Florida International Conservative Reporting & Securities Class Action Lawsuits Paper

User Generated

Znlxry

Writing

Florida International University

Description

This assignment requires to respond to the following introductory questions:

Introductory section:

What is the paper about?

Why are the issues examined interesting?

What do we know from the prior research on related issues?

Unformatted Attachment Preview

ACCOUNTING HORIZONS Vol. 30, No. 1 March 2016 pp. 93–118 American Accounting Association DOI: 10.2308/acch-51308 Conservative Reporting and Securities Class Action Lawsuits Michael L. Ettredge The University of Kansas Ying (Julie) Huang University of Louisville Weining Zhang Cheung Kong Graduate School of Business SYNOPSIS: Positive accounting theory predicts that conservative financial reporting averts GAAP-based litigation. However, very little empirical evidence addresses whether and how accounting conservatism provides these benefits. Using a sample of lawsuits against public companies for alleged violations of U.S. GAAP, we examine the association of accounting conservatism with subsequent initiation of lawsuits and with four litigation outcomes: market reactions to lawsuits, duration of lawsuits, dismissals of lawsuits, and penalties approved by courts. We find that firms with greater degrees of conditional conservatism experience more favorable consequences on all five dimensions of litigation occurrence and outcomes. Several measures of unconditional conservatism are not associated with the litigation variables. Our study provides new empirical evidence supporting Watts’s (2003a) litigation explanation for accounting conservatism. Keywords: accounting conservatism; litigation; lawsuit dismissal; lawsuit settlement. JEL Classifications: M41. INTRODUCTION A ccounting-related class action lawsuits have increased in frequency during recent decades. Every year such lawsuits impose billions of dollars of direct and indirect costs on public companies and, eventually, on their shareholders (Cornerstone Research 2013). Although positive accounting theory asserts that litigation risk engenders accounting conservatism as a preemptive defense (Watts 2003a), there is little empirical evidence on whether and how this defense works.1 Accounting conservatism refers to ‘‘the accountant’s tendency to require a higher degree of verification to recognize good news as gains than to recognize bad news as losses’’ (Basu 1997, 7). In this paper, we provide evidence that conditional accounting conservatism (but not unconditional conservatism) averts the ex ante occurrence and improves the ex post outcomes of GAAPrelated litigation. Accounting-related class action litigation frequently involves plaintiffs attempting to show that companies’ managers reported overstated earnings numbers intentionally or with reckless negligence, and that subsequent revelation of the improper accounting caused stock price decreases that harmed plaintiff shareholders (Chalmers, Naiker, and Navissi 2012). Conservative We are grateful for helpful comments provided on prior versions of this paper by Sudipta Basu, Dain Donelson, Guojin Gong, Karla Johnstone, Teri Yohn, Frank Zhang, and seminar participants at the University of Pittsburgh, the 2015 Canadian Academic Accounting Association Annual Conference, and the 2013 American Accounting Association Annual Meeting. We also thank Paul A. Griffin (editor) and the reviewer. All remaining errors are ours. Editor’s note: Accepted by Paul A. Griffin. Submitted: February 2015 Accepted: October 2015 Published Online: October 2015 1 A number of studies document that higher ex ante litigation risk motivates more preemptive conservative reporting, but this is not the same as evidence that conservatism actually reduces ex post litigation incidence and costs. See Basu (1997), Holthausen and Watts (2001), Ryan and Zarowin (2003), and Seetharaman, Srinidhi, and Swanson (2005). 93 94 Ettredge, Huang, and Zhang financial reporting arguably reduces the expected value of plaintiffs’ contemplated or actual investments in lawsuits by averting or mitigating these elements of alleged harm: overstated earnings and stock price drops. The asymmetric recognition of losses versus gains can lead to systematic understatements of earnings and net assets in audited financial statements (Watts 2003a).2 Courts are more likely to rule that shareholder harm occurs when earnings and net assets have been overstated rather than understated (Kellogg 1984; St. Pierre and Anderson 1984). Conservative reporting also can reduce the expected value of plaintiffs’ investments in litigation by decreasing the likelihood that plaintiff shareholders can prove they suffered financial harm through stock price drops. Timely recognition of bad news via accounting conservatism prevents managers from deferring the revelation of bad news for extended periods. As a result, conservative financial reporting reduces the likelihood of temporarily overpriced stock and subsequent sharp stock price ‘‘crashes’’ (Kim and Zhang 2013).3 Pursuing litigation against public companies in the current legal climate is a risky investment for plaintiffs and plaintiffs’ attorneys. It is costly and the monetary outcomes are uncertain. The expected value of litigation decreases when defendant companies report conservative earnings.4 Consequently, conservative reporting reduces the probability that plaintiffs and their lawyers will prevail should their case go to trial. Proceeding to trial is the plaintiffs’ ultimate bargaining chip. Few cases ever go to trial due to the high costs involved, but reductions in the expected value of the plaintiffs’ case should affect the consequences of litigation that more commonly are observed. This study investigates the implications of this view. If reporting more conservatively reduces the expected value of litigation, then shareholders of more conservative firms are less likely to initiate lawsuits. The initial market reaction to a litigation event should be less negative for a more conservative firm because the market may expect that the lawsuit is more likely to be dismissed. The probability of dismissal is higher because the plaintiffs’ case is likely to be weaker. If a case is not dismissed, then the amount of any court-approved penalty or settlement should be less because the plaintiffs are less likely to prevail at trial. Finally, the duration of the litigation process should be shorter because litigation is costly to plaintiffs and their lawyers, and they will invest less if the expected value of the outcomes is lower. We empirically test these implications using a sample of 363 shareholder lawsuits alleging violations of GAAP by U.S. firms filed in 1996 through 2011 and resolved by mid-2014. Similar to Zhang (2008), our proxies for conditional conservatism consist of (1) Basu’s (1997) measure of the sensitivity of earnings to bad news relative to the sensitivity of earnings to good news, (2) negative non-operating accruals (Givoly and Hayn 2000), (3) the skewness of earnings (Basu 1995; Givoly and Hayn 2000), and (4) a composite measure based on the average rank of the three conditional conservatism measures. We capture unconditional conservatism using (5) Penman and Zhang’s (2002) conservatism index, and (6) a market-to-book equity ratio (Roychowdhury and Watts 2007). To test whether more conservative firms are less likely to be sued, we use a logistic model that regresses a dichotomous variable representing initiation or non-initiation of a shareholder lawsuit on each of the four conditional conservatism measures individually, plus both of the unconditional conservatism measures, and several control variables. Control firms are matched one-to-one with test firms based on the ex ante probability of being sued for GAAP violations. Consistent with expectations, we find that all four measures of conditional conservatism are negatively associated with the occurrence of future lawsuits. Firms in the top percentile of the average conservatism ranking face a 30 percent lower probability of a lawsuit than those in the bottom percentile. Neither of the unconditional conservatism measures is associated with subsequent litigation occurrence, suggesting unconditional conservatism as measured does not play a role in the litigation process. To test whether more conservative firms experience more favorable outcomes, given litigation is initiated, we investigate the four outcomes of the litigation process previously mentioned. We use ordinary least squares (OLS) models to explain the market reaction to litigation initiation, duration of the litigation process, and litigation settlement amount dependent variables, and we use a logistic model to explain courts’ dismissals or non-dismissals of lawsuits. Consistent with expectations, we find that each of the four conditional conservatism measures is significantly associated with more favorable consequences for defendants, and the effects are economically meaningful. The unconditional conservatism metrics generally are statistically insignificant. Specifically, first, the stock market reacts less negatively to lawsuits initiated against more conditionally conservative accounting reporters. Firms in the top percentile of the average ranking of conservatism experience market returns at lawsuit announcements that are less negative by around 930 basis points, 2 3 4 Whether conservatism does so depends on how much bias managers introduce into the measurement system ex ante, for instance by being unconditionally aggressive. See, for example, the discussion of Corollary 2 in Gao (2013, 262). Furthermore, audited financial reports arguably discipline other management disclosures, such as earnings guidance, by providing benchmarks against which managers can be held accountable ex post (Ball, Jayaraman, and Shivakumar 2012, 137). LaFond and Watts (2008) argue that this ‘‘benchmarking’’ role of financial reports is particularly effective when the financial reports are conservative. As a result, conservative firms’ stock prices should be less likely to be poised for crashes, because both ‘‘hard’’ information (audited financial reports) and ‘‘soft’’ information (such as managers’ forecasts) are less likely to overstate past and expected future gains, and to understate past and potential losses. For example, Ettredge, Huang, and Zhang (2012) find that conservative reporters are less likely to have to restate prior financial reports, suggesting that they are less likely to violate GAAP. Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 95 compared to firms in the bottom percentile. Second, cases are resolved more swiftly for more conservative firms. Firms in the top percentile based on the average ranking of conservatism resolve their cases sooner by an average of 385 days than firms in the bottom percentile, which is equivalent to a 36 percent reduction in the number of days for the average firm in our sample. Furthermore, courts are more likely to dismiss lawsuits against conservative defendants: the probability of a dismissal is higher by 31 percent for the top percentile firms than the bottom percentile. Finally, lawsuits that are not dismissed result in smaller court-approved settlement amounts for conservative reporters. The top percentile firms pay a penalty amount that is, on average, $145 million less than firms in the bottom percentile. Overall, these results suggest substantial economic benefits to large differences in conditional conservatism. Our results are robust to controlling for various known firm-specific determinants of litigation risk, as well as to several sensitivity tests. This paper’s main contributions are as follow. First, we contribute to the literature documenting benefits of conservative reporting. Prior studies have examined various country-level and firm-specific factors that explain the contracting demand for conservatism described by Watts (2003a). However, less research has been devoted to the economic consequences of or benefits from conservative accounting in the litigation context. Second, we contribute to the literature on the relation between accounting properties and litigation. Studies such as Kim and Skinner (2012) that explain initiation of lawsuits have not previously investigated the role of conservative reporting. Prior studies of accounting litigation tend not to investigate any event other than the initiation of lawsuits (St. Pierre and Anderson 1984; Kellogg 1984; Palmrose and Scholz 2004; Chalmers et al. 2012). In contrast, we investigate four other outcomes of lawsuits: market reaction to litigation, dismissals of lawsuits, courtapproved penalties, and duration of litigation. Our results suggest that conditional conservatism provides reporting firms with protection against GAAP-based litigation, but unconditional conservatism does not. Finally, our results contribute to a literature that raises questions regarding the FASB’s criticism of accounting conservatism (FASAC 2005; FASB 2010). Conservatism (prudence) has been eliminated from the FASB and IASB’s joint conceptual framework (FASB 2010). However, recent studies arguably support the value of conservatism in reducing information asymmetry and aligning interests between managers and suppliers of capital.5 This study provides new evidence on a different and important benefit arising from conservative financial reporting: reductions in lawsuit occurrences and mitigations of adverse lawsuit outcomes. The remainder of the paper is organized as follows. The second section briefly reviews the relevant literature and develops our hypotheses. The third section describes our sample, empirical model, and variables. The fourth section presents descriptive statistics and the results of multivariate regressions. The fifth section presents our conclusions. BACKGROUND, RELATED LITERATURE, AND HYPOTHESES Background Shareholder litigation against public companies is a persistent feature of the corporate environment and is costly both to plaintiffs and defendants. Firms likely experience negative abnormal stock returns when litigation against them is announced. In addition to the amounts of any settlements, firms targeted by class actions also incur the legal costs of defending themselves, their officers, and directors. These costs include not only payments made to in-house and external defense attorneys, but also the costs of directors and officers insurance, and the opportunity costs of devoting attention to litigation rather than to operating, investing, and financing decisions. When lawsuits are not dismissed, firms bear the costs of court-approved awards or of private settlement costs. On rare occasions claims are litigated in court, adding yet more costs. Litigation is costly not only to the firms that are defendants, but also to plaintiffs and their legal counselors. Significant numbers of lawsuits continue to be filed since passage of the Private Securities Litigation Reform Act (PSLRA) of 1995 increased evidentiary standards in order to deter frivolous lawsuits (Johnson, Nelson, and Pritchard 2007; Chalmers et al. 2012). A study of class action litigation by PricewaterhouseCoopers (2008) reports that 2,159 securities class action lawsuits were initiated during 1996–2007, an average of 180 lawsuits per year. Cornerstone Research (2013) reports that from 1997 through 2012, 191 class action filings were initiated each year on average. Firms subjected to class actions lost $652 billion in market capitalization per year on average, computed over the firms’ class action periods and totaled for all firms sued in a year. Accounting violations are among the factors that courts consider meritorious under PSLRA (Johnson et al. 2007). Accountingrelated class action lawsuits constituted more than 50 percent of the total securities lawsuits in each year during 1996–2007 (PricewaterhouseCoopers 2008). Cornerstone Research (2013) reports that in recent years, securities lawsuits usually have alleged misrepresentations in financial documents (95 percent and 97 percent in 2012 and 2013, respectively) and often have alleged violations of GAAP (23 percent and 24 percent in 2012 and 2013, respectively). We turn now to a discussion of related research. 5 See LaFond and Watts (2008), LaFond and Roychowdhury (2008), Göx and Wagenhofer (2009), Francis and Martin (2010), Kim, Li, Pan, and Zuo (2013), Watts and Zuo (2011), Gao (2013), and Balakrishnan, Watts, and Zuo (2014). Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 96 Related Literature Consistent with the litigation explanation for conservatism (Watts 2003a), prior research documents that changes in legal liability over time appear to motivate changes in conservatism over time (Basu 1997; Ball, Kothari, and Robin 2000; Holthausen and Watts 2001; Ryan and Zarowin 2003; Sivakumar and Waymire 2003; Raonic, McLeay, and Asimakopoulos 2004; Seetharaman et al. 2005). Cross-sectional differences in litigation risk across legal jurisdictions also are associated with cross-sectional differences in conservatism (Huijgen and Lubberink 2005; Lobo and Zhou 2006; Bushman and Piotroski 2006; Qiang 2007; Khan and Watts 2009; Shroff, Venkataraman, and Zhang 2013). For example, Qiang (2007) studies whether the four explanations for accounting conservatism—contracting, litigation, regulation, and taxation—motivate greater levels of conditional and unconditional conservatism. In the litigation context, Qiang (2007) finds that conditional and unconditional conservatism are both increased by proxies for higher litigation risk.6 Although these studies suggest that greater litigation risk and increases in litigation risk motivate higher levels of conservatism, to our knowledge, no published research establishes that conservative reporting of income reduces actual litigation occurrence and costs. Two early studies (St. Pierre and Anderson 1984; Kellogg 1984) report results suggesting that overstatements of net assets and income are more likely to generate litigation than are understatements. Those results can be interpreted as consistent with conservatism averting litigation occurrence. However, the two studies are based on relatively small samples, and they predate both current litigation rules (under the PSLRA of 1995) and contemporary research into conservative reporting, i.e., Basu (1997) and subsequent studies. Thus those early studies do not employ currently accepted measures of reporting conservatism in the contemporary litigation environment. Nor do those studies investigate litigation outcomes other than occurrence. Researchers have identified two broad types of conservative reporting that could affect litigation differently: unconditional and conditional conservatism. A key aspect of the distinction between the two relates to the role of information in recognizing gains and losses. Unconditional conservatism only utilizes information known at the inception of an asset’s life. Conditional conservatism utilizes information when it is received in future periods. Although both types of conservatism eventually reduce income and equity, only conditional conservatism reveals new information and enables contracting responses (Basu 2005).7 Thus it is possible that conditional conservatism and unconditional conservatism affect litigation against reporting firms differently. Lack of conditional conservatism might increase litigation occurrence and costs because such lack potentially ties to the legal concept of scienter or wrongful intent. Plaintiffs can argue that managers employing discretion to overstate earnings in violation of GAAP did so with intent to deceive. In contrast, it would be more difficult to obtain a judgment against a firm whose managers’ report unconservatively in compliance with GAAP. Therefore, we investigate both unconditional and conditional types of conservatism in our litigation setting.8 One unpublished manuscript, Blunck (2009), is somewhat similar to our study in that it investigates the association between a measure of conditional conservatism and several of the litigation occurrence and outcome measures that we study. The marginal contribution of our study over Blunck (2009) arguably lies in our better research designs, which provide stronger results supporting our conclusions. First, the results of our study are more solid and robust than Blunck (2009), as our study employs four measures of conditional conservatism and two measures of unconditional conservatism. In contrast, Blunck (2009) uses only Basu’s (1997) asymmetric timeliness (AT) measure. Second, due to Blunck’s (2009) research design, some of his main hypotheses are not supported by his empirical results, while his statistical support for others is weak. Overall, Blunck’s (2009) four main hypotheses are not strongly supported by his results. The published study most closely related to ours is Chalmers, Naiker, and Navissi (2012; hereafter, CNN). CNN examine the role of earnings quality in the Securities and Exchange Commission’s (SEC) Rule 10b-5 class action lawsuits in both the pre- and post-PSLRA eras. Their proxy for earnings quality consists of a version of the Dechow and Dichev (2002) accruals metric. CNN employ a sample of 359 lawsuits against U.S. corporations alleging misrepresentations of financial performance or condition, and find evidence of positive changes in working capital accruals (and in cumulative working capital accruals) among test firms in the four quarters preceding the onset of litigation. This is followed by a sharp decrease in accruals of test 6 7 8 There is no overlap between our study and Qiang (2007). Qiang (2007) does not study whether conservatism averts litigation (a negative association). In contrast, similar to most prior literature on conservatism and litigation, she studies whether litigation risk motivates higher conservatism (a positive association). While our study focuses on actual litigation occurrence and outcomes, Qiang (2007) does not employ any actual instances of litigation. Conditional conservatism is not limited to conservative recognition of gains and losses in income; it also extends to disclosure of contingent liabilities (losses) but not contingent assets (gains) in the footnotes. In measuring conservatism we do not employ plaintiffs’ specific claims about the timing and amounts of earnings errors in each lawsuit filed. Such information does not exist for control firms that were not sued. Instead, we measure earnings conservatism in the years prior to the initiation (or noninitiation) of litigation. This information is available for all sample firms. Similarly, we do not use information in plaintiff claims about specific stock price decreases attending revelation of GAAP violations. Instead, we control for several characteristics of firms’ stock volume and return behavior in the year prior to the initiation (or non-initiation) of litigation. Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 97 firms over the subsequent four quarters. They observe these results in both the pre- and post-PSLRA eras. CNN (2012, 25) conclude, ‘‘These results suggest that lower earnings quality is a driver of securities class action lawsuits in both [the pre- and post-PSLRA] periods.’’9 We note that CNN investigate only one of the five litigation occurrence and cost measures that we employ, and their test variable consists of a proxy for discretionary accruals rather than reporting conservatism. Accounting conservatism is different from earnings management via discretionary accruals. Aside from the differences in measurement of the respective proxy variables, the goals of conservatism are different from those of earnings management, and this leads to differences in the extent and circumstances under which accruals are used. An extensive literature documents that managers use discretionary accruals to achieve various short-term earnings goals.10 However, conservative reporting of earnings, although it might involve discretionary accruals, is thought to constitute a response to more permanent incentives. For example, firms that report conservatively arguably desire to facilitate contracting and avert litigation.11 If so, then they will pursue a policy of asymmetric income recognition on a consistent basis, resulting in predictable cumulative effects such as the understatement of net assets. As Watts (2003b, 296–297) states, ‘‘Earnings management cannot explain important parts of the conservatism evidence and is not a plausible general explanation of the financial reporting evidence consistent with conservatism over long periods of time’’ (emphasis in the original). We add a variable controlling for discretionary accruals to our models to determine whether conservatism metrics have additional explanatory power beyond discretionary accruals. Hypotheses Our concept of the association between accounting conservatism and the various types of litigation initiation and outcomes is that in the current legal climate, plaintiffs and their lawyers make risky investments when they initiate and pursue securities litigation. Litigation is costly, and the willingness of plaintiffs and their lawyers to bear such costs increases with the expected value of the outcome. The value of a litigation investment hinges on the likelihood that the claims are meritorious.12 In the postPSLRA era the expected value to plaintiffs of investing in weak (less meritorious) cases arguably has decreased. Given that the litigation events we study involve alleged violations of GAAP, and that shareholders must show they suffered harm, claims are more likely to be meritorious if defendant firms can be shown to have overstated reported earnings and net assets. Such overstatement is less likely to have been committed by firms that practice greater conservatism. To be meritorious, claims based on alleged GAAP violations commonly also must show that shareholders suffered a wealth loss due to stock price decreases tied to a revelation of GAAP violations. Conservative reporting disciplines managers to release bad news to investors as it occurs. This prevents stockpiling of deferred bad news (Kothari, Shu, and Wysocki 2009) and helps prevent subsequent sharp price declines or ‘‘crashes’’ (Kim and Zhang 2013). Therefore, consistent with Watts’s (2003a) litigation explanation for conservatism, we expect that shareholders are less likely to initiate lawsuits against conservative reporters. Our first hypothesis, in alternative form, is: H1: Ceteris paribus, the degree of a reporting firm’s accounting conservatism is negatively associated with the likelihood of a future shareholder lawsuit against the firm, alleging violation of GAAP. As discussed in the ‘‘Introduction’’ section, lack of conditional conservatism likely increases litigation occurrence because it can be viewed as possible evidence of scienter or wrongful intent. In contrast, low levels of our unconditional conservatism metric are less likely to be viewed as intentional distortions of earnings because they are inherently less subject to management discretion (Penman and Zhang 2002, 242). Therefore we expect that rejecting a null of no association in favor of alternative H1 is more likely for the conditional conservatism metrics than for the unconditional conservatism metrics. Given that a firm is sued for alleged violation of GAAP, the consequences again ultimately derive from what is likely to happen should the case go to trial. The market reaction to news of litigation initiated against a more conservative firm should be less negative because the probability of dismissal is higher (Griffin 1996; Griffin, Grundfest, and Perino 2004).13 The probability of dismissal should be higher because the case is less likely to have merit if tried. The amount of any settlement or 9 10 11 12 13 See Johnson et al. (2007) for a related study. For a brief discussion of this literature, and a contrary view, see Ball (2013). The other three incentives for conservatism proposed by Watts (2003a) also are long-term in nature: to facilitate long-term contracting, to defer taxes through conforming financial and tax reporting, and to avoid unfavorable attention from regulators. Chalmers et al. (2012, 25) state that to prevail under Rule 10b-5, which is the most common type of investor class action, plaintiffs must establish ‘‘(1) material misrepresentation or omission by the defendant; (2) the defendant acted with scienter; (3) the material misrepresentation or omission was made in connection with the purchase or sale of a security; (4) the plaintiff relied upon the material misrepresentation or omission; (5) the plaintiff suffered an economic loss; and (6) the plaintiff can allege and prove loss causation.’’ An opposite outcome is possible, although we believe it is less likely. Investors might react more negatively to litigation against a conservative reporter if they view the litigation as new information that the firm is not as conservative as previously believed. In essence, investors might be more negatively surprised by litigation against a conservative reporter than by litigation against unconservative reporters. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 98 award, if not dismissed, should be less because the plaintiffs are less likely to prevail at trial. Finally, the duration of the litigation process should be shorter because litigation is costly to plaintiffs and their lawyers, the cost increases as the process continues, and the costs are less likely to be recouped if the plaintiffs’ ultimate bargaining chip, the threat of going to trial, is worth less. Our second hypothesis, in alternate form, is: H2: Ceteris paribus, given that litigation has been initiated against a reporting firm, the degree of the reporting firm’s accounting conservatism is positively associated with more favorable litigation outcomes for the firm. As with H1, we expect that rejecting a null of no association in favor of alternative H2 is more likely for the conditional conservatism metrics than for the unconditional conservatism metrics.14 An important assumption underlying both H1 and H2 is that there is likely to be variation in the degree of conservatism across firms. Conservatism is costly to firms and managers and the costs can differ across firms. First, conservative firms are more prone to debt covenant violations than are non-conservative firms, and the costs associated with covenant violations are economically significant (Beneish and Press 1993; Zhang 2008). Timely loss recognition makes financial covenants more binding by capitalizing bad news into book values (Ball and Shivakumar 2005; Zhang 2008). Second, conservatism can impose explicit and implicit costs on managers. For example, conservatism reduces currently reported earnings and leads to a depressed book value of equity. If managers’ bonus and other compensation contracts are invariant to accounting choices, then managers have incentives to avoid conservative accounting practices (Zhang 2008; Kim et al. 2013). Furthermore, managers who engage in ‘‘empire building’’ by undertaking unprofitable projects find conservative accounting constrains this activity. Timely loss recognition provides early warnings to boards of directors, which enables them to more promptly identify negative net present value projects and to force managers to discontinue them. For these reasons, managers of some firms might avoid adopting conservative accounting policies. Conservative reporting is an equilibrium response to many factors, including the costs imposed on (and the benefits accruing to) both firms and managers. As a result, there is likely to be cross-sectional variation in the degree of conservatism across firms. This is particularly likely to be true of the measures of conditional conservatism. It is possible that variation in unconditional conservatism is more constrained by GAAP, especially among control firms that are not sued. SAMPLE AND RESEARCH DESIGN Measures of Conservatism There is no widely accepted empirical measure of conditional conservatism or of unconditional conservatism. We employ three alternative measures of conditional conservatism (plus an index that combines the three) and one of unconditional conservatism. Each of the five conservatism metrics is measured on a firm-specific basis, up to the year prior to initiation of litigation (or the year prior to non-initiation in the case of matching control firms). The first measure of conditional conservatism, Consv_Coeff, is the sensitivity of earnings to bad news relative to the sensitivity of earnings to good news. It is derived from Basu’s (1997) asymmetric timeliness (AT) regression. The Basu (1997) cross-sectional model is: Xi ¼ b1 þ b2 Di þ b3 Ri þ b4 Di Ri þ ei ð1Þ In Equation (1), Xi is net income before extraordinary items for firm i, scaled by lagged market value of equity; Ri is the firm i market-adjusted compound stock return over the 12-month period ending at the fiscal year-end; and Di is a dichotomous variable defined as 1 if Ri is negative, and as 0 otherwise. Coefficient b3 is the slope coefficient for positive Ri (the sensitivity of earnings to good news). The slope coefficient for negative Ri (the sensitivity of earnings to bad news) is b3 þ b4, and b4 is the differential slope for bad versus good economic news (AT coefficient). Earnings are reported conservatively if net income is more sensitive to bad news than to good news, that is, if b4 is positive. Basu (1997) defines the sensitivity of earnings to bad news, relative to the sensitivity of earnings to good news, as (b3 þ b4 )/b3. Higher values of this metric, which we denote as Consv_Coeff, indicate greater conservatism.15 Ball, Kothari, and Nikolaev (2013) recommend debiasing the cross-sectional AT 14 15 The two hypotheses stated above distinguish between (1) the initiation (or non-initiation) of litigation against firms, and (2) the outcomes of litigation against firms. This distinction is useful because the two hypotheses apply to different samples. Studying the initiation (or non-initiation) of litigation requires both a sample of test firms that were sued and a sample of control firms that were not sued. We study the outcomes of litigation using only the test (sued) firm sample. The AT parameter b4, incorporated into the sensitivity metric, has been used by Zhang (2008) among others. Some studies have challenged the usefulness of Basu’s (1997) AT coefficient, primarily on the basis that it is estimated with bias and indicates the presence of conservatism in datasets where it is thought not to exist (Dietrich, Muller, and Riedl 2007; Givoly, Hayn, and Natarajan 2007; Patatoukas and Thomas 2011). However, these studies are countered by evidence that Basu’s (1997) coefficient is useful in detecting variations in conservatism in settings where theory suggests conservatism does exist (Ettredge et al. 2012; Ball et al. 2013). Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 99 coefficient by estimation using firm fixed effects. In this spirit, we estimate Model (1) for each sample firm individually using as many (with a minimum of seven) years of time-series data as are available up to the year prior to occurrence or non-occurrence of litigation. Our second measure of conditional conservatism originated with Basu (1995) and was adopted by Givoly and Hayn (2000). Variants have been used by Beatty, Weber, and Yu (2008) among others. The variable, Consv_Negskew, is measured as the time-series (with the minimum of seven years) skewness of earnings before extraordinary items, scaled by the skewness of operating cash flows. When bad news is recognized in an asymmetrically timely fashion, earnings will tend to decrease and generate a negatively skewed earnings series. To make the direction of this measure’s association with litigation outcomes consistent with the other conservatism measures, we multiply the measure by 1, so that the higher the value of Consv_ Negskew, the more conservative the firm. Our third conservatism measure, Consv_Accruals, is a measure of the extent to which earnings include negative nonoperating accruals, and consists of accumulated non-operating accruals scaled by accumulated total assets, with a minimum of seven years. We multiply this measure by 1 so that larger values correspond to greater conservatism (Ahmed, Billings, Morton, and Stanford-Harris 2002; Beatty et al. 2008). Consv_Accruals captures the recording of bad non-operating news in earnings, such as restructuring charges and asset write-downs, and has been used by Zhang (2008) among others. Whereas operating accruals can be positive for growing firms, non-operating accruals are more reliably negative for conservative reporters or firms in decline (Givoly and Hayn 2000). Fourth, we employ a composite measure of conditional conservatism, Consv_Avgrank. For each of the individual measures, we rank the annual observations by percentile, then rescale so that rank for each measure ranges from 0 to 1, and so that higher rescaled values represent greater conservatism. We then take the average of the above three ranked measures as our fourth measure. It is possible that different measures of conditional conservatism contain different levels of measurement errors and are differently associated with various measures of litigation cost. Therefore, we draw our inferences based on the tenor of the results across the four conditional conservatism metrics. Our unconditional conservatism proxy is based on Penman and Zhang (2002), and has been used by Aier, Chen, and Pevzner (2014); Givoly, Hayn, and Katz (2010); and Krishnan and Visvanathan (2008), among others. Penman and Zhang (2002, 242) develop a measure of unconditional conservatism that is ‘‘based on the accounting treatment of three investments for which the accounting is relatively immune from managerial discretion after the expenditure has occurred: inventories, R&D, and advertising.’’ Following Penman and Zhang (2002), we compute a reserve component for each of the three categories and scale the sum of the three reserves by a firm’s net operating assets. We label the resulting variable as UConsv, denoting that the variable is intended to proxy for unconditional conservatism. Variable values of UConsv are larger for firms whose net assets are more conservatively stated.16 Sample and Data We use the Securities Class Action Services (SCAS) database from the Institutional Shareholder Services (ISS) to identify all class action securities lawsuits filed in federal courts from 1996–2011. As discussed previously in the ‘‘Background, Related Literature, and Hypotheses’’ section, the passage of PSLRA in 1995 significantly changed securities litigation laws and lawsuit procedures; we thus restrict our sample period to post-1995.17 We end our sample in 2011 because some of our tests of H2 require data on dismissal or settlement decisions, which generally take several years to resolve. We limit our test sample to lawsuits resolved by June of 2014, and do not include any continuing cases in our sample. All the lawsuits that we sample allege violations of U.S. generally accepted accounting principles, (GAAP).18 Conservative reporting per se is unlikely to prevent allegations of non-GAAP violations such as misleading disclosures, insider trading, or various other managerial actions that do not relate to reported earnings. Therefore, for our major analyses we restrict our sample to lawsuits alleging violations of GAAP (Bratton and Wachter 2011; Chalmers et al. 2012). To the extent that some of the alleged GAAP violations in our test 16 17 18 We also tried using the Beaver and Ryan (2000) measure and a modified measure from Beaver and Ryan (2000) as alternative proxies for unconditional conservatism. However, using those metrics shrinks our sample size significantly. The SCAS database covers 28 useable class actions prior to 1996 when we impose the same data requirements. When we add these cases to our sample, and employ a dichotomous variable in models to distinguish between pre-1996 and post-1996 cases, the untabulated results do not differ from those reported in Tables 4–9. The SCAS database has been widely used in the accounting and finance literature, including studies such as Field, Lowry, and Shu (2005); Cheng, Li, Huang, and Lobo (2010); Donelson, McInnis, Mergenthaler, and Yu (2012); and Dai, Jin, and Zhang (2014). To mitigate the potential concern about duplicates, we carefully examined our test sample of sued firms to ensure there are no duplicate lawsuits. Some test firms are sued more than once in our 16-year sample period, but these are separate lawsuits and there are no duplicates. Our results are robust to including or excluding those lawsuits. Furthermore, in an untabulated additional analysis, we add a control variable, representing prior lawsuits, to all of our regressions. We find that our results are robust while the indicator variable of prior lawsuits is insignificant. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 100 TABLE 1 Sampling Procedure Selection Criteria All resolved lawsuits for U.S. firms initiated in 1996–2011, and resolved by June 2014, from the ISS database and covered by Compustat Less: firms without alleged violations of GAAP Less: firms without Compustat and CRSP data to calculate control variables Less: without closely matched control firms Sample to test H1 Less: without data to calculate five-day returns around the lawsuit date Sample to test H2a, H2b, H2c Less: lawsuits that are dismissed Less: without penalty data Sample to test H2d a b Disposition of the Sample 3,518 2,246 905 4 363 5 358a 145 3 210b The tests of H2 reported in Tables 5–7 are based on these 358 firms. The tests of H2 reported in Table 8 are based on these 210 firms. sample do not involve earnings, it should bias against finding that conservatism deters litigation occurrence and mitigates litigation outcomes. Table 1 summarizes the sample-selection process. We collect all completed securities class action lawsuits filed from January 1, 1996 to December 31, 2011 and resolved by June 2014. We eliminate sample firms not covered by Compustat. For the 3,518 lawsuits with data in Compustat, we eliminate 2,246 that do not allege violations of GAAP. We require that the sample firms have necessary data from CRSP and Compustat to calculate control variables, which eliminates 905 firms. After we eliminate four lawsuits that lack closely matched control firms, the final sample is 363 test firms used to test H1. Testing H1 also requires a sample of control (non-sued) firms. Similar to Donelson et al. (2012) we employ a probability-matched sample of control (non-sued) firms to test H1. Using a matched design yields a comparable set of non-sued control firms that plausibly could have been sued for violations of GAAP, but were not. We identify 363 one-to-one matched control firms based on the closest (within plus or minus 1 percent) ex ante litigation probability in the year before the test (sued) firm’s lawsuit is filed. The ex ante litigation probability is calculated as per Kim and Skinner (2012, Model (3) of Table 7).19 Kim and Skinner (2012, 306) recommend their Model (3) among others because ‘‘overall, model 3 yields a relatively large improvement in predictive ability . . . at relatively low cost.’’ Obtaining a set of non-sued firms similar to our sued firms is critical because ex ante litigation risk may endogenously affect firms’ conservatism levels (Watts 2003a). Matching on ex ante litigation likelihood helps ensure that our results are due to differences in conservatism, rather than to uncontrolled differences between test and control firms in other factors affecting litigation occurrence.20 Our tests of H2 are conducted using only the sample of test (sued) firms. Of the 363 test firms, five do not have market returns around the lawsuit filing date. The tests of H2 for market reactions to lawsuits (REACTION), lawsuit resolution window (DURATION), and lawsuit dismissal probability (DISMISSED) are based on the remaining subsample of 358 firms. Of these, 145 are dismissed and three do not have monetary settlement data for the lawsuit. The test of H2 for the monetary settlement amount (PENALTY) is based on the remaining subsample of 210 firms. Table 2 presents the number of lawsuits and their outcomes by year and distribution of lawsuits by industry. As Panel A shows, the number of lawsuits initiated in our sample varies from year to year, ranging from 7 lawsuits in 1996 to 41 in 2004. The number of lawsuits that are dismissed also varies from 1 in 1996 to 16 in 2002; the number of lawsuits that are settled ranges from 2 in 2011 to 28 in 2004.21 The average monetary settlement amount, scaled by firms’ total assets at the year-end 19 20 21 We compute a firm’s ex ante litigation risk for year t using Kim and Skinner’s (2012) estimated coefficients: SUED ¼ 7.883 þ 0.566(FPSt) þ 0.518(LNASSETSt1) þ 0.982(SALES GROWTHt1) þ 0.379(RETURNt1)  0.108(RETURN SKEWNESSt1) þ 25.635(RETURN STD DEVt1) þ 0.00007(TURNOVERt1). Variables are as defined in Kim and Skinner (2012, Appendix A). Alternate use of the largest available control samples results in similar conclusions regarding signs and significance levels of the conservatism test variables (results not tabulated). However, simply controlling for litigation risk in a regression using the largest available sample is problematic because litigation risk is concentrated in a fairly small subsample of firms (Donelson et al. 2012). Lawsuits rarely go to trial and no lawsuits go to trial in our sample. Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 101 TABLE 2 Distribution of Sample Lawsuits Panel A: Distribution of Lawsuits by Year Calendar Year 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Total/Mean Number of Lawsuits Number of Dismissals Number of Settlements Mean of Scaled Penalties 7 12 15 18 20 17 37 32 41 31 33 30 28 18 15 9 363b 1 4 6 8 9 6 16 12 13 14 8 9 13 11 9 7 146 6 8 9 10 11 11 22 19 28 18 25 21 15 7 6 2 218c 0.158 0.402 0.156 0.167 0.177 0.040 0.137 0.079 0.113 0.056 0.091 0.051 0.141 0.606 0.043 0.036 0.077d Mean of Market Reactions to Lawsuit Filingsa Mean of Resolution Periods (days) 0.030 0.075 0.075 0.069 0.048 0.007 0.116 0.028 0.081 0.009 0.045 0.004 0.045 0.079 0.010 0.001 0.049d 1,082.6 933.5 950.4 1,352.1 1,121.8 1,272.9 1,291.7 906.8 1,110.6 958.2 1,117.4 932.8 1,066.5 784.3 697.0 548.9 1,058.5d a The average market reaction is significantly different from 0 for each of the sample years at the 0.05 or less level. H1 is tested using 363 lawsuits plus matches. The number of lawsuits decreases to 358 for tests of H2a, H2b, and H2c due to missing returns data for five firms. See Table 1. c The number of settlements decreases to 210 for tests of H2d due to missing returns data for five firms and missing penalty data for three firms. See Table 1. d Mean values are based on the full sample. b Panel B: Industry Distribution of Lawsuits Industry Food Products Candy and Soda Alcoholic Beverages Recreational Products Entertainment Printing and Publishing Consumer Goods Apparel Healthcare Medical Equipment Pharmaceutical Products Chemicals Rubber and Plastic Products Construction Steel Works, etc. Fabricated Products Machinery Electrical Equipment Miscellaneous Automobiles and Trucks Number of Lawsuits Percentage of Total 8 2 2 3 3 1 4 6 11 11 23 2 1 2 4 2 8 1 5 5 2.20 0.55 0.55 0.83 0.83 0.28 1.10 1.65 3.03 3.03 6.34 0.55 0.28 0.55 1.10 0.55 2.20 0.28 1.38 1.38 (continued on next page) Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 102 TABLE 2 (continued) Industry Aircraft Shipbuilding Railroad Defense Precious Metals Nonmetallic Mines Petroleum and Natural Gas Utilities Telecommunications Personal Services Business Services Computers Electronic Equipment Measuring and Control Equipment Business Supplies Shipping Containers Transportation Wholesale Retail Restaurants Hotel Motel Banking Insurance Real Estate Total Number of Lawsuits Percentage of Total 3 1 2 3 2 3 25 5 7 45 25 40 7 3 1 3 13 18 2 22 22 7 0.83 0.28 0.55 0.83 0.55 0.83 6.89 1.38 1.93 12.40 6.89 11.02 1.93 0.83 0.28 0.83 3.58 4.96 0.55 6.06 6.06 1.93 363 100.00 prior to the lawsuit is 0.077 for the full sample, which is approximately equal to $157 million (¼ e7.619  0.077) per lawsuit.22 The average market reaction around the lawsuit filing days is 4.9 percent. It takes on average about three years (1,058 days) to resolve the lawsuits, ranging from an average of 549 days in 2011 to 1,352 days in 1999. Panel B of Table 2 shows that firms in the Business Services industry frequently are sued, accounting for 12 percent of our sample, followed by Electronic Equipment (11 percent), Computers (7 percent), and Utilities (7 percent). Model for Testing H1 We first test H1, which posits that greater conservatism is associated with reduced occurrence of litigation in the subsequent year. Our logistic regression model is: ProbðSued ¼ 1Þi;tþ1 ¼ a0 þ a1 CConsvi;t þ a2 UConsvi;t þ a3 DUVOLi;t þ a4 CRASHi;t þ a5 NCSKEWi;t þ a6 FPSi;t þ a7 LNASSETSi;t þ a8 SALES GROWTHi;t þ a9 RETURNi;t þ a10 RETURN SKEWNESSi;t þ a11 RETURN STDDEVi;t þ a12 TURNOVERi;t þ a13 DEBT ISSUEi;t þ a14 EQUITY ISSUEi;t þ a15 MKT SHRi;t þ a16 LNMAKTCAPi;t þ a17 LEVERAGEi;t þ a18 NONDISCTCONSVi;t þ a19 DISC ACCRUALi;t þ a20 BIGNi;t þ Year Indicators þ ei;t ð2Þ All variables are defined in Appendix A and are measured at annual intervals unless otherwise noted. Model (2) is estimated using samples of test and control firms. The dependent variable, Suedi,tþ1, is an indicator variable equal to 1 if a firm is a test firm, and 0 if it is a control firm. CConsvi,t denotes one of the four variables proxying for conditional conservatism up to the year prior to a lawsuit. The variable UConsvi,t denotes the Penman and Zhang (2002) measure of unconditional or GAAP-based conservatism measured up to the 22 Settlement amounts for lawsuits that are not dismissed are publicly available only if they are approved by the courts, and some are not. We confirmed this information in discussion with a representative of Institutional Shareholder Services. Although this data limitation does not affect our major conclusions, we are unable to investigate the associations between conservatism metrics and the terms of private out-of-court settlements (i.e., not approved by courts). Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 103 year prior to a lawsuit. Greater conservatism should decrease the expected value to shareholders of a prospective lawsuit, so H1 predicts that a1 , 0, and a2 , 0. As discussed previously, we expect the H1 prediction is more likely to hold for a1 than for a2. We include several control variables known to influence litigation risk. We first control for stock price crash risk, using CRASH and NCSKEW (Hutton, Marcus, and Tehranian 2009), and a third measure, DUVOL, from Chen, Hong, and Stein (2001). A tendency toward higher crash risk should increase the likelihood that investors subsequently will suffer harm in conjunction with a violation of GAAP. As per Hutton et al. (2009), CRASH is an indicator variable equal to 1 for a firm-year that experiences one or more crash weeks, and 0 otherwise. Variable NCSKEW is a measure of negative skewness of firmspecific weekly returns. The variable DUVOL measures crash risk as down-to-up volatility. We expect the three control variables for stock crash risk to be positively associated with Sued in the next year. The second set of control variables is based on Kim and Skinner (2012): a litigious industry indicator variable (FPS), firm size (LNASSETS), sales growth (SALES_GROWTH), stock return (RETURN), return skewness (RETURN_SKEWNESS), return standard deviation (RETURN_STDDEV), and stock turnover (TURNOVER).23 In general, firms with greater size (LNASSETS), higher growth (SALES_GROWTH), greater share turnover (TURNOVER), firms in high litigation risk industries (FPS), and having higher return volatility (RETURN_STDDEV), are more likely to be sued. Firm-years with higher stock returns (RETURN) and greater return skewness (RETURN_SKEWNESS) are less likely to precede litigation.24 The third set of control variables represents firm characteristics other than litigation risk that are thought to motivate reporting conservatism. We wish to ensure that the conservatism metrics do not proxy for underlying constructs that are correlated with both litigation risk and conservatism. Aside from a desire to avert litigation, demand for conservatism primarily arises from debt contracting and equity contracting considerations (Watts 2003a). Lawrence, Sloan, and Sun (2013) employ a debt issuance variable, DEBT_ISSUE, as a proxy for debt contracting demand and an equity issuance variable, EQUITY_ ISSUE, as a contractual incentive favoring conservatism. We therefore add the variables DEBT_ISSUE and EQUITY_ISSUE to our model. The variable MKT_SHR captures a firm’s sales market share in its industry and represents regulatory concerns motivating conservatism (Qiang 2007). We do not specify expected signs of association for these three variables. Further, we add two variables thought to influence conditional conservatism (Kahn and Watts 2009; Lawrence et al. 2013): the log of market capitalization (LNMAKTCAP), and financial leverage (LEVERAGE, Watts 2003a; LaFond and Watts 2008). We expect the first to have a positive coefficient and do not specify the expected sign for the second. We add a measure of nondiscretionary conservatism, NONDISCTCONSV. This variable, equal to the beginning of period book-to-market asset ratio, is intended to proxy for conservatism resulting from the unbiased application of GAAP (Lawrence et al. 2013).25 However, we do not view it as a clear measure of conditional or unconditional conservatism and so do not categorize it as either.26 Nor do we specify an expected coefficient sign. In addition, as discussed previously in ‘‘Related Literature’’ section, we control for earnings management via discretionary accruals (DISC_ACCRUAL), to determine whether our conservatism metrics have additional explanatory power beyond a signed measure of discretionary accruals. Discretionary accruals are measured as in Dechow and Dichev (2002). We expect DISC_ACCRUAL to have a positive coefficient. Finally, we include a dichotomous BIGN control variable because Big N auditors likely affect clients’ accounting choices and might also require more conservative reporting. However, the Big N audit firms are often said to have ‘‘deep pockets’’ and could attract litigation on that basis, so we do not specify a sign of association for BIGN. Model for Testing H2 We next test H2, which posits that, given litigation is initiated, greater conservatism in year t is associated with better outcomes for reporting firms in the subsequent years.27 Our OLS or logit model (depending on the dependent variable) is: 23 24 25 26 27 Although the test and control firms are matched using a litigation probability metric derived from a Kim and Skinner (2012) model, we include some of the Kim and Skinner (2012) model variables as explanatory variables in our models. This controls for residual variation due to imperfect matching. In additional analyses we supplement the litigious industry variable, FPS, with a measure of the risk of litigation against the auditor developed by Shu (2000). All results reported in Tables 4 through 8 are robust to inclusion of this litigation risk variable in the models (not tabulated). Lawrence et al. (2013, 117) employ a book-to-market asset ratio ‘‘defined as total assets scaled by the sum of market capitalization plus total assets minus common equity, all measured as of the end of fiscal year t1.’’ Cheng, Huang, and Li (2015) and Dhaliwal, Huang, Khurana, and Pereira (2014) use book-to-market equity ratios as measures of unconditional conservatism. Roychowdhury and Watts (2007) and some other authors view the market-to-book equity ratio as a proxy for general conservatism rather than as a proxy only for unconditional conservatism. Lawrence et al. (2013, 113) note that ‘‘our model of non-discretionary conservative accounting explains only a small proportion of the observed variation in conservative accounting. This result is consistent with both limitations of our model to perfectly capture the circumstances under which GAAP requires asset write-downs and with the existence of significant discretionary conservatism. We therefore emphasize that our results are consistent with the existence of significant discretionary conservatism.’’ We do not specify a particular time lag n for measurement of these variables since the outcomes can occur in various years relative to the measurement of conservatism in year t. Market reaction occurs in the year the lawsuit is initiated. Other outcomes occur in subsequent years. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 104 Outcomei;tþn ¼ a0 þ a1 CConsvi;t þ a2 UConsvi;t þ a3 DUVOLi;t þ a4 CRASHi;t þ a5 NCSKEWi;t þ a6 FPSi;t þ a7 LNASSETSi;t þ a8 SALES GROWTHi;t þ a9 RETURNi;t þ a10 RETURN SKEWNESSi;t þ a11 RETURN STDDEVi;t þ a12 TURNOVERi;t þ a13 DEBT ISSUEi;t þ a14 EQUITY ISSUEi;t þ a15 MKT SHRi;t þ a16 LNMAKTCAPi;t þ a17 LEVERAGEi;t þ a18 NONDISCTCONSVi;t þ a19 DISC ACCRUALi;t þ a20 BIGNi;t þ a21 CLASSPERIODi;t þ Year Indicators þ ei;t ð3Þ All variables are defined in Appendix A and are measured at annual intervals unless otherwise noted. Model (3) is estimated using only test firms with necessary data. The dependent variable, Outcomei,tþn, is one of the four variables proxying for litigation outcomes. The four outcomes of litigation are the market reaction to the lawsuit announcement (REACTIONi,tþn), the length of the resolution period (DURATIONi,tþn), a dichotomous variable capturing whether a court dismissed the case (DISMISSEDi,tþn), and any courtapproved settlement amount (PENALTYi,tþn). The first dependent variable, REACTION, is the five-day market-adjusted cumulative abnormal stock returns around the filing of a lawsuit, for testing H2a. The second dependent variable, DURATION, is the number of days that lapse from the initiation of a lawsuit until the lawsuit is resolved via dismissal, or settlement, for testing H2b. Given a lawsuit is filed, the third dependent variable, DISMISSED, is an indicator variable equal to 1 if the lawsuit is dismissed, and as 0 otherwise, for testing H2c. Given that a lawsuit is not dismissed, the fourth dependent variable, PENALTY, is the settlement amount (if any), scaled by total assets in the year prior to the lawsuit, for testing H2d. PENALTY reflects settlement amounts approved by courts. Data on private settlements not approved by courts are unavailable to the public. Because H2 predicts that higher degrees of conservatism are associated with more favorable outcomes, we expect that a1 . 0 and a2 . 0 when the dependent variable is REACTION or DISMISSED and that a1 , 0 and a2 , 0 when the dependent variable is DURATION or PENALTY. As discussed previously, we expect that estimated coefficient a1 will provide stronger support for H2 compared to a2. The three equations having continuous dependent variables, REACTION, DURATION, and PENALTY are estimated using OLS regressions; the equation having a dichotomous dependent variable, DISMISSED, is estimated using logistic regression. Our basic concept is that the likely merits of a lawsuit determine both whether it will be initiated and the outcome if it is initiated. Thus we employ almost the same test and explanatory variables in both Models (2) and (3), except that we add two new control variables to Model (3) that are not available for the non-sued firms used to estimate Model (2). The variable CLASSPERIOD captures the number of days that lapse from the beginning to the end of the litigation class period. The longer the class period, the more adverse the litigation outcomes tend to be (Carleton, Weisbach, and Weiss 1996). In the regressions explaining DURATION, DISMISSED, and PENALTY, we also include the five-day market-adjusted abnormal returns to control for the overall market assessed severity of the lawsuit at the filing day (Bajaj, Mazumdar, and Sarin 2003). EMPIRICAL RESULTS Descriptive Statistics We report descriptive statistics in Table 3. All continuous variables are winsorized at the top and bottom 1 percent to mitigate the influences of extreme values. The similarity of Litigation Probability across the test (sued) firms and the control (non-sued) firms indicates essentially no difference in the ex ante probability of being sued, based on the Kim and Skinner (2012) metric.28 The next four rows of Table 3 present the comparisons of four measures of conservatism between the sued and non-sued firms. The three conditional conservatism measures all indicate that lower conservatism exists among sued firms than among non-sued firms prior to litigation. The mean values of Consv_Coeff and Consv_Negskew are negative (positive) for sued firms (non-sued firms) and the differences in means are significant. The differences in means of these two measures of conditional conservatism indicate that compared to the non-sued firms, the sensitivity of the sued firms’ earnings to bad news is lower than the sensitivity of their earnings to good news, and the sued firms’ earnings are positively skewed. The means of Consv_Accruals also suggest that sued firms are less conservative than non-sued firms, as they accumulate more negative nonoperating accruals prior to litigation. The means of the unconditional conservatism variable, UConsv, do not differ across subsamples, likely due to the large standard deviation of this variable’s observations among sued firms. This similarity arguably explains, in part, the lack of association between UConsv and litigation status in subsequent regressions. In untabulated results, we find that the Pearson correlation coefficients for Sued and the three individual measures of conditional conservatism are negative and significantly differ from 0 (p-value ¼ 0.012 or less). The correlation coefficient between Consv_Negskew and Consv_Accruals is 0.115 and is highly significant (p-value ¼ 0.002). Correlations between 28 The first variable shown in Table 3, Litigation Probability, is not a dependent variable or an independent variable but rather is the measure of ex ante litigation probability used to select one-to-one matched control firms. Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 105 TABLE 3 Descriptive Statistics Sued Firms n Litigation Probability Dependent Variables REACTION DURATION DISMISSED PENALTY Test Variables Consv_Coeff Consv_Negskew Consv_Accruals UConsv Control Variables DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD Mean STD Non-Sued Firms Median 363 0.459 0.284 0.402 358 358 358 210 0.049 1058.5 0.408 0.077 0.130 721.5 0.492 0.101 0.020 919.5 0.000 0.042 363 363 363 363 1.143 0.762 0.175 0.642 14.386 3.698 0.269 15.829 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 358 0.082 0.278 0.235 0.380 7.619 0.157 0.048 0.065 0.137 2.951 0.214 0.031 0.623 7.525 0.714 0.631 0.014 0.802 763.6 0.378 0.449 0.857 0.486 2.273 0.317 0.623 0.803 0.082 2.872 0.586 0.087 0.485 1.962 1.913 0.298 0.167 0.399 1812.3 n Mean STD Median Mean Difference t-value 363 0.434 0.285 0.373 0.024 1.16 0.590 0.765 0.165 0.038 363 363 363 363 2.622 0.182 0.331 0.140 15.821 4.518 1.152 1.491 0.538 0.694 0.187 0.022 3.765 0.944 0.156 0.781 3.35*** 3.08*** 2.51*** 0.94 0.032 0.000 0.060 0.000 7.733 0.068 0.100 0.066 0.116 2.093 0.010 0.007 1.000 7.481 0.171 0.606 0.001 1.000 463.5 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 363 0.045 0.185 0.052 0.331 6.750 0.117 0.134 0.278 0.149 1.954 0.218 0.028 0.460 6.465 0.869 0.704 0.003 0.769 0.384 0.388 0.880 0.471 2.625 0.318 0.653 0.877 0.090 2.435 0.610 0.079 0.499 2.399 2.812 0.308 0.154 0.422 0.061 0.000 0.063 0.000 6.736 0.059 0.033 0.201 0.128 1.275 0.008 0.005 0.000 6.538 0.212 0.706 0.001 1.000 0.127 0.094 0.287 0.050 0.869 0.040 0.086 0.213 0.012 0.997 0.004 0.003 0.163 1.060 0.155 0.073 0.011 0.033 4.48*** 3.01*** 4.46*** 1.40 4.77*** 1.69 1.81 3.41*** 1.94* 5.05*** 0.10 0.52 4.45*** 6.51*** 0.87 3.24*** 0.93 1.08 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. Consv_Accruals and Consv_Coeff, and between Consv_Negskew and Consv_Coeff, are not significantly different from 0. Given that all three measures of conditional conservatism are significantly associated with various measures of legal costs in the expected directions in our regression models, the low or insignificant correlation coefficients between them suggest that the three metrics represent very different dimensions of earnings conservatism, each of which reduces legal costs. The correlations of UConsv with the three conditional conservatism measures are insignificant. The control variables exhibit some differences across subsamples. Test firms have greater means of the crash risk variables DUVAL, CRASH, and NCSKEW than control firms. Test firms are larger in total assets, LNASSETS, have lower RETURN_ SKEWNESS and RETURN_STDDEV, and greater trading volume, TURNOVER. Further, test firms face higher regulation costs, as they have higher product market shares, MKT_SHR. Test firms have greater market capitalization, LNMAKTCAP, than control firms. Finally, test firms have lower nondiscretionary conservatism, NONDISCTCONSV, than control firms. Test of H1 Table 4 presents the logistic regressions of the likelihood of litigation on the levels of conditional and unconditional conservatism and the control variables in the prior year (Model (2)). For all three conditional conservatism measures (Consv_ Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 106 TABLE 4 Logistic Regression of the Likelihood of Litigation on the Level of Conservatism (H1) Consv_Coeff Consv_Negskew Consv_Accruals Consv_Avgrank Variables Exp. Sign Coeff. z-value Coeff. z-value Coeff. z-value Coeff. z-value Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN   þ þ þ þ þ þ   þ þ þ/ þ/ þ/ þ þ/  þ þ/ 1.664 0.017 0.010 0.031 0.171 0.166 0.002 0.086 0.183 0.297 0.098 0.006 0.156 0.253 0.371 0.106 0.313 0.045 0.236 0.432 0.281 1.80 2.37** 0.54 0.04 0.66 0.45 0.01 0.48 0.64 1.83* 0.86 0.00 3.13*** 1.43 0.35 0.45 1.51 0.85 0.38 0.80 1.28 1.859 0.059 0.008 0.174 0.132 0.118 0.033 0.075 0.205 0.275 0.091 0.404 0.159 0.290 0.229 0.091 0.312 0.041 0.263 0.367 0.298 2.05 2.53** 0.49 0.23 0.50 0.32 0.17 0.42 0.72 1.67* 0.79 0.27 3.20*** 1.63 0.22 0.39 1.52 0.79 0.43 0.68 1.36 1.685 0.658 0.009 0.125 0.081 0.254 0.033 0.124 0.119 0.279 0.096 0.642 0.165 0.248 0.321 0.099 0.354 0.040 0.192 0.080 0.270 1.85 2.61** 0.54 0.16 0.31 0.70 0.17 0.70 0.41 1.70* 0.84 0.42 3.26*** 1.40 0.30 0.42 1.73*** 0.74 0.32 0.13 1.23 1.201 1.220 0.009 0.063 0.128 0.177 0.032 0.099 0.186 0.292 0.089 0.261 0.158 0.254 0.298 0.046 0.317 0.044 0.186 0.248 0.252 1.26 2.19** 0.55 0.08 0.49 0.48 0.17 0.55 0.65 1.79* 0.78 0.17 3.20*** 1.44 0.28 0.20 1.54 0.82 0.31 0.45 1.15 Year Indicators Pseudo R2 ROC Curve Statistics Adjusted R2 Chi-square Likelihood n Included 0.098 0.636 0.073 54.47 726 Included 0.098 0.631 0.063 47.13 726 Included 0.101 0.623 0.062 46.52 726 Included 0.096 0.633 0.063 47.06 726 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. Coeff, Consv_Negskew, Consv_Accruals), as well as the summary measure (Consv_Avgrank), the coefficients are negative and statistically significant at less than the 5 percent level, indicating that firms reporting higher conditional conservatism are less likely to be sued in the following year in investors’ class action lawsuits alleging violations of GAAP. The effect is also economically significant. For example, the improvement in Consv_Avgrank from the bottom percentile to the top percentile is associated with a decrease in the probability of a lawsuit by 30 percent.29 The coefficients on UConsv are negative but insignificant, suggesting that higher unconditional conservatism does not play an important role in averting litigation occurrence. In Table 4, the coefficients on most of the control variables have the expected signs when significant.30 Larger clients, measured as log of market cap (LNMAKTCAP), are more likely to be sued, probably because they offer more attractive targets for lawyers specializing in investor class actions. Controlling for market cap, larger clients, measured as log of total assets (LNASSETS), are less likely to be sued, perhaps due to higher-quality internal controls. Greater trading volume (TURNOVER) is positively associated with litigation occurrence, probably because greater trading volume increases the likelihood of investors buying and/or selling during periods of alleged improper financial reporting, which better enables investors to claim harm due to relying on that information, consistent with prior work such as Kim and Skinner (2012). 29 30 Economic significance is evaluated by the marginal probability, calculated as 1/(1 þ e(aþbX)), where a þ bX is the predicted probability from the logistic regression. The modest number of significant control variable coefficients likely results to some extent from the Litigation Probability matching of test and control firms. Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 107 TABLE 5 OLS Regression of the Market Reactions to Lawsuit Filings on the Level of Conservatism (H2a) Variables Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD Year Indicators Exp. Sign þ þ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ Adjusted R2 n Consv_Coeff Coeff. t-value 0.048 0.60 0.001 2.23** 0.000 0.12 0.060 1.01 0.013 0.61 0.012 0.41 0.003 0.19 0.005 0.32 0.005 0.21 0.027 2.04** 0.023 2.37** 0.103 0.82 0.002 0.77 0.012 0.83 0.176 2.02** 0.031 1.60 0.001 0.08 0.001 0.15 0.031 0.59 0.036 0.88 0.010 0.57 0.000 0.64 Included 0.0545 358 Consv_Negskew Coeff. t-value 0.056 0.71 0.004 2.09** 0.000 0.01 0.068 1.15 0.013 0.60 0.016 0.55 0.001 0.06 0.001 0.04 0.005 0.23 0.025 1.94* 0.024 2.49*** 0.091 0.72 0.002 0.80 0.012 0.83 0.195 2.25** 0.030 1.55 0.007 0.39 0.000 0.03 0.040 0.76 0.037 0.91 0.010 0.58 0.000 0.76 Included Consv_Accruals Coeff. t-value 0.028 0.49 0.029 2.53** 0.000 0.27 0.048 1.11 0.029 1.89 0.009 0.42 0.001 0.11 0.006 0.57 0.003 0.20 0.008 0.85 0.008 1.16 0.072 0.77 0.002 0.88 0.005 0.49 0.059 0.92 0.013 0.93 0.010 0.73 0.001 0.22 0.033 0.86 0.013 0.40 0.007 0.52 0.000 1.79 Included 0.0527 358 0.0149 358 Consv_Avgrank Coeff. t-value 0.080 0.97 0.093 2.30** 0.000 0.08 0.040 0.68 0.013 0.61 0.003 0.12 0.001 0.07 0.003 0.20 0.004 0.16 0.024 1.85* 0.028 2.90*** 0.080 0.64 0.002 0.85 0.024 1.59 0.090 0.99 0.032 1.63 0.008 0.46 0.004 0.77 0.053 1.02 0.029 0.71 0.013 0.72 0.000 1.30 Included 0.0541 358 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. In summary, the evidence in Table 4 suggests that higher conditional conservatism averts litigation occurrence in the following year.31 Unconditional conservatism does not provide this benefit to the same extent. The lack of significant association between unconditional conservatism and litigation likelihood (and subsequently its lack of association with litigation costs) is puzzling. One possible explanation is that unconditional conservatism is inherently less subject to management discretion. Lower levels of unconditional conservatism metrics are less likely to be viewed as intentional distortions of earnings. In contrast, lack of conditional conservatism likely increases litigation occurrence because it can be viewed as possible evidence of wrongful intent. Another possible explanation is that researchers have invested much effort in developing conditional conservatism measures over several decades. In contrast, less effort has been devoted to developing unconditional conservatism measures until recently. The current proxies for unconditional conservatism in the literature simply might not be very powerful. Tests of H2 Tables 5–8 provide evidence whether conservative firms experience more favorable outcomes of lawsuits (H2a through H2d). Table 5 presents results of OLS regressions explaining market reactions to lawsuits (REACTION) using the conservatism 31 Recent literature documents that conservatism is associated with strong corporate governance (Ahmed and Duellman 2007; Garcia Lara, Garcia Osma, and Penalva 2009; Ettredge et al. 2012; Ramalingegowda and Yu 2012). We try to capture the impact of corporate governance using the G-index, Eindex, and some board characteristics such as board independence, percentage of insider holding, and numbers of committees on which CEOs sit. However, the sample size shrinks substantially after including corporate governance control variables (by more than one-half ) and the coefficients for all conservatism metrics become insignificant. The lack of significance is due to the reduction in sample size, not the inclusion of governance variables. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 108 TABLE 6 OLS Regression of the Lawsuit Resolution Period on the Level of Conservatism (H2b) Variables Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD REACTION Year Indicators Adjusted R2 n Exp. Sign   þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ Consv_Coeff Coeff. t-value 654.32 1.75 3.462 2.20** 3.035 1.53 31.65 0.11 102.76 1.03 6.807 0.05 73.41 1.04 61.34 0.84 140.58 1.28 43.63 0.70 100.11 2.17** 442.30 0.72 8.905 0.65 40.79 0.58 527.13 1.29 31.85 0.35 49.42 0.59 11.30 0.42 242.01 0.96 143.28 0.75 127.44 1.52 0.062 0.97 543.06 2.07** Included 0.0485 358 Consv_Negskew Coeff. t-value 890.10 2.32 17.13 2.25** 1.742 0.96 49.58 0.17 25.02 0.23 43.33 0.31 78.81 1.07 145.32 1.81* 173.25 1.52 103.28 1.64* 78.23 1.65* 924.08 1.55 17.07 1.24 38.09 0.55 302.00 0.69 37.65 0.40 120.30 1.32 41.43 1.15 544.99 2.04** 260.85 1.06 146.16 1.68* 0.102 1.49 260.87 0.98 Included Consv_Accruals Coeff. Consv_Avgrank t-value 596.02 1.57 220.95 2.22** 53.69 1.25 38.45 0.13 63.77 0.63 8.315 0.06 52.21 0.71 66.64 0.86 179.48 1.58 48.59 0.76 101.51 2.10** 662.79 1.05 5.472 0.38 24.33 0.34 288.50 0.70 50.38 0.55 46.36 0.52 8.046 0.28 261.83 1.00 79.89 0.32 138.41 1.62 0.069 1.05 542.87 1.99** Included 0.0449 358 0.0607 358 Coeff. t-value 749.71 1.89 384.60 1.97** 2.331 1.19 24.83 0.09 115.46 1.14 31.02 0.23 48.19 0.66 81.00 1.08 143.72 1.27 50.17 0.79 117.15 2.48*** 792.87 1.29 8.285 0.59 22.64 0.31 257.37 0.61 79.41 0.84 70.16 0.80 17.02 0.59 258.51 1.00 175.03 0.89 99.78 1.15 0.046 0.70 467.96 1.75* Included 0.0534 358 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. metrics. We find the coefficients are significantly positive for all four measures of conditional conservatism, consistent with the prediction that market reactions to news of litigation against more conservative firms are less negative. For example, the coefficient on Consv_Avgrank is 0.093, significant at the 5 percent level. Economically, this coefficient means that an improvement in Consv_Avgrank from the bottom percentile to the top is associated with an increase in REACTION of around 930 basis points. On average, an increase in Consv_Avgrank by one standard deviation (0.168) is associated with an increase of REACTION by 156 basis points (0.168  0.093). This is equivalent to a 32 percent (standard deviation of Consv_Avgrank at 0.168  0.093/0.049) improvement for the average firm in our sample (with REACTION equal to 4.9 percent). The coefficients on UConsv are insignificant. Table 6 presents OLS regression results explaining the duration of lawsuits (DURATION) using the measures of conservatism. We find the coefficients are significantly negative for all four measures of conditional conservatism. These results suggest that the litigation process is resolved sooner when defendant firms are more conditionally conservative. For example, the coefficient on Consv_Avgrank is 384.60, significant at the 5 percent level. Economically, this coefficient means that an improvement in Consv_Avgrank from the bottom percentile to the top is associated with a decrease in DURATION of around 385 days, which is equivalent to a 36 percent (384.60/1,058.5) reduction in the number of days for the average firm in our sample (with DURATION equal to 1,058.5 days). On average, an increase in Consv_Avgrank by one standard deviation (0.168) is associated with a decrease of 65 (384.60  0.168) days to resolve a lawsuit. The coefficients on UConsv do not differ from 0. In logistic regressions of the likelihood of lawsuit dismissal on conservatism (DISMISSED) in Table 7, we find the coefficients are significantly positive for all four measures of conditional conservatism, consistent with the prediction that lawsuits against more conservative firms are more likely to be dismissed by courts, as they are less likely to have merit. For Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 109 TABLE 7 Logistic Regression of the Likelihood of Dismissal on the Level of Conservatism (H2c) Variables Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD REACTION Year Indicators Exp. Sign þ þ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/– þ/ þ/ þ/ Pseudo R2 ROC Curve Statistics Adjusted R2 Chi-square Likelihood n Consv_Coeff Coeff. z-value –0.254 –0.17 0.025 2.52*** –0.007 –0.36 0.898 0.80 0.721 1.86* –0.747 –1.38 –0.129 –0.45 –0.402 –1.42 0.551 1.27 –0.143 –0.55 –0.051 –0.28 –0.974 –0.40 0.116 1.79* 0.027 0.10 –0.443 –0.27 0.191 0.52 0.480 1.45 0.082 0.80 1.142 1.18 –0.861 –1.01 0.315 0.94 –0.001 –2.68*** 0.150 0.15 Included 0.126 0.725 0.157 61.12 358 Consv_Negskew Coeff. z-value –0.347 –0.22 0.075 1.98** –0.016 –0.87 0.556 0.49 0.746 1.93* –0.595 –1.09 –0.176 –0.63 –0.491 –1.69* 0.480 1.10 –0.113 –0.45 –0.010 –0.05 –1.200 –0.50 0.111 1.80* –0.022 –0.08 –0.237 –0.15 0.222 0.61 0.571 1.68 0.114 1.10 1.326 1.33 –0.818 –0.96 0.299 0.90 –0.001 –2.75*** 0.459 0.45 Included Consv_Accruals Coeff. z-value –0.149 –0.10 1.297 2.32** –0.014 –0.79 0.850 0.75 0.888 2.27** –0.776 –1.44 –0.191 –0.67 –0.383 –1.28 0.580 1.34 –0.104 –0.41 –0.029 –0.16 –1.568 –0.65 0.104 1.68* 0.024 0.09 –0.225 –0.14 0.194 0.53 0.452 1.31 0.100 0.97 1.085 1.05 –0.366 –0.38 0.228 0.68 –0.001 –2.74*** 0.254 0.25 Included 0.118 0.719 0.148 57.22 358 0.125 0.724 0.156 60.52 358 Consv_Avgrank Coeff. z-value –1.009 –0.63 1.791 2.11** –0.014 –0.77 0.683 0.61 0.808 2.08** –0.689 –1.28 –0.175 –0.62 –0.403 –1.41 0.537 1.24 –0.128 –0.51 –0.023 –0.12 –1.258 –0.53 0.110 1.75* 0.004 0.01 –0.229 –0.14 0.123 0.33 0.498 1.50 0.098 0.96 1.113 1.14 –0.553 –0.65 0.222 0.66 –0.001 –2.64*** 0.294 0.29 Included 0.118 0.719 0.148 57.27 358 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. example, the coefficient on Consv_Avgrank is 1.791, significant at the 5 percent level. Economically, this coefficient means that an improvement in Consv_Avgrank from the bottom percentile to the top is associated with an increase in the probability of a dismissal by 31 percent (the marginal effect). This is equivalent to 95.6 percent of the improvement in the chances of dismissals for the average firm in our sample (with DISMISSED equal to 40.8 percent).32 The coefficients on UConsv are insignificant. Table 8 presents OLS regression results using conservatism measures to explain the scaled settlement amount (PENALTY). The coefficients are significantly negative for each of the four measures.33 These results suggest that the court-approved monetary settlement is lower in amount even when lawsuits against more conservative firms are not dismissed in court. For example, the coefficient on Consv_Avgrank is 0.071, significant at the 5 percent level. Economically, this coefficient means an improvement in Consv_Avgrank from the bottom percentile to the top is associated with a decrease in PENALTY by 0.071. This is equivalent to an average amount of $145 million (¼ e7.619  0.071), equal to 92.2 percent of the penalty reduction for the average firm in our sample (with PENALTY equal to 0.077). The coefficients on UConsv are insignificant in general, except for the Consv_Negskew regression. 32 33 The 40.8 percent is equal to the 146 dismissals divided by the 358 total lawsuits in our sample. The conditional conservatism coefficients are not conventionally significant when estimated using penalty data winsorized at the top and bottom 1 percent level. Investigation suggested that this result is due to extreme values in the tails of the penalty distribution. When we winsorize the penalty variable amounts at the top and bottom 3 percent, as in Table 8, the conditional conservatism metrics are significant. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 110 TABLE 8 OLS Regression of the Settlement Penalty on the Level of Conservatism (H2d) Variables Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD REACTION Year Indicators Adjusted R2 n Exp. Sign   þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ/ þ þ/ þ/ þ/ þ/ þ þ/ þ/ þ þ/ þ þ/ Consv_Coeff Coeff. t-value 0.077 1.17 0.001 2.13** 0.015 0.97 0.000 0.01 0.020 1.22 0.008 0.39 0.002 0.18 0.009 0.86 0.009 0.42 0.004 0.40 0.005 0.68 0.117 1.18 0.006 2.01** 0.008 0.57 0.076 0.91 0.011 0.76 0.001 0.05 0.000 0.01 0.026 0.70 0.069 1.75* 0.020 1.54 0.000 1.81* 0.060 1.51 Included 0.2848 210 Consv_Negskew Coeff. t-value 0.065 1.11 0.002 2.08** 0.031 2.21** 0.071 1.78* 0.018 1.22 0.021 1.10 0.002 0.20 0.054 5.55*** 0.015 0.82 0.010 1.22 0.011 1.63* 0.332 3.79*** 0.001 0.34 0.006 0.50 0.048 0.64 0.006 0.51 0.050 4.31*** 0.006 1.44 0.152 4.62*** 0.064 1.83* 0.015 1.27 0.000 2.08** 0.079 2.24** Included Consv_Accruals Coeff. t-value 0.061 0.84 0.027 2.00** 0.023 1.32 0.058 1.16 0.011 0.63 0.025 1.08 0.002 0.16 0.053 4.35*** 0.004 0.17 0.011 1.02 0.006 0.76 0.142 1.30 0.002 0.56 0.005 0.34 0.017 0.18 0.017 1.10 0.051 3.51*** 0.004 0.85 0.151 3.71*** 0.098 2.24** 0.029 1.96** 0.000 1.66* 0.052 1.17 Included 0.2816 210 0.2789 210 Consv_Avgrank Coeff. t-value 0.027 0.35 0.071 2.03** 0.012 0.67 0.072 1.40 0.021 1.14 0.033 1.35 0.003 0.25 0.054 4.28*** 0.004 0.16 0.007 0.65 0.007 0.78 0.209 1.86* 0.003 0.83 0.013 0.78 0.011 0.12 0.010 0.62 0.049 3.29*** 0.006 1.18 0.151 3.57*** 0.067 1.47 0.029 1.88* 0.000 1.68* 0.044 0.97 Included 0.3068 210 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. Collectively, the results reported in Tables 4–8 demonstrate the benefits of conditional conservatism (but not unconditional conservatism) to reporting firms in averting litigation occurrence and reducing negative outcomes. These results hold after controlling for other factors that affect firms’ litigation risk and factors that motivate accounting conservatism. Our results provide empirical evidence for the Watts’s (2003a) litigation explanation of accounting conservatism. They are consistent with the view that conditional conservatism plays an important role in reducing the expected value of litigation to plaintiffs and to plaintiffs’ attorneys and therefore in averting litigation occurrence and mitigating subsequent litigation outcomes. Additional Analysis: Falsification Test Our main results suggest that conditional conservatism benefits reporting firms by averting future lawsuits alleging GAAP violations and by improving lawsuit-related outcomes. It is possible that these results are due to uncontrolled firm characteristics that both (1) are positively associated with conservative reporting, and (2) are negatively associated with litigation occurrence and costs. In order to mitigate the concern that our results are due to other uncontrolled firm characteristics that might avert litigation, such as some aspect of corporate governance, we perform additional analyses. Our concept is that conditional conservatism deters GAAP-related litigation and lowers the costs of such litigation by reducing the expected merits of plaintiffs’ cases. Conditional conservatism should not reduce litigation risk arising from nonGAAP issues such as insider trading. Yet such problems could be deterred or reduced by stable firm characteristics other than conservatism, such as superior corporate governance. The availability of securities lawsuits dealing with non-GAAP issues therefore provides us with an opportunity to falsify our main results. If the conditional conservatism metrics are negatively associated with litigation occurrence using a sample of lawsuits not alleging violations of GAAP, then we must conclude that Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 111 conditional conservatism simply proxies for underlying firm characteristics that are determinants of both conservatism and all types of litigation risk. We start with the 2,246 lawsuits that do not allege violation of GAAP (see Table 1). We impose the same requirement that these test firms have necessary data from CRSP and Compustat, which reduces our test sample size to 416. As in the main analyses testing H1, we identify one-to-one matched control firms based on the closest ex ante litigation probability in the year before the test firm is sued. We eliminate seven lawsuits due to lack of closely matched control firms. The result is a final sample of 409 lawsuit firms and 409 matching control firms. We note that the test sample size of 409 is larger than the test sample of 363 for our main results in Table 4. Thus, any lack of significance of conditional conservatism metrics in the additional analysis should not be due to small sample size. We estimate the same models reported in Table 4 but using the alternative ‘‘falsification’’ test sample rather than the original test sample. In untabulated results, we find that the coefficients for each of the three conditional conservatism measures, as well as the summary measure Consv_Avgrank, are insignificantly different from 0 and the signs are mixed. The most significant z-value for a conditional conservatism measure in the four regressions is 0.89 (with a corresponding p-value of 0.37), indicating that the degree of conditional conservatism does not avert the future litigation occurrence. The coefficients for UConsv are consistently negative but none differ significantly from 0 (p-values are 0.61 and higher). These results suggest that neither conditional nor unconditional conservatism plays a role in averting future occurrences of lawsuits alleging issues other than GAAP violations.34 Turning to the control variables in these tests, we find that likelihood of lawsuits is positively associated with firm size measured as the log of market cap (LNMAKTCAP), with lower market returns (RETURN), and with greater trading volume (TURNOVER). The results for these control variables that capture the firm-specific characteristics are similar to those in Table 4. In addition, we find that firms with higher discretionary accruals (DISC_ACCRUAL) are more likely to be sued, consistent with Chalmers et al. (2012).35 Overall, the falsification tests provide additional evidence suggesting that it is conditional conservatism that averts the future occurrences of lawsuits alleging GAAP violation, and mitigates their effects, and probably not uncontrolled firm characteristics. Robustness Analysis: Using the C-Score Conditional Conservatism Metric In this section, we employ the Khan and Watts’s (2009) C-Score metric as an alternative measure for conditional conservatism. The C-Score is estimated using annual Basu (1997) regressions, specifying the differential timeliness coefficient as a linear function of company size, market-to-book equity ratio, and financial leverage. The C-Score metric has the advantage that it provides estimated differential timeliness parameters for individual companies and individual years. It has merit in our setting because it allows us to investigate whether less conservative reporting of income in a given year is followed by investor class action lawsuits in the next year. To obtain the C-Score we follow the procedures described in Khan and Watt’s (2009, Section 2). In particular, we estimate their extended Basu (1997) model as an annual cross-sectional regression: Xi;t ¼ b1 þ b2 Di;t þ Ri;t ðl1t þ l2t Sizei;t þ l3t M=Bi;t þ l4t Levi;t Þ þ Di;t Ri;t ðk1t þ k2t Sizei;t þ k3t M=Bi;t þ k4t Levi;t Þ þ ðd1t Sizei;t þ d2t M=Bi;t þ d3t Levi;t þ d4t Di;t Sizei;t þ d5t Di M=Bi;t þ d6t Di Levi;t Þ þ ei;t ð4Þ To get company i’s differential timeliness parameter for a given year t, we estimate Model (4) for year t. Then we compute Consv_Cscoreit as: Consv Cscorei;t ¼ k1t þ k2t Sizei;t þ k3t M=Bi;t þ k4t Levi;t ; ð5Þ where k1t is the estimate of k1t from Model (4), Sizei,t is the size of company i in year t, and so on. Results are presented in Table 9. The coefficients of Consv_Cscore have the expected signs and are statistically significant in all four regressions except when explaining PENALTY. These additional results are consistent with our main results that more conservative firms are less likely to be sued and that increases in conservatism result in more favorable litigation outcomes. Thus, our main results are robust to using the Khan and Watt’s (2009) C-Score measure for conditional conservatism. 34 35 We also examine the association between conservatism measures and subsequent litigation outcomes using the ‘‘falsification’’ samples. We find no significant associations between conservatism measures and litigation outcomes. This result raises the question whether discretionary accruals actually proxy for earnings quality in Chalmers et al. (2012), or whether they proxy for other firm characteristics that are correlated both with discretionary accruals and litigation occurrence. Accounting Horizons Volume 30, Number 1, 2016 Ettredge, Huang, and Zhang 112 TABLE 9 Multivariate Regressions of the Effect of Khan and Watts’s (2009) C-Score on Litigation Occurrence and Outcomes Sued REACTION Variables Coeff. z-value Coeff. t-value Intercept CConsv UConsv DUVOL CRASH NCSKEW FPS LNASSETS SALES_GROWTH RETURN RETURN_SKEWNESS RETURN_STDDEV TURNOVER DEBT_ISSUE EQUITY_ISSUE MKT_SHR LNMAKTCAP LEVERAGE NONDISCTCONSV DISC_ACCRUAL BIGN CLASSPERIOD REACTION Year Indicators 0.266 2.247 0.009 0.069 0.132 0.167 0.008 0.063 0.226 0.293 0.088 0.074 0.157 0.243 0.251 0.158 0.185 0.004 0.206 0.428 0.276 0.23 2.13** 0.50 0.09 0.51 0.46 0.04 0.35 0.79 1.80* 0.77 0.05 3.16*** 1.35 0.23 0.68 0.85 0.07 0.34 0.79 1.26 0.153 0.162 0.000 0.059 0.017 0.011 0.002 0.003 0.006 0.024 0.025 0.090 0.002 0.020 0.143 0.024 0.016 0.004 0.044 0.033 0.007 0.000 1.65 2.05** 0.14 1.02 0.82 0.40 0.13 0.23 0.26 1.86* 2.56*** 0.72 0.68 1.34 1.61 1.28 0.84 0.76 0.85 0.83 0.38 2.73*** 2 Adjusted R Pseudo R2 ROC Statistics Chi-square Likelihood n Included 0.129 0.096 0.710 100.13 726 DURATION Coeff. t-value Included 1002.22 2.40 919.83 2.03** 1.357 0.79 19.80 0.08 57.49 0.60 34.69 0.28 123.33 1.82 85.82 1.24 187.08 1.81 145.08 2.28** 61.23 1.37 762.73 1.35 11.816 0.92 0.211 0.00 165.68 0.38 57.84 0.66 85.18 0.98 6.757 0.25 440.91 1.86* 78.60 0.42 119.11 1.48 0.035 0.54 321.48 1.29 Included 0.282 0.279 358 358 DISMISSAL Coeff. z-value 4.801 2.73 4.044 2.14** 0.016 0.87 0.625 0.55 0.734 1.90* 0.621 1.15 0.074 0.26 0.454 1.58 0.536 1.24 0.119 0.46 0.002 0.01 1.241 0.51 0.115 1.83* 0.016 0.06 0.014 0.01 0.333 0.93 0.725 2.03** 0.043 0.40 1.127 1.14 0.870 1.01 0.329 0.99 0.001 2.87*** 0.551 0.54 Included 0.149 0.101 0.716 57.672 358 PENALTY Coeff. t-value 0.043 0.80 0.040 0.86 0.017 1.61* 0.034 1.12 0.013 1.17 0.012 0.80 0.004 0.52 0.016 2.06** 0.004 0.31 0.014 2.16** 0.008 1.64 0.124 1.85* 0.003 1.32 0.003 0.28 0.000 0.00 0.010 1.00 0.012 1.23 0.002 0.59 0.009 0.35 0.062 2.32** 0.002 0.28 0.000 3.50** 0.047 1.74* Included 0.307 210 *, **, *** Represent significance at the 0.10, 0.05, and 0.01 levels, respectively, for two-tailed tests. See Appendix A for variable definitions. CONCLUSION While prior theory asserts that litigation risk engenders accounting conservatism as a preemptive defense, we know relatively little about whether and how this defense actually works. This paper investigates the role of accounting conservatism in the securities class action litigation process from the initiation (or non-initiation) of lawsuits to their resolutions. We use four measures to capture conditional accounting conservatism, including the relative sensitivity of earnings to bad news versus good news, negative non-operating accruals, the skewness of earnings, and a composite measure based on the average percentile ranks of the three individual conditional conservatism measures. We report a variety of evidence on the mitigating benefits of conservative financial reporting on the litigation occurrence and outcomes. We find that all four conditional conservatism metrics are negatively associated with subsequent lawsuit occurrence. An improvement in the average conservative rankings from the bottom to top percentile reduces the probability of a lawsuit by 30 percent. Given that firms are sued, market reactions to lawsuits are less negative for more conservative firms. Firms in the top percentile of the average rankings of conservatism experience less negative market abnormal returns by 930 basis points. Furthermore, their cases tend to be resolved sooner; and their cases are more likely to be dismissed by courts. An improvement in the average ranking of conservatism from the bottom percentile to the top is associated with a decrease in resolution window of around 385 days and an increase in the probability of a dismissal by 31 percent, respectively. Finally, even when their cases are settled rather than dismissed, more conservative firms face smaller monetary settlement amounts. On average, firms reporting conservatism in the top percentiles pay $145 million less penalty than those with the conservatism Accounting Horizons Volume 30, Number 1, 2016 Conservative Reporting and Securities Class Action Lawsuits 113 ranking in the bottom percentile. Taken together, our results are consistent with conditional accounting conservatism serving as an effective mechanism for deterring lawsuit occurrences and mitigating adverse outcomes in shareholder class action lawsuits alleging violations of GAAP. We view our results as providing new empirical evidence to support Watts’s (2003a) litigation explanation for accounting conservatism. REFERENCES Aier, J., L. Chen, and M. Pevzner. 2014. Debtholders’ demand for conservatism: Evidence from changes in directors’ fiduciary duties. Journal of Accounting Research 52 (5): 993–1027. doi: 10.1111/1475-679X.12062 Ahmed, A., B. Billings, R. Morton, and M. Stanford-Harris. 2002. The role of accounting conservatism in mitigating bondholdershareholder conflicts over dividend policy and in reducing debt costs. The Accounting Review 77: 867–890. Ahmed, A., and S. Duellman. 2007. Accounting conservatism and board of director characteristics: An empirical analysis. Journal of Accounting & Economics 43: 411–437. Bajaj, M., S. Mazumdar, and A. Sarin. 2003. Securities class action settlements. Santa Clara Law Review 43: 1001–1033. Balakrishnan, K., R. Watts, and L. Zuo. 2014. The Effect of Accounting Conservatism on Corporate Investment during the Global Financial Crisis. Available at: http://www.papers.ssrn.com/abstract¼1952722 Ball, R. 2013. Accounting informs investors and earnings management is rife: Two questionable views. Accounting Horizons 27: 847– 853. Ball, R., and L. Shivakumar. 2005. Earnings quality in U.K. private firms: Comparative loss recognition timeliness. Journal of Accounting & Economics 39: 83–128. Ball, R., S. Kothari, and A. Robin. 2000. The effect of international institutional factors on properties of accounting earnings. Journal of Accounting & Economics 29: 1–51. Ball, R., S. Jayaraman, and L. Shivakumar. 2012. Audited financial reporting and voluntary disclosure as complements: A test of the confirmation hypothesis. Journal of Accounting & Economics 53: 136–166. Ball, R., S. Kothari, and V. Nikolaev. 2013. On estimating conditional conservatism. The Accounting Review 88: 755–788. Basu, S. 1995. Conservatism and the Asymmetric Timeliness of Earnings. Ph.D. thesis, University of Rochester. Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics 24: 3–37. Basu, S. 2005. Discussion of ‘‘Conditional and unconditional conse...
Purchase answer to see full attachment
User generated content is uploaded by users for the purposes of learning and should be used following Studypool's honor code & terms of service.

Explanation & Answer

Thank you so much

Running head: conservative reporting & securities class action lawsuits

CONSERVATIVE REPORTING & SECURITIES CLASS ACTION LAWSUITS:
Name:
Institution affiliation:
Date:

1

Conservative reporting & securities class action lawsuits

2

Introduction
Companies have an obligation of creating value for shareholder’s investments. This goal
is usually achieved by ensuring that a company remains profitable not only in the short-term but
also in the long-term. However, the ever-changing market conditions may result in a decrease in
a company’s earnings, thereby prompting companies to overstate their profits in order to remain
attractive to potential investors. In so doing, companies usually find themselves facing
accounting –related class action lawsuits from their shareholders. In this paper, an article by
Ettredge et al. (2016) titled “Conservative Reporting and Securities Class Action Lawsui...


Anonymous
Great! Studypool always delivers quality work.

Studypool
4.7
Trustpilot
4.5
Sitejabber
4.4

Similar Content

Related Tags