts. At the same time, firms may want to increase disclosure of pro forma earnings that s insensitive to litigation risk changes, to compensate for the drop in disclosure as a result of the reduction in management forecasts. The finding in Houston et al. (2015) is similar to our finding related to the SGI lawsuit. The biggest difference between our paper and Houston et al. (2015) lies in the research design. Houston et al. (2015) focus almost exclusively on the SGI ruling, while we examine both the SGI ruling and the Tellabs ruling. This research design difference is vital because the SGI lawsuit is concurrent with many macro-events, such as the burst of internet bubble, and the crash of the stock market. Therefore, it s very hard to pinpoint the SGI ruling as the cause of the change in firms disclosures. By using two shocks in similar macro-economic situations and documenting different disclosure patterns, we are able to establish a much more reliable link between the lawsuits and firms disclosures, avoiding the common criticism on drawing inferences from one single event. Our use of the U.S. data may limit the generalizability of our results to other countries. The legal system differs across countries. While class action suits are available in the U.S., it is not in the European Union 4, Singapore 5 or Hong Kong 6 (prior to 2012). Our conclusion therefore isn t readily applicable to non-u.s. countries. What s more, cross-border differences likely influence the impact of litigation risk on firms disclosures by altering firms cost-and-benefit calculations. For example, while we show that litigation risk increases voluntary disclosures, conceivably, in countries where law enforcement is poor, litigation risk may have little impact on firms disclosure, because the consequence of litigations is unlikely to 4 See https://money.cnn.com/2018/04/11/news/europe-class-action-lawsuits-legal/index.html 5 See http://www.asiaone.com/business/are-class-actions-good-investors-and-singapore-markets 6 See https://www.hkreform.gov.hk/en/publications/rclassactions.htm 9

materialize. The rest of the paper proceeds as follows. Section 2 reviews prior literature and develops hypotheses. Section 3 discusses research design, variable measurement and sample construction. Section 4 presents our main empirical results. Section 5 concludes. 2. Literature review and hypothesis development Conceptually, it is unclear how litigation risk affects voluntary disclosure. When firms face higher litigation risk, firms may increase their voluntary disclosure due to the following two reasons. First, more disclosures help to invalidate potential litigants claim that the firm withholds information. Second, frequent disclosures reduce the likelihood of large stock price decline by gradually adjusting investors expectations. In contrast, infrequent disclosures may result in dramatic market reactions. Research consistent with the positive impact of litigation risk on voluntary disclosure dates back to Skinner (1994), who examines voluntary disclosure for a sample of NASDAQ firms and shows that these firms exhibit a tendency to pre-announce large negative earnings surprises. He suggests that this behavior could be driven by litigation risk concerns. Francis et al. (1994) however find that management forecasts are more common for firms that are ultimately sued than for firms that are at risk. Their findings cast doubt on the ability of voluntary disclosure to deter shareholder lawsuits. Skinner (1997) shows that although voluntary disclosure does not affect the likelihood of lawsuits, it reduces the payment the defendant corporation has to make in settlements, i.e., it reduces the firm s litigation costs. Field et al. (2005) recognize the endogenous relation between litigation risk and voluntary disclosure. After they control for the 10

endogeneity, they find that voluntary disclosure deters litigations. 7 Cao and Narayanamoorthy (2011) advocate the Directors and Officers liability insurance premium as a new measure of firms ex-ante litigation risk. Using this measure, they find that, when faced with ex ante litigation risk, managers with bad news are more likely to issue an earnings warning. Donelson et al. (2012) find that timely revelation of bad earnings news lowers the likelihood of litigation while Billings and Cedergren (2015) show that the firm s litigation risk increases when the management fails to warn investors of impending bad news. Naughton et al. (2018) examine foreign firms cross-listed on U.S. exchanges whose private litigation costs are lowered by the Supreme Court ruling in Morrison v. National Australia Bank. They find that these firms reduce their voluntary disclosure relative to a matched sample of U.S. firms. Their results are based on a small sample of cross-listed foreign firms, and it is unclear whether their results apply to U.S. firms or to firms not cross-listed. Levy et al. (2016) report that firms whose CFO s personal litigation risk is increased by the Delaware Supreme Court s Gantler v. Stephens (2009) decision tend to disclose negative news early. Their results are mainly related to managerial personal litigation risk but not firms litigation risk. Hopkins (2014) uses the same setting as our study. He examines how litigation risk affects earnings management while our paper focuses on the impact of litigation risk on voluntary disclosure. However, there also exist arguments that firms reduce voluntary disclosure when their litigation risk increases. While voluntary disclosure may be unbiased at the time of issuance, they may be inaccurate ex post, triggering shareholder lawsuits. In addition, once a forward-looking disclosure is made, firms face the obligation to update it in a timely fashion. Failing to do so may provoke shareholder litigations. 7 While prior study (i.e., Field et al. 2005) uses a simultaneous equations methodology to deal with this issue, this methodology is subject to limitations and concerns. For example, Field et al. (2005) acknowledges that the effectiveness of this approach hinges on the appropriateness of the identifying variable, which should be directly correlated with either litigation risk or voluntary disclosure, but not directly correlated with the other. Failure to satisfy this requirement will lead to biased coefficient estimates. 11

Several papers offer empirical evidence that litigation risk reduces voluntary disclosure. Johnson et al. (2001) find that their sample firms increase voluntary disclosure when their litigation risk is lowered by the safe harbor provision of the PLSRA. Baginski et al. (2002) show that firms in Canada, a country with a less litigious environment, in general provide more voluntary disclosure than firms in the U.S. Rogers and Van Buskirk (2009) examine the change in disclosures for firms involved in shareholder litigations, and they find that these firms curtail voluntary disclosure after the lawsuit. Lowry (2009) however points out that their evidence is based on a sample of firms that are actually sued and it may be un-generalizable to the larger population of firms that are not sued. Bourveau et al. (2017) document an increase in disclosure after the adoption of universal demand (UD) laws, which make it more difficult to file derivative lawsuits. They also show that UD laws have no asymmetrical effect on good news and bad news disclosures. This result is inconsistent with the well-established finding that litigations have a greater impact on disclosures of bad news than on disclosures of good news (Skinner, 1994; Francis et al., 1994; Kasznik and Lev, 1995; Field et al., 2005, Billings and Cedergren, 2015). Our interpretation of Bourveau et al. (2017) is that they speak more to the indirect effect of litigations on disclosure (i.e., through corporate governance) than to the direct effect of litigations on disclosure, and the latter is the focus of our study. Given the conceptual ambiguity and mixed prior results, we present our hypothesis in its null form: H1: Litigation risk has no impact on voluntary disclosure. While prior studies do not offer a clear answer as to how litigation risk affects voluntary disclosure, they provide unambiguous evidence that there is mutual dependence between voluntary disclosure and litigation risk. Therefore, there is a need to control for the endogenous 12

relationship between the two. In this paper, we use the SGI and Tellabs rulings on class action litigations as natural experiments to control for the endogeneity and we provide more descriptions about the experiments in the next section. Francis et al. (1994) show that bad news firms have higher likelihood of being sued, because bad news typically results in investors wealth losses. Given that litigation risk is a more important concern for bad news firms, if the decrease (increase) in treatment firms voluntary disclosure is due to the reduced (increased) litigation risk caused by the two rulings, respectively, we hypothesize that the decrease (increase) is more pronounced for treatment firms with bad news. This yields H2: H2: The impact of litigation risk on voluntary disclosure is more pronounced for firms with bad news. An influential early study, Skinner (1994) shows that to reduce litigation risk, firms preempt bad news through voluntary disclosures, because disappointing earnings news may result in shareholder lawsuits. Kasznik and Lev (1995) find consistent evidence that firms provide earnings warnings to avoid large negative earnings news. Field et al (2005) and Billings and Cedergren (2015) demonstrate that early disclosure of earnings-related bad news lowers litigation risk. These papers highlight the role of disclosures in reducing litigation risk. Therefore, we hypothesize that firms are more likely to pre-empt bad news when litigation risk increases. H3: Firms likelihood to preempt bad news is higher when litigation risk increases. 3. Research design, variable measurement and sample construction 3.1. Research design We use a difference-in-differences design to examine the impact of the two court rulings in the Ninth Circuit on voluntary disclosure. Specifically, we estimate the following model: 13

y it = α i + α t + β Treat i Post t + γ Control it 1 + ε it (1) Where y it represents one of the measures of voluntary disclosure for firm i in year t. α i is the firm fixed effects and α t is the year fixed effects. Treat i is a dummy, which equals one for treatment firms, i.e., firms in the Ninth Circuit, and zero otherwise. Post t equals one if the current year is after year 1999 and zero otherwise in the SGI ruling; and it equals one if the current year is after year 2007 and zero otherwise in the Tellabs ruling. Treat i and Post t are not included as standard-alone variables, because they are absorbed by firm and year fixed effects. The independent variable of interest is the interaction term, Treat i Post t. Its coefficient indicates the difference in the post-ruling change in voluntary disclosure between treatment firms and control firms. We also control for several firm characteristics which may affect firms disclosure strategies. The standard errors are corrected for clustering at the firm level. 8 Our next subsections discuss how we measure voluntary disclosure and control variables. 3.2. Voluntary disclosure measures Following prior literature (Baginski et al., 1993, 2002; Baginski and Hassell, 1997; Bamber and Cheon, 1998; Rogers and Van Buskirk, 2009; Vashishtha, 2014), we use management-forecast based measures to assess firms voluntary disclosure. Our measures are as follows. Forecast frequency is the total numbers of annual and quarterly EPS forecasts provided by management in the current year. A higher value indicates more frequent disclosures. Forecast horizon is the end date of the fiscal period being forecasted minus the date 8 Our inferences remain unchanged if the standard errors are clustered at the industry level or at the state level. 14

when the management forecast is issued. 9 For example, if a firm makes a forecast on October 31, 2001 for the fiscal year ending on December 31, 2001, the horizon for the forecast is 61 days. A higher value indicates that the forecast is made earlier and conceivably more informative. Forecast precision equals 4 for point estimates, 3 for range estimates (i.e., both minimum and maximum estimates provided), 2 for open-ended estimates (i.e., minimum or maximum provided, but not both), and 1 for qualitative estimates (e.g., about breakeven to slightly positive ). A Higher value indicates a more precise forecast. Forecast specificity equals zero for point estimates. For range estimates, it s computed as the negative value of the forecast range (the top minus the bottom), deflated by pre-forecast share price (*100). A higher value indicates that the forecast is more specific. Prior research finds that more specific estimates are associated with stronger price reactions and greater liquidity benefits (Baginski et al., 1993; Rogers, 2008). Bertrand et al. (2004) provide evidence that serial correlation among observations leads to under-estimated standard errors in OLS regressions when researchers implement the difference-in-differences research design. They recommend using the mean value of the data as a way to reduce serial correlation and produce unbiased inferences. Following their recommendation, when firms issue multiple management forecasts in one year, we compute Forecast horizon, Forecast precision, and Forecast specificity for each management forecast and then take the average of the values. Our inferences are the same if we conduct our analyses without aggregating to the firm-year level. To accommodate the possibility that firms use other channels, such as conference calls or earnings press releases, for voluntary disclosure, we examine two indirect measures: absolute 9 Using this definition, preannouncements have negative forecast horizons. Our results are robust to the sample excluding the preannouncements. 15

earnings surprises and absolute earnings announcement returns, to capture the aggregate effect of a firm s voluntary disclosure. Both measures are on a firm-year basis. Absolute earnings surprisesit is the average absolute value of the actual quarterly earnings of firm i less the most recent analyst consensus estimate, deflated by the stock price prior to the earnings announcement in year t (*100). Absolute abnormal returnsit is the average absolute value of the size-adjusted returns for the three-day window centered on the quarterly earnings announcement date for firm i in year t (*100). If managers provide more voluntary disclosure, investors will be less surprised by earnings announcements and both variables will take on lower values. 3.3. Control variables We include control variables that have been used in prior literature in their enquiries on voluntary disclosure: Size, Leverage, ROA, Log(Analyst), Volatility, BTM, and IO. Firm size (Size) is measured by the natural logarithm of the market value of equity at the end of the fiscal year (Rogers and Van Buskirk, 2009; Ali et al., 2014). Leverage is defined as long-term debt plus current liabilities deflated by total assets and it controls for the effect of capital structure on voluntary disclosure (Rajan, 1992; Dass and Massa, 2011). ROA is operating income before depreciation divided by total assets and it captures the impact of firm performance on firms disclosures (Vashishtha, 2014). Log (Analyst) is the natural logarithm of one plus the number of analysts following the firm and it reflects analysts demand for voluntary disclosure (Baginski and Hassell, 1997; Balakrishnan et al., 2014; Rogers and Van Buskirk, 2013). Volatility, a measure of uncertainty, is computed as the standard deviation of daily stock returns over the prior year. Prior literature shows that firms are less likely to disclose information when there is high uncertainty (Li, 2010; Ali et al., 2014; Balakrishnan et al., 2014). BTM is the 16

book-to-market ratio and it reflects growth opportunities. Prior research has shown that firms growth opportunities affect firms voluntary disclosure (Bamber and Cheon, 1998; Rogers and Van Buskirk, 2009; Vashishtha, 2014). IO is the percentage of common shares held by institutional investors. 10 Prior studies find that institutional ownership increases firms information environment such as the quantity and quality of management forecasts (Boon and White, 2016). Since our treatment firms are located in specific states, the state economy may play a role in explaining our finding. To address this concern, we add two variables representing state economy: state GDP growth rate, which is obtained from the Census Bureau, and state unemployment rate, which is obtained from the Bureau of Labor Statistics. 3.4. Sample construction We obtain firms financial information, stock return information and analyst forecast information from Compustat, CRSP and I/B/E/S. Our sample firms have to be U.S. firms so that our results are unaffected by differences in regulations across countries. Financial firms and utility firms are eliminated from our sample because these firms are highly regulated and face a different legal environment. We obtain annual and quarterly management forecasts of earnings per share (EPS) from Company Issued Guidance section under First Call Historical Database (FCHD) by Thompson Reuters for our sample firms. For the SGI ruling, we choose our sample period to be between 1996 and 2006 to avoid important confounding legal changes. Our sample starts in year 1996, because in December 1995, Congress enacted the PSLRA. If we extend our sample to the 10 Bourveau et al. (2017), Appel et al. (2016) and Crane and Koch (2017) show that firms may change their corporate governance in response to changes in litigation risk. It is therefore possible that the impact of litigation risk on voluntary disclosure goes through corporate governance. We find that several coefficients in Table 3 exhibit lower absolute value after we control for institutional ownership, which seems to suggest that concurrent changes in corporate governance are responsible for the changes in voluntary disclosure. 17

periods prior to 1996, our difference-in-differences research design may be affected by the enactment of the PSLRA. For a similar reason, our sample period ends in 2006, as the Supreme Court issued the Tellabs ruling in 2007, which affected the litigation risk of all firms. Since the SGI ruling occurred in July of 1999 and it is ambiguous whether we shall classify observations in 1999 as either before or after the ruling, we remove these observations from our sample. For the Tellabs ruling, we choose our sample period to be between 2006 and 2010. Since the ruling occurred in June 2007, we remove the observations in 2007 from our sample. When firms issue multiple forecasts for a year, we compute forecast-based measures (Forecast horizon, Forecast precision, and Forecast specificity) for each forecast and then take the average of the values so that our forecast-based measures are on a firm-year basis. 11 As we discussed earlier, the aggregation to the firm-year level reduces the serial correlation among observations and helps to produce unbiased inferences in OLS regressions which implement the difference-in-differences research design (Bertrand et al., 2004). Since we use a difference-in-differences research design, we require our sample firm to appear both before and after the SGI ruling and the Tellabs ruling, respectively. 12 We require all control variables (Size, Leverage, ROA, Log(Analyst), Volatility, BTM, and IO) to be non-missing. Our final sample for the SGI ruling consists of 12,495 firm year observations with 2,668 of them belonging to treatment firms, while our Tellabs sample consists of 6,382 firm year observations with 1,269 of them belonging to treatment firms. 4. Main results 4.1. Summary statistics 11 In un-tabulated analyses, we show that our results are robust to the sample using the firm-forecast observations. 12 In un-tabulated analyses, we use a constant composition sample requiring the firms exist over the whole sample period. And we find similar results of the treatment effect with our baseline regression models. The drawback of such a sample is the potential concerns for survivorship bias. 18

Table 1 shows the summary statistics of all the variables used in the regression analysis. All the variables are measured on the fiscal year basis and detailed definitions are presented in Appendix A. Panel A presents the descriptive statistics for the SGI sample. The mean value of Forecast frequency shows that in general, a firm in our sample issues about one earnings forecast per year. Forecast horizon has a mean value of 99 days, suggesting that management forecasts on average are issued about 3 months before the corresponding fiscal period end. This statistic is comparable to prior findings: it is 120 days in Ball et al. (2012). Forecast precision has an average score of 2.790 (i.e. between an open-ended estimate and a range estimate) with a standard deviation of 0.773, whereas in Ball et al. (2012) the average value of this variable is around 2.89 with a standard deviation of 0.64. The mean value of Forecast specificity is -0.221%, which implies that for a $50-stock company, the average distance between the top and bottom of all the range EPS forecasts is around $0.111 (0.221%X50). The mean value of Absolute earnings surprise is 0.428%, suggesting that for a $50-stock company, the absolute difference between the actual earnings and analysts forecasts is about $0.214. Absolute abnormal return has a mean and median value of 0.371% and 0.150%, respectively. The rest of the variables show that the average firm in our sample has the natural logarithm of market capital of 6.497 (or $0.663 billion), its leverage ratio is 40.6%, its ROA is 10.4%, the natural logarithm of one plus its analyst coverage is 1.661 (or it is followed by 5 analysts), the standard deviation of its daily stock returns over the prior year is about 0.025, its book-to-market ratio is 0.609, and its institutional ownership is 55.8%. Panel B presents the industry distributions of the SGI sample, high-tech vs. non high-tech, and internet vs. non-internet. We define the high-tech industries as the Electrical Equipment, Computers, and Drugs industries in the Fama-French 48 Industry Classification. We collect a list 19

of internet firms from Hand (2000). Hand (2000) identifies internet firms as those with more than 51% of the revenues coming from the internet. Consistent with our earlier discussion, the proportions of high tech firms and internet firms are higher for treatment firms than for control firms. In later robustness tests in the online appendix, we remove the high-tech firm and internet firms to see whether the causal impact that we document in the paper is driven by firms in these industries. Panel C reports the mean value for all variables separately for treatment and control firms in 1998, i.e. one year before the SGI ruling. Except Forecast specificity, there is no significant difference between the two groups of firms in direct and indirect voluntary disclosure measures. Treatment firms are on average smaller, their leverage ratio is lower, they are less profitable, they are followed by fewer analysts and their book-to-market ratio is lower. There are no significant differences in stock return volatility or the institutional ownership between the two groups of firms. Nevertheless, we control for all these firm characteristics in our regressions. Panel D presents the summary statistics for the Tellabs sample. The mean value of Forecast frequency shows that in general, a firm in the Tellabs sample issues about two earnings forecast per year. Forecast horizon has a mean value of 143 days, suggesting that management forecasts on average are issued about 4.8 months before the corresponding fiscal period end. Forecast precision has an average score of 2.913 (i.e. between an open-ended estimate and a range estimate) with a standard deviation of 0.807. The mean value of Forecast specificity is -0.192%, which implies that for a $50-stock company, the average distance between the top and bottom of all the range EPS forecasts is around $0.10 (0.192%X50). The mean value of Absolute earnings surprise is 0.507%, suggesting that for a $50-stock company, the absolute difference between the actual earnings and analysts forecasts is about $0.25. Absolute abnormal return has 20

a mean and median value of 0.415% and 0.209%, respectively. The average firm in the sample has the natural logarithm of market capital of 6.635 (or $0.761 billion), its leverage ratio is 45.2%, its ROA is 11.3%, the natural logarithm of one plus its analyst coverage is 1.821 (or it is followed by 6 analysts), the standard deviation of its daily stock returns over the prior year is about 0.031, its book-to-market ratio is 0.584, and its institutional ownership is 73.5%. Panel E reports the industry distribution of the sample, high-tech vs. non high-tech, and internet vs. non-internet. Similar to Panel B, treatment firms have a higher proportion of high-tech firms and internet firms than control firms. Panel F shows the comparison for all variables between the treatment and control firms in 2006, i.e., one year before the Tellabs ruling. Our results are largely similar to what s reported in Panel C. Except Forecast specificity, there is no significant difference between the two groups of firms in voluntary disclosure measures. Treatment firms are on average smaller, their leverage ratio is lower, they are less profitable, and they are followed by fewer analysts. 4.2. Regression results We estimate the model specified in Eq. (1) and report our results in Table 2. Panel A and B present the regression results for the SGI sample and Tellabs sample, respectively. As a recap, our direct voluntary disclosure measures include: Forecast frequency, Forecast horizon, Forecast precision, and Forecast specificity. The variables are coded so that a larger value indicates a more informative disclosure. The coefficient estimate of Treat i Post t indicates the difference in post-ruling change in voluntary disclosure between treatment firms and control firms. Panel A reports the results for the SGI sample. Since we use the firm-year average value for voluntary disclosure measures, all these measures become continuous and we use OLS 21

regressions for our analyses. Our results are reported in Columns (1) to (4). Column (1) shows that, relative to control firms, treatment firms on average disclose 0.662 fewer forecast after the SGI ruling. Comparing with an average of 1.289 forecasts in our sample, this change represents more than half of the reduction in the disclosure frequency. Column (2) shows that relative to non-treatment firms, treatment firms on average experience a decrease of 24 days in the forecast horizon after the SGI ruling. Compared to an average of 99 days of forecast horizon in our sample, this change amounts to a 23.8% decrease in forecast horizon. Column (3) reports that, relative to control firms, forecasts issued by treatment firms become less precise after the ruling. Forecast precision drops by 0.26 for treatment firms, which is about 9.4% of the average value of the sample 13. The coefficient on the interaction term in Column (4) is -0.192, indicating that forecast specificity is reduced by 0.192% after the ruling, compared to control firms. This change represents an 86.9% decrease in forecast specificity relative to the sample mean. The control variables generally take on signs which are consistent with prior literature. The coefficients on Size suggest that larger firms tend to disclose more frequently (Rogers and Van Buskirk, 2009). The coefficients on ROA demonstrate that more profitable firms tend to disclose more frequently, and their disclosures have longer forecast horizon and are more precise and specific (Vashishtha, 2014). The coefficients on Log(Analyst) show that firms followed by more analysts tend to disclose more frequently, and their disclosures are more precise (Baginski 13 The PSLRA of 1995 was enacted to provide corporations a safe harbor from frivolous lawsuits. It provides that a company is not liable with respect to any forward-looking statement if (1) the forward-looking statement is identified as forward-looking and is accompanied by meaningful cautionary statements identifying important factors that could cause actual results to differ materially from those in the forward-looking statement or is immaterial; or (2) the plaintiff fails to prove that the forward-looking statement was made with actual knowledge that the statement was false or misleading. Under this situation, a company will be more likely to issue range forecasts rather than point ones, because range forecasts could be defended as more cautionary statements, and help the firm avoid the finding of deliberate recklessness. 22

and Hassell, 1997; Lang and Lundholm, 1996). If a firm has more volatile stock returns (i.e., when Volatility is higher), it tends to disclose less frequently (Li, 2010; Ali et al., 2014; Balakrishnan et al., 2014). The coefficients on BTM indicate that high growth firms tend to issue management forecasts more frequently. This finding is also available in Rogers and Van Buskirk (2009). The coefficients on IO indicate that firms with higher institutional ownership tend to disclosure more precisely and their disclosures are more specific. To take into account other channels to disclose information, we use two indirect measures of voluntary disclosure: Absolute earnings surprise and Absolute abnormal return. As we discussed earlier, higher values of these two measures indicate less informative disclosures. Our regression results are reported in Columns (5) and (6). We find that after the SGI ruling, the absolute value of the earnings surprises (absolute abnormal returns upon quarterly earnings announcement) increases by approximately 0.046% (0.056%) for treatment firms, relative to control firms, representing a 10.7% (15.1%) increase relative to the average absolute earnings surprises (absolute abnormal returns) in our sample. In addition, larger firms, more profitable firms, firms with more analysts following and firms with high growth opportunities tend to have smaller earnings surprises and smaller announcement returns, consistent with them having more voluntary disclosure than other firms. Panel B reports results for the Tellabs sample. Since the Tellabs ruling significantly increases the litigation risk for firms in the Ninth Circuit, if high litigation risk encourages voluntary disclosure, we expect that the Tellabs ruling increases treatment firms voluntary disclosure, relative to control firms. Column (1) shows that, relative to control firms, treatment firms on average disclose 0.483 more forecast after the Tellabs ruling, which represents 27.6% of the sample mean of 23

disclosure frequency. Column (2) to (4) show that treatment firms experience increases in forecast horizon, forecast precision, and forecast specificity. In terms of economic magnitude, these increases represent 12.5%, 3.2%, and 50% of the sample mean, respectively. Columns (5) and (6) report the results related to indirect measures of voluntary disclosures. We find that absolute earnings surprises decreases by 2.0% for treatment firms relative to control firms, which is about 3.9% of the sample mean value. The result based on the absolute abnormal return is however insignificant. In sum, Table 2 shows that the SGI ruling (Tellabs) ruling reduces (increases) both the quantity and quality of treatment firms voluntary disclosure, relative to control firms. These effects are economically meaningful and statistically significant. Collectively, our results from both experiments offer strong evidence that litigation risk causally increases voluntary disclosure. 4.3. Propensity Score Matching While the above section suggests that the parallel trend assumption is likely to be satisfied, our difference-in-differences results may still be explained by significant differences between treatment firms (i.e., firms located in the Ninth Circuit) and control firms before the rulings. To alleviate this concern, we use the propensity score matching method to form the control sample. Specifically, we first estimate a probit model on all firms in 1998 for the SGI sample, and in 2006 for the Tellabs sample. The dependent variable equals one if the firm is in the Ninth Circuit, and zero otherwise. The independent variables include industry fixed effects and the set of firm characteristics used in the baseline regression in Table 2. Then we match each treatment firm to a control firm with the closest estimated propensity score. This process 24

produces 306 unique treatment-control pairs for the SGI sample and 251 unique pairs for the Tellabs sample. Panel A of Table 3 examines the matching between treatment firms and control firms. It shows that there are in general no significant differences between the treatment and matched control groups for the firm characteristics in 1998 before the ruling (except Leverage). In Panel B we report the average changes in the six disclosure measures for the treatment firms (Treatment_Dif), control firms (Control_Dif) and the differences between the two (Dif-in-Dif). Our results are consistent with the conclusion that firms reduce both the quantity and quality of their voluntary disclosure as a result of the SGI ruling. We then re-estimate Eq. (1) using the matched treatment-control firms. Our results are presented in Panel C of Table 3, and they are qualitatively similar to our baseline results. Panel D to F report the results for the Tellabs sample. Panel D shows that there are in general no significant differences between the treatment and matched control groups for the firm characteristics in 2006, the year before the Tellabs ruling (except Log(Analyst)). Panel E shows that compared with control firms, treatment firms issue forecasts more frequently, their forecasts have longer horizon, and their forecasts are more precise and specific. Panel F reports the results when we use the regression approach, and our results are similar to our baseline results. 4.4. The importance of ex ante litigation risk Our existing results suggest that firms in the Ninth Circuit reduce their voluntary disclosure, after the SGI ruling. We attribute this finding to the fact that the SGI ruling significantly reduces the litigation risk of firms located in the Ninth Circuit. If our claim is true, we predict that the treatment effect is more pronounced for firms with higher ex ante litigation 25

risk, because the benefits brought by the ruling are more substantial for them. This prediction is however unavailable, if unobservable state or geographic economic factors explain our finding. As for the Tellabs ruling, we find that firms in the Ninth Circuit increase their voluntary disclosure after the ruling, and we attribute the finding to the increased litigation risks after the ruling. If our claim is true, we predict that the effect of the ruling should be more pronounced for firms with lower ex ante litigation risk. This subsection empirically tests our predictions. Specifically, we use the model in Kim and Skinner (2012) to estimate firms ex ante litigation risk in the year before the rulings. If a treatment firm s litigation risk level in the year prior to the SGI or Tellabs ruling is above (below) the median, the treatment firm is identified as a high (low) risk firm. We repeat our prior analyses separately for the subsamples of treatment firms with high and low risk and report our results in Table 4. Table 4 has six panels, corresponding to the six voluntary disclosure measures. In each panel, the first (second) row reports the coefficient on Treat i Post t and its related t-statistics from the regression based on the subsample with high (low) risk. The third (fourth) row reports the equality of the estimated coefficients in the two subsamples respectively using the Wald tests. Column (1) and (2) present the results for the SGI ruling and Tellabs ruling, respectively. Our results in Column (1) indicate that relative to control firms, high litigation risk treatment firms experience a significant decline in voluntary disclosure while low litigation risk treatment firms experience no significant change after the SGI ruling. The difference between the two types of treatment firms is in general statistically and economically significant for most measures of voluntary disclosure (except Forecast precision and Absolute Earnings Surprise). This finding is consistent with our hypothesis that the effect of the SGI ruling concentrates on firms with high ex ante litigation risk. Our results in Column (2) indicate that relative to control 26

firms, low litigation risk treatment firms experience a significant increase in voluntary disclosure while high litigation risk treatment firms experience no significant change, after the Tellabs ruling. The difference between the two types of treatment firms is in general statistically and economically significant for all direct measures of voluntary disclosure, but not the two indirect measures. This finding is consistent with our prediction that the effect of the Tellabs ruling concentrates on firms with low ex ante litigation risk. In sum, our analyses show that the treatment effect we document is more pronounced for firms with high (low) ex ante litigation risk in the SGI (Tellabs) ruling, which is consistent with our claim that our results are due to the exogenous shock in litigation risk brought by the two rulings. 4.5. Test of H2 This subsection empirically tests H2. To do the test, we sort treatment firms into two groups according to the nature of the news, which is based on the sign of either the average quarterly earnings surprises or the average quarterly earnings announcement returns. The quarterly earnings surprise is the actual earning minus the most recent analyst consensus estimate. If a treatment firm s average quarterly earnings surprise is positive or equal to zero, the treatment firm is deemed a firm with good news; otherwise, it is deemed a firm with bad news. The average quarterly earnings announcement return is the average of the size-adjusted returns for the three-day window centered on the quarterly earnings announcement date. If the average quarterly earnings announcement return in a year is positive or equal to zero, the treatment firm is deemed a firm with good news; otherwise, it is deemed a firm with bad news. 27

We estimate the model as specified in Table 2 and report the coefficient on Treat i Post t and its related t-statistics for the regression separately for treatment firms with good and bad news in Table 5. We also report the equality of the estimated coefficients in the two subsamples respectively using the Wald tests. Column (1) and (2) present the results for the SGI ruling, while column (3) and (4) present the results for the Tellabs ruling. We follow Houston et al. (2015) and infer the timeliness of the forecast through Forecast Horizon, defined as the end date of the fiscal period being forecasted minus the date when the forecast is released. A higher value of Forecast Horizon indicates that managers provide an earlier and timelier forecast to the market. Given prior evidence that firms preempt bad news through voluntary disclosure (Skinner 1994; Kasznik and Lev, 1995; Field et al., 2005; Billing and Cedergren, 2015), we expect that the impact of litigation risk on timeliness of forecasts is more pronounced for firms with bad news. Column (1) reports results where we define news according to the sign of average quarterly earnings surprises. It shows that treatment firms in the SGI ruling with bad news experience a statistically significant drop in voluntary disclosures while treatment firms with good news experience no significant change in voluntary disclosure, for all voluntary disclosure measures except Forecast horizon. The Wald tests indicate that the treatment effect is significantly more pronounced for firms with bad news. When the dependent variable is Forecast horizon, the coefficient on Treat*Post (good) is about -14, suggesting that for firms with good news, compared to pre-sgi period, the management s forecast is issued on average 14 days later in the post-sgi period; the coefficient on Treat*Post (bad) is about -26, suggesting that, for firms with bad news, the management s forecast is issued on average 26 days later in the post-sgi period. The difference between the coefficients is statistically significant, indicating that the 28

delay in issuing management forecasts is more severe for bad news firms, when the litigation risk is reduced by the SGI ruling. This result is consistent with the notion that litigation risk has a greater impact on timeliness of disclosure. Column (2) reports results where we define news according to the sign of the average quarterly earnings announcement returns. Our results are consistent with those in Column (1) and indicate that the treatment effect is more pronounced for firms with bad news. Column (3) reports results for the Tellabs ruling and we define bad news according to the sign of the average quarterly earnings surprises. We find that the effect of the ruling is significantly more pronounced for all measures of voluntary disclosures, except Absolute Abnormal Return. When the dependent variable is Forecast Horizon, the coefficient on Treat*Post (good) is about 2, not significant, while the coefficient on Treat*Post (bad) is about 13, significant at the 1% level. The difference between the two coefficients is significant at the 5% level. This result is consistent with the notion that the effect of litigation risk on timeliness of management forecasts is more pronounced for bad news firms. Column (4) reports results that are similar to Column (2). In sum, the results in Table 5 provide strong evidence that the treatment effect is more pronounced for firms with bad news. This finding is difficult to explain if the treatment effect is due to reasons unrelated to litigation risks. 4.6. Test of H3 This subsection empirically tests H3. If H3 is true, we expect that firms will be more likely to issue bad news forecasts, and bad news forecasts will be timelier, when litigation risk increases. We therefore examine changes in the frequency and horizon of bad (good) news 29

forecasts after the SGI and Tellabs rulings. The nature of the news carried by the forecasts is based on the CIG news code, following Anilowski, Feng, and Skinner (2007). Frequency of bad (good) news forecasts is the number of bad (good) news forecasts disclosed by a firm in a fiscal year. Horizon of bad (good) news forecasts is defined as the average forecast horizon of bad (good) news forecasts in the fiscal year. We estimate the model as specified in Table 2 and report the coefficients on Treat i Post t and their related t-statistics in Table 6. Panel A reports the frequency results while Panel B reports the results of forecast horizons. Within each panel, Column (1) and (2) respectively present the results for bad news forecasts and good news forecasts for the SGI ruling, while Column (3) and (4) report the results for bad news forecasts and good news forecasts for the Tellabs ruling. In Panel A, Column (1) reports that the coefficient on Treat i Post t is about -0.399, suggesting that relative to control firms, treatment firms on average disclose 0.399 fewer bad news forecasts after the SGI ruling; the coefficient on Treat i Post t is about -0.189 in Column 2, suggesting that relative to control firms, treatment firms on average disclose 0.189 fewer good news forecast after the SGI ruling. Comparing the two coefficients shows that when the litigation risk is reduced by the SGI ruling, firms reduce the frequency of bad news forecasts to a greater extent. The coefficient on Treat i Post t is about 0.276 in Column (3), significant at the 5% level, while in Column (4) the coefficient on Treat i Post t is about 0.115, not significant. The difference between the two coefficients suggests that when the Tellabs ruling increases treatment firms litigation risk, firms issue more bad news forecasts than good news forecasts. In Panel B, Column (1) reports that the coefficient on Treat i Post t is about -20, 30

suggesting that after the SGI ruling, relative to control firms, treatment firms delay their bad news forecasts by 20 days ; the coefficient on Treat i Post t in Column (2) is about -9, suggesting that the delay for good news forecasts is only 9 days. A comparison between the two coefficients indicates that bad news forecasts become less timely, when the litigation risk is reduced by the SGI ruling. The coefficient on Treat i Post t in Column (3) is about 11, significant at the 5% level, while it is about 3, not significant, in Column (4). The difference between the two coefficients suggests that when the Tellabs ruling increases litigation risk, bad news forecasts become timelier than good news forecasts. In sum, our results are consistent with H3 and suggest that litigation risk increases the likelihood of firms preempting bad news through management forecasts. 5. Conclusion How does litigation risk affect voluntary disclosure is a matter of considerable debate in the accounting literature. On one hand, litigation risk may increase disclosures, because disclosures help to fend against the claim that the firm withholds information from investors and they help to reduce stock price crashes, which are often the trigger of investor lawsuits. On the other hand, litigation risk may deter disclosures, because forward-looking disclosures, unbiased at the time of issuance, may invite lawsuits if they are proven untrue ex post. Dating back to Skinner (1994) and Francis et al. (1994), several empirical works have provided support for both sides. The long-running academic debate implies more evidence is necessary. To establish the casual relationship between litigation risk and voluntary disclosure, we take advantage of two natural experiments, i.e., the Ninth Circuit Court s ruling Re: Silicon 31

Graphics Inc. Securities Litigation in 1999 and the Supreme Court s ruling Tellabs Inc, v. Makor Issues & Right Ltd. in 2007. Our research employs a difference-in-differences approach. Firms located in the Ninth Circuit are treatment firms and other firms are control firms. We find that after the SGI (Tellabs) ruling, treatment firms reduce (increase) their voluntary disclosure and that the pre-treatment trends are indistinguishable between treatment and control firms. We conduct a battery of robustness checks to ensure that our results are not driven by other covariates. We show that our results continue to hold when we use propensity score matching to select control firms, the treatment effect is stronger for firms with higher out-of-state sales and for firms with bad news. In addition, our findings are not driven by high-tech or internet firms, and the effect of the SGI (Tellabs) ruling is more pronounced for firms with high (low) ex ante litigation risk. We also hypothesize and find that the impact of litigation risk is more pronounced when firms have bad news, and that firms are more likely to preempt bad news through voluntary disclosures when there is an increase in litigation risk. While we can never completely rule out alternative explanations, our results overwhelmingly lend support to our conclusion: litigation risk causally increases voluntary disclosure. We however caution that our results are based on U.S. data and our conclusion may not be applicable to countries with different legal and institutional settings. 32

References Acharya, V., R. Baghai, & K. Subramanian, 2014. Wrongful discharge laws and innovation. Review of Financial Studies, 27, 301-346. Ali, A., Klasa, S., & Yeung, E. (2014). Industry concentration and corporate disclosure policy. Journal of Accounting and Economics, 58(2), 240-264. Agrawal, A. K., & Matsa, D. A. (2013). Labor unemployment risk and corporate financing decisions. Journal of Financial Economics, 108(2), 449-470. Anilowski, C., Feng, M., & Skinner, D. J. (2007). Does earnings guidance affect market returns? The nature and information content of aggregate earnings guidance. Journal of Accounting and Economics, 44, 36-63. Appel, I. R., Gormley, T. A., & Keim, D. B. (2016). Passive investors, not passive owners. Journal of Financial Economics 121, 111 141. Baginski, S. P., Conrad, E. J., & Hassell, J. M. (1993). The effects of management forecast precision on equity pricing and on the assessment of earnings uncertainty. The Accounting Review, 913-927. Baginski, S. P., & Hassell, J. M. (1997). Determinants of management forecast precision. The Accounting Review, 303-312. Baginski, S. P., Hassell, J. M., & Kimbrough, M. D. (2002). The effect of legal environment on voluntary disclosure: Evidence from management earnings forecasts issued in US and Canadian markets. The Accounting Review, 77(1), 25-50. Balakrishnan, K., Billings, M. B., Kelly, B., & Ljungqvist, A. (2014). Shaping liquidity: On the causal effects of voluntary disclosure. The Journal of Finance, 69(5), 2237-2278. Ball, R., Jayaraman, S., & Shivakumar, L. (2012). Audited financial reporting and voluntary disclosure as complements: A test of the confirmation hypothesis. Journal of Accounting and Economics, 53(1), 136-166. Bamber, L. S., & Cheon, Y. S. (1998). Discretionary management earnings forecast disclosures: Antecedents and outcomes associated with forecast venue and forecast specificity choices. Journal of Accounting Research, 167-190. Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates? The Quarterly Journal of Economics, 119(1), 249-275. 33

Billings, M. B., & Cedergren, M. C. (2015). Strategic silence, insider selling and litigation risk. Journal of Accounting and Economics, 59(2-3), 119-142. Boone, A., White, J. (2015). The effect of institutional ownership on firm transparency and information production. Journal of Financial Economics 117, 508 533. Bourveau, T., Lou, Y., & Wang, R. C. (2018). Shareholder litigation and corporate disclosure: evidence from derivative lawsuits. Journal of Accounting Research 56, 797-842. Cao, Z., & Narayanamoorthy, G. S. (2011). The effect of litigation risk on management earnings forecasts. Contemporary Accounting Research, 28(1), 125-173. Cazier, R. A., Christensen, T. E., Merkley, K. J., & Treu, J. S. (2017). Litigation Risk and Non-GAAP Reporting. Working paper. Choi, S. J., & Pritchard, A. C. (2012). The supreme court s impact on securities class actions: an empirical assessment of Tellabs. J Law Econ Organ 28 (4): 850-881. Cox, J. D., Thomas, R. S., & Bai, L. (2009). Do differences in pleading standards cause forum shopping in securities class actions: Doctrinal and empirical analyses. Wis. L. Rev., 421. Crane, A. D., & Koch, A. (2016). Free riding and ownership structure: Evidence from a natural experiment. Management Science, Forthcoming. Dass, N., & Massa, M. (2011). The impact of a strong bank-firm relationship on the borrowing firm. Review of Financial Studies, 24(4), 1204-1260. Donelson, D. C., McInnis, J. M., Mergenthaler, R. D., & Yu, Y. (2012). The timeliness of bad earnings news and litigation risk. The Accounting Review, 87(6), 1967-1991. Field, L., Lowry, M., & Shu, S. (2005). Does disclosure deter or trigger litigation? Journal of Accounting and Economics, 39(3), 487-507. Francis, J., Philbrick, D., & Schipper, K. (1994). Shareholder litigation and corporate disclosures. Journal of Accounting Research, 137-164. Gande, A., & Lewis, C. M. (2009). Shareholder-initiated class action lawsuits: Shareholder wealth effects and industry spillovers. Journal of Financial and Quantitative Analysis, 44(04), 823-850. Gibney, B. C. (2001). The end of the unbearable lightness of pleading: Scienter after Silicon Graphics. UCLA L. Rev., 48, 973-1633. Grundfest, J. A., & Pritchard, A. C. (2002). Statutes with multiple personality disorders: The value of ambiguity in statutory design and interpretation. Stanford Law Review, 627-736. 34

Hand, J. R. (2000). Profits, losses and the non-linear pricing of Internet stocks. SSRN elibrary. Hopkins, J. (2014). Does the Threat of Securities Litigation Deter Earnings Management? SSRN elibrary. Houston, J. F., Lin, C., Liu, S., & Wei, L. (2015). Shareholder litigation and management earnings forecasts. Working paper. Jhong, C. P. 2016. Application of common-law doctrine of forum non conveniens in federal courts after enactment of 28 U.S.C.A. 1404(a) authorizing transfer to another district. 10 A.L.R. Fed. 352. Johnson, M. F., Nelson, K. K., & Pritchard, A. C. (1999). In Re Silicon Graphics Inc.: shareholder wealth effects resulting from the interpretation of the Private Securities Litigation Reform Act's pleading standard. S. Cal. L. Rev., 73, 773. Johnson, M. F., Kasznik, R., & Nelson, K. K. (2001). The impact of securities litigation reform on the disclosure of forward Looking information by high technology firms. Journal of Accounting Research, 39(2), 297-327. Kasznk, R., & Lev, B. To Warn or Not to Warn: Management Disclosures in the Face of an Earnings Surprise. The Accounting Review 70: 113 34. Kim, I., & Skinner, D. J. (2012). Measuring securities litigation risk. Journal of Accounting and Economics, 53(1), 290-310. Lang, M. H., & Lundholm, R. J. (1996). Corporate disclosure policy and analyst behavior. The Accounting Review, 467-492. Levy, H., Shalev, R., & Zur, E. (2016). The effect of CFO personal litigation risk on firms disclosure and accounting choices. Conditionally accepted by Contemporary Accounting Research. Li, X. (2010). The impacts of product market competition on the quantity and quality of voluntary disclosures. Review of Accounting Studies, 15(3), 663-711. Lowry, M. (2009). Discussion of Shareholder litigation and changes in disclosure behavior. Journal of Accounting and Economics, 47(1), 157-159. Naughton, J. P., Rusticus, T. O., Wang, C., & Yeung, I. (2018). Private Litigation Costs and Voluntary Disclosure: Evidence from the Morrison Ruling. Forthcoming, The Accounting Review https://doi.org/10.2308/accr-52203 35

Rajan, R. G. (1992). Insiders and outsiders: The choice between informed and arm s length debt. The Journal of Finance, 47(4), 1367-1400. Rogers, J. L. (2008). Disclosure quality and management trading incentives. Journal of Accounting Research, 46(5), 1265-1296. Rogers, J. L., & Van Buskirk, A. (2009). Shareholder litigation and changes in disclosure behavior. Journal of Accounting and Economics, 47(1), 136-156. Rogers, J. L., & Van Buskirk, A. (2013). Bundled forecasts in empirical accounting research. Journal of Accounting and Economics, 55(1), 43-65. Skinner, D. J. (1994). Why firms voluntarily disclose bad news. Journal of Accounting Research, 38-60. Skinner, D. J. (1997). Earnings disclosures and stockholder lawsuits. Journal of Accounting and Economics, 23(3), 249-282. Usmani, F. Z. (2001). Inequities in the resolution of securities disputes: Individual or class action; arbitration or litigation. Fordham J. Corp & Fin. L. 7, 193. Vashishtha, R. (2014). The role of bank monitoring in borrowers discretionary disclosure: Evidence from covenant violations. Journal of Accounting and Economics, 57(2), 176-195. Zhang, H., & Zheng, L. (2010). The valuation impact of reconciling pro forma earnings to GAAP earnings. Journal of Accounting & Economics, 51(1), 186-202. 36

Figure 1 Geographic boundaries of United States Courts of Appeals and United States District Courts 37