Are Lengthy and Boilerplate Risk Factor Disclosures Inadequate? An Examination of Judicial and Regulatory Assessments of Risk Factor Language - PDF Document

Presentation Transcript

  1. Are Lengthy and Boilerplate Risk Factor Disclosures Inadequate? An Examination of Judicial and Regulatory Assessments of Risk Factor Language Richard A. Cazier University of Texas at El Paso racazier@utep.edu Jeff L. McMullin* Indiana University jemcmull@indiana.edu John Spencer Treu University of West Virginia john.treu@mail.wvu.edu April 2018 ABSTRACT Prior research finds that lengthy and boilerplate risk factor disclosures are associated with negative capital market consequences. Yet regulators and users of financial statements continue to criticize corporate risk factor disclosures as excessively long and boilerplate. We examine two factors that may influence firms’ incentives to disclose lengthy, boilerplate risk factor disclosures by examining how measures of disclosure length and disclosure boilerplate correlate with judicial and regulatory assessments of firms’ risk factor disclosures. Our results suggest that longer and more boilerplate risk factor disclosures are less likely to be flagged as inadequate under judicial and regulatory review. Specifically, we find that longer and more generic risk factor language is positively associated with favorable assessments for purposes of the Private Securities Litigation Reform Act’s safe harbor, and that standardized risk factor language is less likely to be targeted by an SEC comment letter during the SEC’s filing review process. JEL classification: D8; G38; M4 Keywords: risk disclosure; boilerplate; litigation risk; securities lawsuits; SEC comment letters; disclosure regulation We acknowledge helpful comments from John Campbell, Ted Christensen, Lauren Cooper, Todd Kravet, Josh Madsen, Rick Mergenthaler, Ken Merkley, and workshop participants at the 2017 BYU Accounting Research Symposium and at the University of Texas at El Paso. We also thank Shea Boothe, Jangho Gil, and Brandon Nicholas for valuable research assistance. * Corresponding Author.

  2. 1.Introduction The SEC mandates that firms disclose the most significant factors that make their stock speculative or risky in Item 1A of their periodic SEC filings (SEC 2005). Recent research suggests that firms with risk factor disclosures that are lengthy or more boilerplate (i.e., less firm-specific) experience negative capital market consequences, such as higher cost of capital, greater stock price volatility, weaker market responses, and declines in analysts’ ability to assess fundamental risk (e.g., Kravet and Muslu 2013, Campbell et al. 2014, Hope, Hu, and Lu 2016). Yet practitioners and regulators lament that firms’ risk factor disclosures continue to be generic and excessively long (e.g., Johnson 2010; IRRC 2016; SEC 2016; Berkman 2018). Existing research offers little explanation for the persistence of these disclosure practices in light of the potentially adverse capital market consequences. In this study, we consider two factors that may influence firms’ incentives to disclose lengthy and boilerplate risk factor disclosures. Specifically, we examine how measures of disclosure boilerplate and length correlate with judicial and regulatory assessments of risk factor disclosure adequacy. Risk factor disclosures are subject to regulatory assessment under the SEC’s filing review process, and to judicial assessment when federal judges consider whether a sued firm’s forward-looking statements should be granted safe harbor protections. The Private Securities Litigation Reform Act (PSLRA) of 1995 provides a statutory safe harbor for firms’ forward-looking statements accompanied by cautionary language describing the risks that could cause actual results to vary from projections. Thus, firms may be willing to accept the capital market consequences of providing boilerplate and/or lengthy risk factor disclosures if such disclosures reduce the likelihood of unfavorable regulatory or judicial outcomes. 1

  3. Official disclosure guidance suggests that lengthy and boilerplate risk factor disclosures may be disfavored by regulators and judges. The SEC’s Item 1A risk factor disclosure mandate indicates that risk factor disclosures should be concise and tailored to the specific risks faced by the firm (SEC 2005; SEC 2016).The Congressional Record accompanying the passage of the PSLRA indicates that boilerplate risk factor disclosures are not sufficient to receive the safe harbor protections and that only important factors should be listed (Cong. Rec. 28 Nov. 1995 13703).1 Despite such official guidance, boilerplate and lengthy risk factor disclosures may actually be associated with more favorable judicial and regulatory assessments for a number of reasons. First, firms may adopt peer firm disclosure language that has successfully passed judicial or regulatory review processes to increase the likelihood their own disclosures will be considered acceptable (Kahan and Klausner 1997). McMullin (2016) documents firms’ tendency to borrow disclosure language from industry peers that share the same auditor.2 This process may lead firms to adopt lengthier disclosure language that is not necessarily tailored to their specific risk profile, but which is more likely to be deemed adequate under judicial or regulatory review. This association may be self-reinforcing as regulators and judges use the disclosures of firms’ industry peers to set their own benchmark of what constitutes “adequate” or “normal” disclosure. 1The Congressional Record is the official daily record of the proceedings and debates occurring in the United States Congress (See https://www.congress.gov/congressional-record.). While the congressional record does not carry the weight of law, the record provides insight into the purpose for the enactment of laws and courts frequently rely upon the congressional record to interpret the law. 2 Large auditing firms provide disclosure consulting services which include template disclosures based on peer firm disclosures for their clients. We spoke with a director of audit and assurance at one of the big four accounting firms who stated his firm tags and databases disclosures from Item 1A risk factors as well as other sections of the 10-K to use as a reference for aiding client firms prepare their own disclosures. He indicates his firm also tags and databases the comments in SEC comment letters, which allows the firm to model the likelihood their client will receive an SEC comment letter based on the client’s disclosures. 2

  4. In contrast, disclosures dissimilar to those of industry peers may be more susceptible to greater regulatory or judicial examination and a greater likelihood of being flagged as inadequate. Second, firms’ inclination to adopt standardized risk factor language may be amplified by the legal doctrine of stare decisis (precedent), which leads judges to consider how closely a given set of risk factor disclosures resemble those already assessed as adequate in similar lawsuits. For example, a federal court granted a favorable ruling of risk factor disclosure adequacy based on the fact that the firm’s cautionary statements were “virtually identical to language approved by the Ninth Circuit in instances in which forward-looking statements were immunized by the PSLRA Safe Harbor.”3 Third, although the PSLRA requires risk factor disclosures to convey “substantive” firm-specific information about risks, the PSLRA does not outline how “substantive” risk disclosures are identified. Published judicial opinions suggest that disclosure length is sometimes relied on as a heuristic.4 Finally, broad risk factor language may provide greater legal protection if projected results fail to materialize, as risk factor disclosures that are more specific and concise inherently cover a narrower subset of adverse potential outcomes. Thus, in spite of official guidance and capital market costs documented in prior research, firms may continue to provide lengthy and boilerplate risk factor disclosures if these attributes are associated with more favorable judicial and regulatory outcomes. 3 In Re Fusion-IO, Inc. Securities litigation, 2015 U.S. Dist. LEXIS 18304; see also Plumbers & Pipefitters Local Union No. 630 Pension-Annuity Trust Fund v. Allscripts-Misys Healthcare Solutions, Inc., 778 F. Supp. 2d 858 (2011). 4 For instance, in dismissing a suit against Motorola’s allegedly false and misleading forward-looking statements, one judge reasoned that “the risk factors found in Motorola’s SEC 10-K filings are extensive and specific, extending beyond eight pages.” 3

  5. We examine how judicial and regulatory assessments of risk factor disclosure adequacy correlate with disclosure length and two distinct, but related measures of disclosure boilerplate used in prior research. Our measure of disclosure length is based on a word count of the firm’s Item 1A risk factor disclosures (Campbell et al. 2014). Our two boilerplate measures capture the extent to which language is generic, rather than firm-specific. The first measure, based on the Stanford Named Entity Recognition (NER) tool, measures language specificity by counting the number of specific entities in the disclosure (Hope et al. 2016). The second measure, based on the use of phrases (trigrams) that are very commonly found in risk factor disclosures, captures the extent to which risk factor language is so standardized that it is unlikely to contain important firm-specific information (Lang and Stice-Lawrence 2015). We construct two distinct samples to examine judicial and regulatory assessments of risk factor disclosures. The first consists of judicial decisions regarding firms’ motion to dismiss a securities lawsuit alleging false or misleading forward-looking statements between 2005 and 2015. We read each published judicial decision to determine the judge’s assessment of the defendant’s risk factor language as adequate or not for purposes of the safe harbor.5 To better understand the judicial assessment of risk factor disclosures and motivate our hypotheses, we also identify the reasons cited for why the risk factor disclosure was or was not adequate. We find significant variation in the rationale judges offer to support their assessments of risk factor disclosure adequacy, suggesting this determination is far from straightforward. For our second sample, we use Audit Analytics’ SEC Comment Letter database to identify firms that received an 5 To infer how attributes of firms’ risk factor disclosures influence judicial safe harbor determinations, we focus directly on judges’ explicit evaluation of firms’ risk factor language as our dependent variable. It is possible that risk factor language also has an impact on shareholders’ propensity to file the initial lawsuit alleging false or misleading forward-looking statements, but plaintiffs’ decision to file a lawsuit is a function of multiple inputs and is a much less direct indicator of the relationship between risk factor attributes and safe harbor protections. 4

  6. SEC comment letter identifying a deficiency in their risk factor disclosures. Our review of a sample of comment letters indicates the SEC comments on a variety of perceived deficiencies in firms’ risk factor disclosures. Our multivariate regression analyses suggest that the probability a federal judge rules that a firm’s risk factor disclosures are inadequate for safe harbor protection is decreasing in disclosure length and each measure of boilerplate. Including both measures of boilerplate in the model reveals only length and the measure of boilerplate based on Stanford’s NER tool remain statistically significant. Together, our results suggest that, conditional on a firm being sued, risk factor disclosures are less likely to be assessed as inadequate under judicial review if they are longer, more generic and less firm-specific. Results are robust to controlling for determinants of firm risk, industry affiliation, fiscal year, the U.S. circuit in which the firm is headquartered, and the topics discussed in the firm’s risk factor disclosures. Turning to the SEC comment letter sample, we find a strong negative association between our measure of boilerplate language based on standardized language and the probability of receiving an SEC comment letter related to risk factor disclosures. We do not find that the probability of receiving an SEC comment letter significantly correlates with disclosure length or our measure of boilerplate based on Stanford’s NER tool. The negative association between standardized risk factor language and the probability of receiving an SEC comment letter is consistent with the notion that more firm-specific risk factor language may subject firms’ disclosures to greater regulatory scrutiny. Combined with our findings relating to judicial assessments of risk factor disclosures, these results suggest that one reason firms may be willing to bear the negative capital market consequences of lengthy and non-specific risk factor disclosures is that these attributes are 5

  7. associated with more favorable judicial and regulatory assessments of disclosure adequacy. We run various robustness tests that support these main findings. For example, our main results are robust to alternative approaches to measuring boilerplate and length and to controlling for the specific topics disclosed in the risk factor section. Our paper contributes directly to the stream of research examining the quality of firms’ risk factor disclosures. Whereas recent research focuses on whether risk factor disclosures contain anyinformation content, our study is among the first to propose and test an explanation for why lengthy, non-specific risk factor disclosure continues to persist despite the potential for adverse capital market consequences. Our results suggest that lengthy, non-specific risk factor disclosures may help minimize expected legal and regulatory costs arising from judicial and regulatory enforcement. These findings also have direct implications for regulators and policy makers seeking to understand the forces that shape current risk factor disclosure practices. The evidence we find suggests that one reason firms continue to provide lengthy, standardized, and generic risk factor language is that regulatory and legal outcomes generally do not reward concise, firm-specific disclosures. This also suggests that judges and the SEC may assess the adequacy of risk factor disclosures in a manner that is inconsistent with official guidance. 2.Institutional Background and Hypothesis Development 2.1CAPITAL MARKET CONSEQUENCES OF RISK FACTOR DISCLOSURE LENGTH AND BOILERPLATE Several recent studies have examined capital market outcomes associated with firms’ risk factor disclosures. Kravet and Muslu (2013) find that increases in risk disclosure length is associated with subsequent increases in stock return volatility and trading volume and conclude that longer risk disclosures increase investors’ perception of firm risk. Campbell et 6

  8. al. (2014) find that increases in risk factor disclosures increase firms’ cost of capital, and suggest that this relation may provide strong incentives for managers to resist disclosure. Hope et al. (2016) find that more specific disclosures are associated with stronger market responses and with greater analyst ability to assess fundamental risk, and concurrent research finds that risk factor disclosures can predict firm-specific adverse events (Gaulin 2017). Overall, these studies suggest that risk factor disclosures that are less specific (i.e., more boilerplate) and more lengthy can subject firms to negative capital market consequences. Despite the benefits of concise and specific risk disclosure documented by prior research, critics allege that risk factor disclosures continue to be excessively boilerplate and lengthy (e.g. Johnson 2010; IRRC 2016; SEC 2016; Berkman 2018). Consistent with critics’ concerns about the informativeness of risk factor disclosures, Beatty, Cheng, and Zhang (2018) report that the market’s response to unexpected 10-K risk factor disclosures has decreased in recent years. However, existing research sheds little light on the source of firms’ preferences for lengthy and boilerplate risk factor disclosures that could explain the persistence of these attributes. We contribute to this literature by examining how risk factor disclosure attributes impact the likelihood of an adverse regulatory or legal intervention. Specifically, we examine how length and firm-specificity correlate with the likelihood regulators and federal judges will flag risk factor disclosures as inadequate. 2.2CAUTIONARY LANGUAGE AND THE PSLRA’S SAFE HARBOR Risk factor disclosures play an important role in providing legal protections to firms making forward-looking statements. In response to concerns that securities lawsuits were discouraging managers from providing useful forward-looking information to investors, Congress enacted a safe harbor provision to protect forward-looking statements as part of the 7

  9. Private Securities Litigation Reform Act of 1995 (PSLRA) (15 U.S.C. § 78u5(c)).6 In order to receive the full protection of the PSLRA safe harbor, firms must accompany forward looking statements with meaningful cautionary language and claims brought by shareholders for such disclosures are preliminarily dismissed.7 Congress did not explicitly define what makes cautionary language “meaningful”, though the statute indicates that cautionary language must identify important risks that could cause actual results to vary from projections. The Congressional Conference Report accompanying the PSLRA asserted that risk factor disclosures should focus on specific risks, that boilerplate cautionary language would not suffice for purposes of the statute’s safe harbor, and that only important factors should be listed (Cong. Rec. 28 Nov. 1995 13703). After the SEC began mandating the disclosure of firms’ principal risks as Item 1A in their periodic filings starting in 2005, Item 1A of periodic filings became the primary location for much of the cautionary language firms use to invoke safe harbor protection for their forward-looking statements (Nelson and Pritchard 2016). Ultimately, whether a firm’s risk factor disclosures are deemed as adequate cautionary language depends on judicial interpretation and application of the statutory safe 6 While the PSLRA codified a safe harbor protecting forward looking statements, even prior to the PSLRA a common law rule called the Bespeaks Caution Doctrine protected projections and estimates accompanied by meaningful cautionary language because misrepresented statements that were accompanied by such warnings were deemed to be immaterial. In re Trump Casino Sec. Litig., 7 F.3d 357, 371 (3d Cir. 1993); Harris v. IVAX Corp., 998 F. Supp. 1449, 1454 (S.D. Fla. 1998). 7 The safe harbor of the PSLRA has two prongs and the first protects forward looking statements that are immaterial and statements accompanied by “meaningful cautionary statements identifying important factors that could cause actual results to differ materially” (15 U.S.C. § 78u-5(c)(1)(A)). Claims based on statements covered by the first prong are typically dismissed by courts on preliminary motion without inquiry into the state of mind of the firm that made the statement. In the absence of such cautionary statements, lesser protections may apply under the second prong that permits an inquiry into the state of mind of the firm making the statement (15 U.S.C. § 78u-5(c)(1)(B)). The second prong protections are more similar to the weaker general PSLRA protections for non-forward looking statements and potentially opens the firm up to costly discovery proceedings. 8

  10. harbor provisions (e.g., Olazábal 2000; De Simone, Ingber, and Creutz 2004; Pritchard and Sale 2005; Cornerstone 2016). In order to gain institutional insight into the factors that judges claim influence their assessments of risk factor disclosure adequacy and to guide the development of our hypotheses, we read over 500 hundred judicial opinions published between 1996 and 2015 relating to defendants’ request to dismiss securities lawsuits which allege false or misleading forward- looking statements. We find that in over 400 of these decisions, the judge makes an explicit ruling over the firms’ cautionary language as either adequate or inadequate to avail the firm’s forward-looking statements of safe harbor protection. Next, we categorize the judges’ rationale and tabulate in Table 1 the reasons cited by judges to support their rulings. Panel A indicates that the extent to which risk factor language is boilerplate versus firm- specific is commonly cited as a factor influencing judicial assessments of risk factor disclosure adequacy. However, consistent with similar views expressed by legal scholars (Bloomenthal and Wolff 2012), our reading of the underlying judicial opinions suggests many of these assessments are largely ad hoc and are not based on any standard, objective benchmark of what constitutes sufficient specificity.8 By far the most common factor cited by judges ruling risk factors are adequate is that the disclosure simply fulfils the requirement to warn investors of risks that could cause actual results to vary. The most commonly cited reason for why risk factor language is inadequate is the firm’s omission of a material risk factor. Because specific risk factor disclosures inherently cover a narrower subset of potential adverse outcomes, it is possible that 8Bloomenthal and Wolff (2012) questioned whether the 7th circuit court correctly identified boilerplate language in determining the adequacy of risk factor disclosures in the case Asher v. Baxter International, Inc., 377 F.3d 717 (7th Cir. 2004) because such disclosures, although deemed adequate by the court, appeared to “cover as many conceivable generic factors that could relate to the company’s business as possible.” 9

  11. more generic risk factor disclosures provide superior legal protection. These considerations suggest that even generic risk factor disclosures may be adequate to avail firms of safe harbor protection. Consistent with the notion that broad risk factors provide superior legal protection, one partner at a national law firm recently asserted that all 10-Ks contain boilerplate risk factor language to “make it as all-encompassing as possible,” (see Berkman 2018). This incentive may help explain Hope et al.’s (2016) finding that Item 1A risk factor disclosures are less specific than other sections of the 10-K. Thus, despite legislative guidance that risk factors should not be boilerplate, the actual association between firm-specific risk factor disclosure and judicial assessments of adequacy is unclear. We formalize the tension in the relationship between firm- specificity and the likelihood risk factors are flagged as inadequate for purposes of the PSLRA’s safe harbor as our first hypothesis, stated in null form as follows: H1a: There is no association between risk factor disclosure firm-specificityand the likelihood risk factor disclosures will be deemed inadequate for purposes of the PSLRA’s safe harbor. The PSLRA and its supporting legislative guidance indicate that only important risk factor disclosures entitle firms to safe harbor protections. However, there are a number of reasons that longer risk factor disclosures may provide superior legal protection. First, judges may view lengthier risk factor disclosures as better fulfilling the PSLRA’s requirement that risk factor disclosures convey “substantive” information about the risks that could cause actual results to vary. Panel A of Table 1 indicates that judges sometimes cite the length of a firm’s risk factor disclosures as supporting evidence for their assertion that the firm’s cautionary language was adequate. Second, uncertainty regarding which risks are most important may lead assessors of risk disclosure adequacy to view lengthier risk factor disclosures as representing a good faith 10

  12. effort to warn of all potentially material risks the firm may face. Third, the inclusion of a longer list of risk factor disclosures helps ensure no obvious omissions have been made. Given that the congressional record and some evidence from our reading of judicial opinions may be at odds with one another, we formalize our next hypothesis in null form as follows: H1b: There is no association between risk factor disclosure lengthand the likelihood risk factor disclosures will be deemed inadequate for purposes of the PSLRA’s safe harbor. 2.3THE SEC’S ITEM 1ARISK FACTOR DISCLOSURE MANDATE In 2005, the SEC began mandating disclosure of firms’ most significant risk factors in Item 1A of firms’ periodic filings to help investors understand the nature of the risk inherent in the company. Prior to 2005, any cautionary language that firms provided for purposes of the PSLRA’s safe harbor was not standardized or consistently reported in any particular section of firms’ periodic filings. SEC reporting requirements stipulate that Item 1A risk factor disclosures should avoid vague boilerplate (17 C.F.R. 230.421(b)) and be clear, concise, and tailored to the specific risks facing the reporting firm (e.g., SEC 2004; SEC 2005). The SEC reviews Item 1A risk factor disclosures, along with firms’ other periodic filings, as part of its filing review process. Firms whose risk factor disclosures are non-compliant with disclosure guidelines may receive SEC comment letters asking them to rectify the perceived deficiency. The SEC states it “concentrates its review resources on disclosures that appear to be inconsistent with Commission rules or applicable accounting standards, or that appear to be materially deficient in their rationale or clarity.” The SEC Division of Corporate Finance’s 11 11

  13. offices possess industry-specific accounting and disclosure expertise and manage the review of filings of firms in these industries.9 The SEC’s review of Item 1A risk factor disclosures varies in important respects from judicial review of risk factor disclosures to determine disclosure adequacy. Perhaps most importantly, the SEC is under no obligation to explicitly rule on the adequacy of risk factor disclosures that it reviews. Rather, the SEC only sends comment letters relating to risk factor disclosures if it identifies a perceived deficiency. Thus, no observable documentation exists regarding the SEC’s assessments when risk factor disclosures are deemed to be adequate for regulatory compliance. To identify the specific disclosure deficiencies the SEC asks firms to rectify and to guide the development of our hypotheses, we read a randomly selected sample of 200 SEC comment letters identified by Audit Analytics as relating to firms’ risk factor disclosures. We place the SEC’s comments into one of six separate categories and tabulate them as Panel B in Table 1. Because the SEC may have multiple comments relating to risk factor disclosures within a single comment letter, the total number of comments tabulated in Panel B exceeds 200. Panel B indicates that the most common SEC comments relating to risk factor disclosures request additional specificity or additional clarity relating to risk factors that are disclosed. In 57 of the comment letters, the firm was asked to add an additional risk factor. In only 7 cases (3.5 percent of the comment letters examined) did the SEC ask the firm to remove an existing risk factor.10 Thus, how disclosure length and boilerplate language relate to the probability risk factors become targeted by the SEC is an open empirical question. 9 See https://www.sec.gov/divisions/corpfin/cffilingreview.htm (accessed on March 20, 2018). 10 One caveat to note is that Panel B sheds light on the issues the SEC raises conditional on the SEC deciding to issue a comment letter relating to a firm’s risk factor disclosures. That is, while Panel B summarizes comments 12

  14. Despite regulatory admonitions for firms to tailor their risk factor disclosures to be firm- specific, it is possible that risk factor disclosures that mirror those of industry peers may be less likely to be flagged as inadequate under regulatory review. Recent research suggests that firms preparing their own disclosures often look to peer firm disclosures to form their own expectation of what constitutes adequate disclosure, and that borrowing language from industry peers’ disclosures is common (e.g., McMullin 2016, Berkman 2018; see also Brown, Tian, and Tucker 2017). Firms have incentives to adopt industry peers’ disclosure language that has been vetted by successfully passing through regulatory review. This process is likely facilitated by large audit firms’ statutory reporting and disclosure assistance services, which provide financial reporting templates for their clients based on their knowledge of the types of disclosures that are likely to be flagged as inadequate by regulators. Firms may thus gravitate toward risk factor language that is seen as “safe” in lieu of language that is specifically tailored to the actual risks faced by the firm. This process may lead to language that is more generic and boilerplate having a lower likelihood of being flagged for further scrutiny under regulatory review. Thus, despite the SEC’s formal guidance that risk factor disclosures should be firm- specific, it is possible that disclosures that look more similar to those of industry peers are less likely to attract regulatory attention. We formalize the tension in the relationship between boilerplate (i.e., not firm-specific) risk factor disclosures and the likelihood of an SEC comment letter as our next hypothesis, stated in null form as follows: H2a: There is no association between risk factor disclosure boilerplate and the likelihood risk factor disclosures will be deemed inadequate under the SEC’s filing review process. made by the SEC about a firm’s risk factors, we cannot conclude whether these issues are the factors that initially attracted greater scrutiny to these disclosures in the first place. 13

  15. Although SEC guidance cautions firms to make risk factors concise, lengthier risk factor disclosure may decrease the likelihood regulators will observe an obvious omission that could spawn further scrutiny. In addition, the descriptive evidence in Panel B of Table 1 suggests the SEC is much more likely to request a firm to lengthen its risk factor disclosures rather than remove an existing factor. This evidence suggests that, despite the SEC’s formal guidance that risk factors should be concise, firms may be less likely to have their existing risk factor disclosures targeted by an SEC comment letter if those risk disclosures are lengthy. We formulate our final hypotheses in null form as follows: H2b: There is no association between risk factor disclosure length and the likelihood risk factor disclosures will be deemed inadequate under the SEC’s filing review process. 3.Sample Construction Our analysis of judicial and regulatory assessments of firms’ risk factor disclosures requires the construction of two distinct samples. The first sample consists of firms whose risk factor disclosures were assessed for adequacy by judges making safe harbor determinations during the dismissal phase of securities class action lawsuits. To do this, we search the Lexis Advance legal database for all published judicial opinions between 2006 and 2015 which relate to the motion to dismiss securities litigation alleging false or misleading forward-looking statements. Keyword searches retrieved an initial judicial opinion sample for 328 securities lawsuits that we were able to match with lawsuit data from Stanford’s Securities Class Action Clearinghouse database.11 11 Our initial search includes securities cases between 1996 and 2015 with judicial opinions that use the following phrases: “risk factor,” “cautionary language,” and “safe harbor.” This search returned over 500 cases. This sample is reduced to 328 align the time period with the presence of Item 1A (2006-2015). 14

  16. We eliminate several cases due to missing Compustat or CRSP data and cases where either the judge determined the statements were not forward-looking or failed to explicitly rule on the adequacy of the cautionary language. Our procedure for selecting this sample is detailed in Panel A of Table 2. Ultimately, our regression analyses examining judicial assessments of risk factor disclosure are based on a sample of 144 distinct lawsuits in which the judge explicitly rules regarding the adequacy of the firm’s cautionary language. We measure the judicial assessment of risk factor disclosure adequacy for each lawsuit and the boilerplate for the Item 1A Risk Factor disclosures in the firm’s 10-K filed during the lawsuit’s class period.12 Because the class period of some lawsuits includes more than one year, our final sample for our main analysis of judicial assessments includes 231 unique firm-years corresponding to 231 unique 10-K filings made during the class period. Our second sample consists of firm-years in which a firm received an SEC comment letter relating to at least one Item 1A risk factor disclosure between 2005 and 2015. We use Audit Analytics’ SEC Comment Letter database to construct this sample. As illustrated in Panel B of Table 2, we begin with an initial sample of 4,450 firm-years for which Audit Analytics indicates the corresponding SEC filings were the target of a comment letter relating to risk factors. After eliminating firm-years for which Item 1A Risk Factor disclosures were not retrievable from the SEC website, firm-years for which textual analysis variables could not be computed from the extracted Item 1A disclosure, and firm-years missing necessary CRSP or Compustat data, we are left with a final sample of 1,607 firm-years for which risk factor disclosures were the target of an SEC comment letter in the following year. We use this same 12 In some cases, the class period did not include a 10-K filing date. In these cases, we selected the 10-K filed immediately prior to the start of the class period. 15

  17. dataset to identify firm-years in which a firm did not receive an SEC comment letter relating to risk factor disclosures to form a control sample. After identifying these firms, our final sample for the SEC comment letter analysis includes a total of 33,725 firm-years. 4.Variable Measurement 4.1MEASURING JUDICIAL AND REGULATORY ASSESSMENT OF RISK FACTOR INADEQUACY Hypotheses 1a and 1b center on the likelihood risk factor disclosures will be ruled inadequate under judicial review as a function of risk factor disclosure length and the extent to which that language is firm-specific. To identify risk factor disclosures deemed inadequate under judicial review, we search LexisNexis’ legal database for judicial opinions relating to dismissal decisions for shareholder lawsuits alleging false or misleading forward-looking statements. We examine each opinion to identify the judge’s assessment of whether the firm’s cautionary language was sufficient to avail itself of the PSLRA’s safe harbor. We set the indicator variable RF_INADEQUATE_JG equal to one for firm-years in class periods in which the cautionary language was ultimately deemed inadequate by the judge, and zero for firm-years in class periods for which the judge ruled the cautionary language adequate. Hypotheses 2a and 2b consider the probability risk factor disclosures will be flagged as inadequate under regulatory review as a function of risk factor disclosure length and firm- specificity. We use the SEC’s decision to issue a risk factor-related comment letter as a proxy for the SEC’s assessment that the firm’s risk factor disclosure was inadequate. We set the indicator variable RF_INADEQUATE_SEC equal to one if a firm-year’s risk factor disclosures were targeted by an SEC comment letter, and zero otherwise. We use Audit Analytics’ Comment Letter database to identify firms that received comment letters relating to their risk factor disclosures in the 10-K filing for any given year. 16

  18. Our objective in testing our hypotheses is to capture the association between risk factor disclosure boilerplate and judicial or regulatory assessments of risk factor disclosure adequacy, holding constantother aspects of the firm’s risk factor disclosure, the firm’s underlying risks, industry affiliation, headquarter location, and fiscal year. We next discuss our, measure of disclosure length, two measures of boilerplate disclosure, and the construction of each of our control variables. 4.2MEASURES OF DISCLOSURE LENGTH AND BOILERPLATE We measure Item 1A risk factor disclosure length based on the total number of words, consistent with how disclosure volume is measured by a host of prior studies (e.g., Li 2008; You and Zhang 2009; Miller 2010; Lawrence 2013; Campbell et al 2014). LENGTH is the natural logarithm of the number of words in the Item 1A risk factor disclosure in the firm-year’s 10-K. Because we are interested in controlling for the influence of length holding other determinants of disclosure constant, we benchmark this measure against that of each firm’s most similar industry peer as follows: First, we calculate the similarity between all firms in the same two-digit SIC industry and identify a firm’s peer as the firm with the most similar 10-K Item 1 Business Description disclosure. To measure similarity, we first construct a topic model using Latent Dirichlet Allocation with 200 topics over all firms’ Item 1 and calculate the topic probability vector for each firm (Dyer et al. 2017). Our similarity metric is in the same spirit as Hoberg and Phillips (2016), except that we calculate the cosine similarity score between two firms’ topic probability vectors rather than between vectors indicating word presence. The motivation for using the LDA topic model relaxes the assumption made by Hoberg and Phillips (2016) that each word is it unique topic. Second, for each firm in our sample, we subtract the natural logarithm of the number of words of the most similar peer to obtain a measure of abnormal disclosure length. 17

  19. We next construct measures of boilerplate language, which by definition is generic and not firm-specific. Prior literature has examined two measures of boilerplate that reflect the extent to which firms’ disclosures do not convey firm-specific information. The first is based on the Stanford Named Entity Recognition (NER) tool that determines the extent to which specific entities are discussed in the disclosure (e.g., Hope et al. 2016). This measure captures the extent to which disclosures are specific rather than boilerplate based on the inherent precision of the language. Following Hope et al. (2016), we use a computing algorithm to identify words that specifically describe an entity (name or numeric expression), and count the number of specific entities described in a firm’s risk factor section.13 We scale this word count by the total number of words in the risk factor section to adjust for differences in the length of this disclosure across firms. We label this measure BP_NER.Because we are interested in examining the effects of boilerplate after controlling for determinants of the underlying disclosure, we also benchmark this variable against the most similar industry peer. The second measure of boilerplate captures the extent to which disclosure language is so standardized that it is unlikely to be informative (e.g., Lang and Stice-Lawrence 2015). To capture firms’ use of boilerplate language in their risk factor disclosure, we first identify commonly used trigrams (3-word phrases) in risk factor disclosure sentences for all firms in the same two-digit SIC industry.14 Next, we identify boilerplate sentences as those that either use 10 or more of these commonly used trigrams or for which 10 percent or more of the trigrams in the sentence are commonly used. We then count the number of words in these boilerplate sentences and divide this 13 We use Python’s natural language tool kit (NLTK) package (http://www.nltk.org) to implement Stanford’s Named Entity Recognition algorithm (Lafferty et al. 2001, Finkel et al. 2005). This algorithm identifies specific references to entities in seven categories: person, organization, location, money, date, time, and percent. Hope et al. (2016) explain in detail the implementation and assess the construct validity of this measure. 14 We define commonly-used trigrams are those that appear in at least 10% and not more than 90% of the risk factor disclosures produced by firms in the same SIC 2-digit industry. 18

  20. number by the total number of words in the risk factor section. Our approach to measuring this variable, which we label BP_STANDARD, is similar to the approached used by Lang and Stice- Lawrence (2015) to measure boilerplate in annual reports of non-US firms. Untabulated analyses indicate that BP_NER and BOILERPLATE_ STANDARDIZED have a 22 percent correlation. Thus, these two variables capture related but distinct dimensions of firms’ risk factor disclosures. We expect BP_NER will correlate negatively with unfavorable assessments of risk factor disclosure to the extent that inherently generic risk factor disclosures sufficiently cover a broader set of potentially adverse outcomes. We expect BP_STANDARD will negatively correlate with unfavorable assessments of risk factor disclosure to the extent that adjudicators are less likely to flag disclosure practices that are common in the industry as inadequate. 4.3CONTROL VARIABLES We include several control variables that may confound the relationship between risk factor length, boilerplate, and adjudicators’ assessments of risk factor disclosure adequacy. First, we include the correlation of firms’ stock returns and industry peer firm stock returns to control for the fact that firms with similar underlying economic events may adopt similar language to describe for those events. RET_COR is the mean correlation between each firm’s 24-month return ending at the end of the current fiscal year and that of its 2-digit SIC industry peers. We also construct several measures of firm risk that may influence firms’ risk factor disclosures as well as adjudicators’ assessments of disclosure adequacy. RET_VOL is the standard deviation of daily returns for the year ending two days prior to the 10-K filing date. BETA is the firm’s market beta measured using daily returns for the one year ending two trading days prior to the 10-K filing date. SIZE is the natural logarithm of the market value of equity. 19

  21. LEV is total liabilities scaled by total assets, measured as of the end of the fiscal year. BIG_N is an indicator variable equal to one if the firm is audited by one of the largest auditors (the Big N), and zero otherwise. SKEW is measured as the negative coefficient of skewness for daily returns over the year ending two days prior to the 10-K filing date. TURNOVER is the mean number of shares traded during the fiscal year scaled by the number of common shares outstanding. ETR is the firm’s effective tax rate, computed as tax expense scaled by pre-tax income. ROE is income before extraordinary items scaled by the market value of equity. We also control for a host of other factors that may influence the relationship between disclosure boilerplate and adjudicator assessments of adequacy. We include various fixed effects to control for industry affiliation, fiscal year, and the U.S. circuit in which each firm is headquartered because research indicates judicial practices and litigation risk sometimes vary by circuit (Hopkins 2018; Cazier, Christensen, Merkley, Treu 2018).15 We note that over-controlling for the underlying content of the disclosure may wash out some of the effect that we are seeking to measure in the relationship between disclosure boilerplate and assessments of adequacy. However, in some specifications we also include controls for the actual topics discussed in the firm’s risk factor disclosures. We estimate a topic model using a Bayesian machine-learning approach proposed by Blei et al. (2003).16 Specifically, we split each risk factor section into paragraphs, remove stop words, estimate an LDA topic model with 200 topics, and calculate a topic probability vector (TPV) for each risk factor paragraph. We identify the topic for each paragraph as the topic with the highest 15 We measure industry affiliation at the 2-digit SIC level for the SEC comment letter analysis. Because of the relatively small sample size in the lawsuit sample, we measure industry affiliation at the Fama-French 12 industry classification level for those tests to maintain a practical number of degrees of freedom (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/det_12_ind_port.html). 16 Blei (2012) provides an overview of topic models. Also, prior accounting studies have estimated topic models for 10-Ks (Dyer et al. 2017), analyst reports (Huang et al. 2016), and risk factor disclosures (Campbell et al. 2014). 20

  22. probability and create 200 indicator variables that equal one if the topic was disclosed in that firm’s risk factor disclosures, and zero otherwise. To preserve degrees of freedom, we replicate this analysis using only 20 topics when controlling for topic inclusion in the lawsuit analysis. Equation 1 below shows the form of our primary regression model: Pr($%&'()*+,-&.&&/&&/0)&12)0/34)+/%,+) = β0 + β1LENGTH%,+ + β2BP_NER%,+ + β3BP_STANDARD%,+ + β4RET_COR%,+ + β5RET_VOL%,+ + β6BETA%,+ + β7SIZE%,+ + β8LEV%,+ + β9BIG_N%,+ + β10SKEWNEWSS%,+ + β11TURNOVER%,+ + β12ETR%,+ + β13ROE%,+ + Industry controls + Circuit controls + Year controls + Topic controls + ε%,+ (1) 4.4DESCRIPTIVE STATISTICS Table 3 presents sample means for each of our independent variables of interest. Panel A displays the descriptive statistics for our sample of firms sued for false or misleading forward- looking statements. Panel B displays the descriptive statistics for firm-years in our comment letter analysis. We note that our variables do not appear to exhibit extreme skewness, as the mean for each continuous variable is within the inner-quartile range in nearly every case. The one exception is in the sample of firms used for the SEC comment letter analysis, which has a mean ROE lower than the first quartile, suggesting significant skewness in that variable. To mitigate the effects of extreme outliers, all continuous variables are winsorized at the 1st and 99th percentiles. 5.Empirical Results 5.1JUDICIAL ASSESSMENT OF RISK FACTOR DISCLOSURE ADEQUACY Hypotheses 1a and 1b focus on the judicial assessment of risk factor disclosure adequacy as a function of risk factor disclosure boilerplate and length. To substantiate the notion that judicial assessments of risk factor disclosure inadequacy are associated with costly judicial outcomes and to validate our coding of RF_INADEQUATE_JG, we first estimate a logistic 21

  23. regression of the likelihood of lawsuit dismissal as a function RF_INADEQUATE_JG and control variables. Column 1 of Table 4 displays results from estimating this regression over the entire sample period 1996 to 2015. Column 2 of Table 4 re-estimates this regression only on post-2005 observations for which we have Item 1A data. Both columns report a significantly negative coefficient on RF_INADEQUATE_JG (p < 0.01), confirming that the ruling that risk factor disclosures are inadequate for purposes of the safe harbor has significant negative consequences for defendant firms. Results from modeling judicial assessments of risk factor disclosure adequacy as a function of Item 1A boilerplate and length are reported in Table 5. In Column 1 of Table 5, we present results from including only the first measure of boilerplate (BP_NER). In this specification, we find a significantly negative (p < 0.05) coefficient estimate on LENGTH, suggesting lengthier risk factor disclosures are less likely to be ruled inadequate for purposes of the PSLRA’s safe harbor. We also find a negative and strongly significant (p < 0.01) coefficient estimate on BP_NER, suggesting that risk factor disclosures containing less specific language are less likely to be assessed as inadequate by judges for purposes of safe harbor protections. Column 2 reports results when we only include our second measure of boilerplate BP_STANDARD. In this specification, the coefficient on this measure of boilerplate is insignificant. Column 3 of Table 5 indicates that when we include both measures of boilerplate, the coefficient estimate on BP_NER remains significantly negative (p < 0.05). Column 4 presents results after including indicator variables identifying the topics discussed in the firm’s risk factor disclosure. Results indicate that even after controlling for disclosure content, BP_NER is still significantly negatively associated with the likelihood risk factor language is deemed inadequate 22

  24. for safe harbor purposes.17LENGTH continues to be negatively associated with the likelihood risk factors are ruled inadequate, though the inclusion of the specific topic indicators attenuates the statistical significance of this relationship. Results in Table 5 suggest that, despite the caution expressed in the congressional record and by courts that risk factor disclosure should be specific and avoid boilerplate, more specific risk factor disclosure does not lead to better judicial outcomes. These results are consistent with the view that lengthy and generic risk factor disclosure may provide greater legal protection by covering a broader set of potentially adverse events and, as occasionally expressed in judicial opinions, that relatively generic language can meet the requirements of the statute. To provide a sense of the economic magnitude of our results, we compute the mean marginal effects for the two explanatory variables of interest that load significantly in Model 4 of Table 5. Untabulated marginal effects suggest that going from the first quartile to the third quartile of LENGTH decreases the probability risk factor disclosures are ruled as inadequate under judicial review by 7.6 percent. The probability a firm’s risk factor disclosures are ruled inadequate under judicial review decreases by 13.7 percent as those disclosures move from the first to the third quartile of BP_NER. 5.2REGULATORY ASSESSMENT OF RISK FACTOR DISCLOSURE ADEQUACY Panel A of Table 6 presents results from estimating equation 1 on our sample of firms for our SEC comment letter analysis. We replace the dependent variable with RF_INADEQUATE_SEC, which is an indicator variable equal to one in firm-years if the firm 17 Stanford’s NER tool identifies entities in seven categories, three of which refer to proper nouns (persons, organizations, and locations), and four of which refer to numeric values (time, money, percent, and date). In untabulated analyses, we split the BP_NER variable into the portion made of proper nouns versus numeric values and find that the results in Table 5 appear to be driven primarily by a lack of specificity relating to proper nouns. 23

  25. receives an SEC comment letter relating to their Item 1A risk factor disclosures. The first column of Panel A presents results from including only the first boilerplate measure, BP_NER, in the regression model. Results indicate that the coefficient estimate on BP_NER is negative but not statistically significant. We do not find LENGTH to be significantly correlated with the likelihood of an SEC comment letter. Column 2 presents results from replacing that measure of boilerplate with BP_STANDARD. We find the coefficient estimate on BP_STANDARD is negative and strongly significant (p < 0.01). Results from including both measures of boilerplate into the model are displayed in Column 3, which indicates that BP_STANDARD continues to be strongly negatively associated with the likelihood that risk factor disclosures are the target of an SEC comment letter. In Column 4, we display results from re-estimating the model reported in Column 3 after also including controls for the topics discussed in firms’ risk factor disclosures. Results continue to indicate that RF_INADEQUATE_SEC has a strong negative statistical association with BP_STANDARD, but not with BP_NER or LENGTH. An analysis of marginal effects suggests that the association between standardized language (BP_STANDARD)and the probability of an SEC comment letter targeting risk factor disclosures is relatively modest. Going from the first quartile to the third quartile of BP_STANDARD from Model 4 of Panel A in Table 6 decreases the probability of an SEC letter by only approximately 0.5 percent. However, results from Table 6 do suggest that, despite the SEC’s frequent admonition to tailor risk factor disclosures to the specific risk factors faced by the firm, risk factor language that more closely resembles that of industry peers is less likely to be flagged as inadequate under regulatory review. 24

  26. 5.3ROBUSTNESS TESTS One potential limitation of the sample examined in Panel A of Table 6 is that not all firms undergo the SEC filing review process each year. The SEC reviews filings of every publicly traded firm at least every three years, with some firms’ filings being reviewed more frequently. However, the SEC does not reveal its formula for determining which firms to review more frequently, and in the absence of a comment letter, we cannot discern which firms were under review each year. To test the sensitivity of the results in Panel A of Table 6, we rerun our estimation only on firms that received at least one SEC comment letter during the year to ensure our sample includes only firms whose filings were actually under review. We tabulate results from this sample, which includes 13,419 firm-year observations, in Panel B of Table 6. We find the association between BP_STANDARD and RF_INADEQUATE_JG to be qualitatively and quantitatively similar to those reported in Panel A, providing further support for the view that the likelihood risk factor disclosures are targeted by an SEC comment letter is decreasing in disclosure boilerplate. We continue to find no association between LENGTH and the likelihood of an SEC comment letter. A second potential limitation of the results in Panel A of Table 6 is that we rely on Audit Analytics’ comment letter database to identify comment letters relating to Item 1A risk factor disclosures. Our reading of a significant subsample of the SEC comment letters relating to risk factors as identified by Audit Analytics indicates that the majority (approximately 75%) relate to Item 1A risk factor disclosures. However, we find that a significant minority of SEC comment letters that Audit Analytics indicates as relating to “risk factors” do not directly relate to Item 1A Risk Factor disclosures, which potentially induces noise or even bias in our main analyses. To examine the impact of this potential bias in our sample, we test the sensitivity of our results to an 25

  27. alternative method of identifying comment letters relating to Item 1A risk factors. We use a regular expression to measure an alternative variable equal to one if the SEC comment letter includes the text “Item 1A,” and zero otherwise. We find this alternative indicator variable to be 45 percent correlated with RF_INADEQUATE_SEC. We report results from estimating equation (1) after substituting this variable as the dependent variable in Panel C of Table 6. Results in the first two columns of Panel C indicate that both BP_NER and BP_STANDARD aresignificantly negatively associated with the likelihood of receiving an SEC comment letter related to Item 1A risk factor disclosures. However, results in Columns 3 and 4 of Panel C suggest that when both boilerplate measures are included in the model, only BP_STANDARD has a significantly negative coefficient estimate. We continue to find no evidence of an association between risk factor disclosure length and the likelihood of an SEC comment letter. We next examine the sensitivity of our results to alternative cutoffs to identify boilerplate trigrams and sensitives to measure BP_STANDARD. Our original construction of this variable in our main analyses defines boilerplate sentences as those with at least 10% of all trigrams in the sentence being boilerplate trigrams, where boilerplate trigrams are those appearing in at least 10% of all Item 1A disclosures in the same 2-digit SIC industry on average, but not in more than 90% of all industry peers’ Item 1A disclosures. We test the robustness of our results to alternative cutoffs and report the results in Table 7. The first column of Table 7 (Model 1) displays results after changing those cutoffs to 20%, 20%, and 80%, respectively. The second column of Table 7 (Model 2) displays results after changing those cutoffs to 25%, 25%, and 75%, respectively. Under both alternative sets of cutoffs, BP_STANDARD continues to have a strong negative association (p < 0.01) with the likelihood of an SEC comment letter. 26

  28. The SEC has repeatedly admonished firms to avoid risk factor language that could apply to any issuer. We next examine results from the SEC comment letter analysis after broadening our measure of BP_STANDARD to include trigram similarity with all firms in the economy rather than those in just the same industry. The resulting measure of standardized language arguably captures greater reliance on extremely generic risk factor disclosure language. We apply the original cutoffs (10%, 10%, and 90%) to our construction of BP_STANDARD where these percentages apply to trigrams used in all Item 1A risk factors in our sample, rather than those only in the same 2-digit SIC industry and report results in the last column of Table 7 (Model 3). We find this variable does not correlate significantly with the probability of receiving an SEC comment letter. In untabulated analyses, we fail to find any association between market- wide measures of standardized language based any alternative cutoffsand the likelihood of an SEC comment letter. Overall, our results suggest the SEC does not look favorably or unfavorably upon risk factor language that is so broad that it could apply to any issuer, though the SEC appears to reward firms using risk factor language that is pervasive among firms within the same industry.18 We emphasize that the negative correlation between industry-level standardized risk factor language and SEC comment letters holds after controlling for the actual underlying risk factor topics disclosed. In untabulated analyses, we also include a control variable that measures the cosine similarity between a firm’s risk factor topic probability vector of each firm’s risk factors with those of its industry peers. Results are consistent with BP_STANDARD being associated with a lower likelihood of risk-related SEC common letters, even after controlling for the similarity of disclosed risk topics. These results appear to be inconsistent with the SEC’s 18 We emphasize that our results hold after including controls for the actual underlying risk factor topics being disclosed. In untabulated analyses, we also find the negative relationship between 27

  29. admonition for firms to tailor their risk factor language to reflect the specific risks faced by the firm. Finally, we test the robustness of our results to an alternative measure of risk factor disclosure length proposed by Campbell et al (2014). This alternative measure of length is based on the natural logarithm of a word count of specific risk-related words identified in their study. To test the robustness of our results to the inclusion of this alternative measure of the length of risk disclosures, we rerun our main analyses using this measure in place of LENGTH. We find that this measure also correlates significantly (p < 0.05) with the probability risk factor disclosures will be ruled as inadequate under judicial review, and that this variable has a marginally significant negative correlation with the likelihood of receiving an SEC comment letter (p<0.10). 6.Conclusion Despite official reporting requirements from the SEC that Item 1A risk factor disclosures should be specific and tailored to the reporting firm, practitioners and regulators continue to lament that these disclosures are boilerplate and uninformative. Whereas other academic studies have focused on documenting that risk factor disclosures do convey some information content, our study is among the first to propose and test an explanation for why these disclosures continue to be boilerplate. We provide evidence that specific risk factor disclosures do not lead to more favorable regulatory or judicial outcomes, after controlling for other determinants of risk factor disclosure. Our results provide insight that may inform policy makers and standard setters regarding how legal standards and SEC mandates are implemented in practice. Our results highlight a potential disconnect between formal regulatory and legislative disclosure guidance and how that guidance is enforced via regulatory and judicial outcomes. 28

  30. The results from our study are subject to certain caveats. For instance, we note that the correlation we document between boilerplate and the likelihood risk factor disclosures will subsequently be deemed adequate is not necessarily causal. We cannot rule out the possibility that risk factor disclosures that contain more boilerplate happen to be superior along other dimensions not included in our model, and that it is these correlated omitted variables to which adjudicators are responding. However, the link between favorable judicial and regulatory outcomes and disclosure characteristics the SEC asserts are undesirable provides one possible explanation for the persistence of boilerplate and generic disclosures that run contrary to formal SEC disclosure guidance. 29

  31. REFERENCES Beatty, A., Cheng, L., & Zhang, H. 2018. Are Risk Factor Disclosures Still Relevant? Evidence from Market Reactions to Risk Factor Disclosures Before and After the Financial Crisis. Contemporary Accounting Research, forthcoming. Berkman, O. 2018. Disclosure Effectiveness Weakened By Complicated Ownership. Financial Executives International Daily. Blei, D. M. 2012. Probabilistic topic models. Communications of the ACM 55 (4): 77–8. Blei, D., Ng, A., & Jordan, M. 2003. Latent Dirichlet allocation. Journal of Machine Learning Research 3 (4-5): 993–1022. Bloomenthal, H., & Wolff, S. 2016. Pleading Securities Fraud and the PSLRA. Securities Law Handbook Vol.2, Ch.29. Brown, S., Tian, X., & Tucker, J. 2017. The Spillover Effect of SEC Comment Letters on Qualitative Corporate Disclosure: Evidence from the Risk Factor Disclosure: 1–58. Campbell, J., H. Chen, H., Dhaliwal, D., Lu, H., & Steele, L. 2013. The information content of mandatory risk factor disclosures in corporate filings. Review of Accounting Studies 19 (1): 396–455. Cazier, R., Christensen, T., Merkley, K., and Treu, J. 2017. Litigation Risk and Non-GAAP Reporting. Working Paper: 1–55. Johnson, S. 2010. SEC Pushes Companies for More Risk Information. CFO.com. Cornerstone Research. 2017. Securities Class Action Filings 2016 Year in Review: 1–44. De Simone, J., Ingber, M. & Creutz, E. 2004. High Courts Should Review Ruling on Securities Fraud “Safe Harbor.” Legal Opinion Letter - Washington Legal Foundation 14 (26): 1–2. Dyer, T., Lang, M. & Stice-Lawrence, L.. 2016. The Evolution of 10-K Textual Disclosure: Evidence from Latent Dirichlet Allocation. Journal of Accounting and Economics: 1–68. Finkel, J., Grenager, T., & Manning, C. 2005. Incorporating Non-local Information into Information Extraction Systems by Gibbs Sampling. In, 363–370. Stroudsburg, PA. Gaulin, M. 2017. Risk Fact or Fiction: The information content of risk factor disclosures: 1–51. Working paper, University of Utah. Hoberg, G., & Phillips, G. 2016. Text-Based Network Industries and Endogenous Product Differentiation Gerard Hoberg. Journal of Political Economy 124 (5): 1423–1465. 30

  32. Hope, O., Hu, D., & Lu, H. 2016. The benefits of specific risk-factor disclosures. Review of Accounting Studies 21 (4): 1005–1045. Hopkins, J. 2018. Do Securities Class Actions Deter Misreporting? Contemporary Accounting Research. Huang, A., Lehavy, R., Zang, A., & Zheng, R.. 2017. Analyst Information Discovery and Interpretation Roles: A Topic Modeling Approach. Management Science: mnsc.2017.2751– 24. Investor Responsibility Research Center Institute (IRRC). 2016. The Corporate Risk Factor Disclosure Landscape. Kahan, M., & Klausner, M. 1997. Standardization and Innovation in Corporate Contracting (Or “The Economics of Boilerplate”). Virginia Law Review 83 (4): 713–770.? Kravet, T., & Muslu, V. 2013. Textual risk disclosures and investors’ risk perceptions. Review of Accounting Studies 18 (4): 1088–1122. Lafferty, J., McCallum, A., & Pereira, F. 2001. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. In Proceedings of International Conference on Machine Learning (ICML-2001), 282–289. Lang, M., Stice-Lawrence, L. 2015. Textual analysis and international financial reporting: Large sample evidence. Journal of Accounting and Economics 60 (2-3): 110–135. Lawrence, A. 2013. Individual investors and financial disclosure. Journal of Accounting and Economics 56 (1): 130-147. Li, F. 2008. Annual report readability, current earnings, and earnings persistence. Journal of Accounting and Economics 45 (2-3): 221-247. McMullin, J. 2016. Can I Borrow Your Footnotes? Learning and Network Benefits of Footnote Similarity. Working Paper, Kelley School of Business. Miller, B. 2010. The Effects of Reporting Complexity on Small and Large Investor Trading. The Accounting Review 85 (6): 2107-2143. Nelson, K.., & Pritchard, A. 2016. Carrot or Stick? The Shift from Voluntary to Mandatory Disclosure of Risk Factors. Journal of Empirical Legal Studies 13 (2): 266–297. Olazábal, A. 2000. Safe Harbor for Forward-Looking Statements Under the Private Securities Litigation Reform Act of 1995. Dickson Law Review 105 (1): 1–30. Pritchard, A., & Sale, H. 2005. What Counts as Fraud? An Empirical Study of Motions to Dismiss Under the Private Securities Litigation Reform Act. Journal of Empirical Legal Studies 2 (1): 125– 149. 31

  33. Securites and Exchange Commission (SEC). 2004. Proposed rule: securities offering reform, release nos. 33-8501; 34-50624; ic-26649, international series release no. 1282, file no. s7-38-04. Available at: https://www.sec.gov/rules/proposed/33-8501.htm#VII. Securites and Exchange Commission (SEC). 2005. Securities and exchange commission final rule, release no. 33–8591: 1–468. Available at: https://www.sec.gov/rules/final/33-8591.pdf Securites and Exchange Commission (SEC). 2016. Business and financial disclosure required by regulation S-K, release no. 33–10064; 34-77599; File No. S7-06-16: 1–341. Available at: https://www.sec.gov/rules/concept/2016/33-10064.pdf You, H. & Zhang, X. 2009. Financial reporting complexity and investor underreaction to 10-K information. Review of Accounting Studies 14 (4): 559-586. 32

  34. TABLE 1 Factors Cited in Judicial and Regulatory Assessments of Risk Factor Disclosure Adequacy Panel A: Judicial Assessments of Risk Factor Disclosure Adequacy for Safe Harbor Protections Supporting Reasons Cited When Risk Factor Disclosure is Ruled Adequate Risk factor language fulfils the statutory requirement to warn investors of risks that could cause actual results to vary. Risk factor language is specific/detailed/tailored to the firm Risk factor language included the risk that actually transpired. Assertion that risk factor language is adequate with no clear supporting explanation. Risk factor language is extensive/lengthy/numerous Risk factor language is similar to that found to be adequate by prior courts. Other 94 57 56 47 30 13 12 309 Total Supporting Reasons Cited when Risk Factor Disclosure is Ruled Inadequate Failure to disclose a material risk factor Failure to disclose important facts made risk factors disclosures misleading Failure to provide any risk language with the forward-looking statements Risk factor disclosure was too vague/boilerplate/non-specific Adverse events disclosed as "risk factors" were actually happening Other 51 46 35 35 20 12 199 Total Panel B: SEC Comments Relating to Risk Factor Disclosures Increase specificity relating to information already disclosed Explain further or clarify a risk factor Add an additional risk factor Remove qualifying language from risk factors Change heading or formatting to make content of risk factor more clear Remove a risk factor 84 83 57 40 9 7 280 Total 33

  35. Panel A of Table 1 categorizes the factors cited by judges in explaining why a given firm’s risk factor disclosures are deemed adequate or inadequate to provide legal protection to a defendant’s forward-looking statements under the safe harbor of the PSLRA. Panel A is based on 409 unique judicial opinions regarding defendants’ motion to dismiss lawsuits filed between 1996 and 2015. The total factors cited sum to more than 409 because in some cases the judge cites more than one unique factor. These lawsuits consist of 264 unique cases for which the risk factor disclosures were deemed adequate, and 145 unique cases for which the firm’s risk factor disclosures were deemed inadequate. Panel B categorizes the SEC comments relating to firms’ risk factor disclosures for a random sample of 200 risk factor related SEC comment letters issued between 2006 and 2015. The total issues cited in the SEC comment letters sums to more than 280 because in some comment letters the SEC requests more than one change to the risk factor disclosure. 34

  36. TABLE 2 Sample Selection Panel A: Sample of Lawsuits Alleging False or Misleading Forward-Looking Statements Securities lawsuits with published judicial opinions retrieved from keyword searches on Lexis Nexis Legal with a filing date in 2006 or later Eliminate: Lawsuits against firms missing necessary Compustat or CRSP data or for which the CIK could not be found Lawsuits that do not involve forward-looking statements Lawsuits in which the judge does not explicitly rule regarding the adequacy of the firm’s cautionary language Lawsuits missing Item 1A textual analysis data Unique lawsuits examined in post-2005 multivariate regression analyses 328 (78) (37) (62) (7) 144 Panel B: Sample of SEC Comment Letters Relating to Item 1A Risk Factor Disclosures 4,450 SEC Comment Letters relating to Item 1A Risk Factor Disclosures from Audit Analytics Database (June 2017) Eliminate: Firm-years for which Item 1A textual analysis variables could not be computed Firm-years missing CRSP and Compustat data necessary for model variables SEC Comment Letters relating to Item 1A Risk Factors examined in multivariate regression analyses (1,606) (1,178) 1,666

  37. TABLE 3 Descriptive Statistics Panel A: Descriptive Statistics for Firm-Years in Lawsuit Sample (N = 246) Variable LENGTH BP_NER BP_STANDARD RET_COR RET_VOL BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Mean 0.0917 -0.4687 0.5460 0.2151 0.0343 1.2215 7.4862 0.2299 0.8252 0.3406 0.6031 0.1824 -0.0117 Median 0.0915 -0.2724 0.5518 0.2159 0.0299 1.1775 7.2264 0.1500 1.0000 0.2648 0.4753 0.2888 0.0430 Q1 Q3 0.4911 5.8015 0.6421 0.2878 0.0424 1.5446 8.5297 0.3620 1.0000 0.9282 0.8551 0.3702 0.0747 Std 0.6612 11.3285 0.1473 0.1222 0.0179 0.4826 1.8350 0.2478 0.3806 1.5483 0.4609 0.3679 0.1903 -0.3757 -6.5086 0.4539 0.1247 0.0223 0.9057 6.3059 0.0028 1.0000 -0.3176 0.2665 0.0000 -0.0326 Panel B: Descriptive Statistics for Firm-Years in SEC Comment Letter Sample (N = 33,725) Variable LENGTH BP_NER BP_STANDARD RET_COR RET_VOL BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Mean -0.0267 -0.3969 0.5493 0.2259 0.0314 1.0426 6.2914 0.2215 0.7145 0.4469 0.5820 0.1171 -0.0739 Median -0.0067 -0.1995 0.5576 0.2173 0.0265 1.0491 6.2730 0.1552 1.0000 0.2731 0.3822 0.2700 0.0416 Q1 -0.4747 -6.2545 0.4725 0.1258 0.0184 0.6625 4.8966 0.0224 0.0000 -0.1261 0.1898 0.0000 -0.0274 Q3 0.4231 5.5401 0.6366 0.3203 0.0385 1.4030 7.6394 0.3346 1.0000 0.7907 0.6859 0.3610 0.0717 Std 0.7298 10.4461 0.1334 0.1403 0.0192 0.5758 1.9964 0.3656 0.4517 1.5538 1.0851 11.9401 24.5087 LENGTH is the natural logarithm of the total words in the 10-K Item 1A risk factor disclosure. BP_NER is the number of specific entities identified in the firm’s risk factor section using the Stanford Named Entity Recognition tool, scaled by the total number of words in Item 1A and multiplied by -1,000. LENGTH and BP_NER are benchmarked against the firm’s closest industry

  38. peer, where closest industry peers are determined as discussed in Section 4.2. BP_STANDARD is the number of words in boilerplate sentences in Item 1A scaled by the total number of words in Item 1A. Our measure of boilerplate sentences is based on commonly used trigrams in risk factor disclosures of other firms in the same 2-digit SIC industry, and is described in greater detail in Section 4.2 of the text. RET_COR is the mean correlation between firm’s 24-month return ending at the current fiscal year end and the same window return for each other firm in its same 2-digit industry. RET_VOL is the standard deviation of daily returns for the year ending two days prior to the 10-K filing date. BETA is the firm’s market beta measured using daily returns for the one year ending two trading days prior to the 10-K filing date. SIZE is the natural logarithm of the market value of equity. LEV is total liabilities scaled by total assets, measured as of the end of the fiscal year. BIG_N is an indicator variable equal to one if the firm is audited by one of the largest auditors (the Big N), and zero otherwise. SKEWNEWSS is measured as the negative coefficient of skewness for daily returns over the year ending two days prior to the 10-K filing date. TURNOVER is the mean number of shares traded during the fiscal year scaled by the number of common shares outstanding. ETR is the firm’s effective tax rate, computed as tax expense scaled by pre-tax income. ROE is income before extraordinary items scaled by the market value of equity.

  39. TABLE 4 Inadequate Risk Factor Disclosures and the Probability of Lawsuit Dismissal Dependent Variable DISMISSAL Parameter Estimate Standard Error Model 1 -2.5204 *** 0.3101 Model 2 -4.7078 *** 1.0318 RF_INADEQUATE_JG Risk control variables: RET_COR RET_VOL -0.3751 1.1379 -6.7256 10.1762 0.4432 0.2753 0.2262 ** 0.0906 -0.9149 0.5658 -0.0528 0.4581 0.1648 * 0.0857 0.6353 ** 0.3116 0.4095 0.4066 0.6396 1.1935 Yes Yes Yes 0.882 524 -0.1565 2.6781 -34.3907 22.2959 0.9254 0.7253 0.0276 0.1979 -0.5613 1.5958 0.1559 0.9534 0.5271 *** 0.1822 1.9174 ** 0.7671 1.3952 1.0752 0.3569 2.8737 Yes Yes Yes 0.955 237 BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Area under ROC curve N Table 4 displays the results of a logistic regression of the probability of a lawsuit dismissal on the indicator variable, RF_INADEQUATE_JG, and control variables. RF_INADEQUATE_JG is equal to one for lawsuits in which the judge explicitly rules the firm’s risk factor disclosures are inadequate to avail the firm of the PSLRA’s safe harbor, and zero if the judge explicitly rules the risk disclosures are adequate. All control variables are defined in the notes to Table 3. The sample includes all firm-years that have a 10-K filed within the class period of the lawsuit. The first column displays regression results for our entire sample of lawsuits from 1996 to 2015. The second column displays results for the post-2005 sample for which we have Item 1A textual data. *, **, and *** denote statistical significance at the .10, 0.05, and 0.01 significance levels, respectively (two-tailed tests).

  40. TABLE 5 Judicial Assessments of Risk Factor Disclosure Inadequacy Dependent Variable RF_INADEQUATE_JG Parameter Estimate Standard Error Model -1.0928 ** 0.4390 -0.0593 *** 0.0212 Model 2 -0.9393 ** 0.4099 Model 3 -1.1381 *** 0.4344 -0.0548 ** 0.0218 -1.9703 2.2231 Model 4 -1.2169 * 0.7337 -0.1534 ** 0.0592 -1.8226 3.5529 1.5920 4.6544 13.2772 61.1065 -1.8347 * 1.0237 -0.2322 0.4036 2.5108 2.2288 -2.5460 * 1.4124 0.1336 0.3544 0.0157 0.8864 0.6465 1.0233 9.5632 ** 4.6253 Yes Yes Yes Yes 0.955 246 LENGTH 1 BOILERPLATE_NER -3.1431 2.1375 BOILERPLATE_STANDARDIZED Risk control variables: RETURN_CORRELATION RETURN_VOLATILITY BETA SIZE LEVERAGE BIG_N_AUDITOR SKEWNESS TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Topic fixed effects Area under ROC curve N Table 5 reports the results from a regression of the probability the sued firm’s risk factor disclosures are assessed as inadequate by a judge for purposes safe harbor protection under the PSLRA. These assessments -0.1334 2.4103 16.009 22.295 -0.3331 0.5863 -0.0238 0.2007 2.5916 ** 1.0735 -0.5318 0.7879 0.2012 0.1307 -0.0202 0.4427 -0.0677 0.7546 6.7966 *** 1.7972 Yes Yes Yes No 0.901 246 -0.2172 2.3049 17.9858 21.4423 -0.3061 0.5316 0.0362 0.1840 1.9115 ** 0.9526 -0.5000 0.8527 0.2071 * 0.1247 0.0094 0.4621 -0.1483 0.7238 4.9326 ** 1.5132 Yes Yes Yes No 0.887 246 -0.0672 2.4100 16.9580 22.6897 22.6897 -0.3317 05647 0.5647 -0.0296 0.2005 2.5124 ** 1.0821 -0.4737 0.8582 0.2149 0.1330 0.0352 0.4577 -0.0794 0.7559 6.5524 *** 1.8389 Yes Yes Yes No 0.900 246 0 2 **

  41. are based on the published judicial opinion supporting the judicial decision whether to grant the defendant’s motion to dismiss the securities lawsuit. Variable definitions are provided in the notes to Table 3. Industry fixed effects are measured at the Fama/French 12-industry classification level. Topic indicators are included based on twenty topics estimated using Latent Dirichlet Allocation over the entire corpus of 10-K Item 1A risk factor disclosures over our sample period. All continuous variables are winsorized at the 1st and 99th percentiles. Robust standard errors are used to assess statistical significance, which is denoted by *, **, and *** for statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.

  42. TABLE 6 Regulatory Assessments of Risk Factor Disclosure Inadequacy Panel A: Full Sample Dependent Variable RF_INADEQUATE_SEC Parameter Estimate Standard Error Model 1 0.0347 0.0372 -0.0038 0.0025 Model 2 0.0293 Model 3 0.0264 0.0368 -0.0010 0.0025 -0.9459 *** 0.2377 Model 4 -0.0607 0.0418 -0.0024 0.0027 -0.7792 *** 0.2759 LENGTH BP_NER -0.9725 *** 0.2293 BP_STANDARD Risk control variables: RET_COR RET_VOL -0.1408 0.2621 12.3648 *** 1.7766 0.0755 0.0545 0.1928 *** 0.0194 -0.0305 0.0519 -0.2398 *** 0.0719 -0.0155 0.0148 -0.0034 0.0226 0.0003 0.0027 0.0001 0.0002 Yes Yes Yes No 0.712 33,407 -0.0764 0.2629 12.1190 *** 1.7802 0.0792 0.0546 0.1825 *** 0.0194 -0.0223 0.0470 -0.2395 *** 0.0719 -0.0148 0.0148 -0.0004 0.0220 0.0003 0.0028 0.0001 0.0002 Yes Yes Yes No 0.713 33,407 -0.0784 0.2629 12.1044 *** 1.7803 0.0795 0.0546 0.1828 *** 0.0195 -0.0230 0.0475 -0.2394 *** 0.0719 -0.0149 0.0148 -0.0004 0.0220 0.0003 0.0028 0.0001 0.0002 Yes Yes Yes No 0.714 33,407 0.0080 0.2751 9.6053 *** 1.9478 0.0612 0.0573 0.1830 *** 0.0217 -0.0515 0.0836 -0.1585 ** 0.0752 -0.0142 0.0159 -0.0141 0.0254 0.0000 0.0023 0.0001 0.0003 Yes Yes Yes Yes 0.740 33,407 BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Topic fixed effects Area under ROC curve N Table 6 Panel A reports the results from a regression of the probability that a firm’s risk factor disclosures are targeted by an SEC comment letter during the SEC’s filing review process. Variable definitions are provided in

  43. the notes to Table 3. Industry fixed effects are measured at the two-digit SIC classification level. Topic indicators are included based on two hundred topics estimated using Latent Dirichlet Allocation over the entire corpus of 10-K Item 1A risk factor disclosures over our sample period. All continuous variables are winsorized at the 1st and 99th percentiles. Robust standard errors are used to assess statistical significance, which is denoted by *, **, and *** for statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.

  44. TABLE 6 (cont’d) Panel B: Sample of Firms Receiving Any SEC Comment Letter During the Year Dependent Variable RF_INADEQUATE_JG Parameter Estimate Standard Error Model 1 -0.0013 0.0397 -0.0050 * 0.0026 Model 2 -0.0031 0.0385 Model 3 -0.0097 0.0394 -0.0023 0.0027 -0.9391 *** 0.2557 Model 4 -0.0625 0.0445 -0.0044 0.0029 -0.7470 ** 0.2969 LENGTH BP_NER -0.9990 *** 0.2461 BP_STANDARD Risk control variables: RET_COR RET_VOL -0.2430 0.2727 8.6757 *** 2.1002 0.0263 0.0598 0.0545 *** 0.0207 -0.1578 0.1103 -0.1880 ** 0.0750 0.0078 0.0184 0.0000 0.0339 -0.0002 0.0081 0.0007 ** 0.0003 Yes Yes Yes No 0.687 14,582 -0.772 0.2737 8.3943 *** 2.0997 0.0318 0.0599 0.0436 * 0.0207 -0.1479 0.1089 -0.1855 ** 0.0750 0.0085 0.0184 0.0040 0.0339 -0.0001 0.0082 0.0007 ** 0.0003 Yes Yes Yes No 0.688 14,582 -0.1829 0.2738 8.3738 *** 2.0991 0.0325 0.0599 0.0442 * 0.0207 -0.1509 0.1096 -0.1850 ** 0.0750 0.00839 0.0184 0.0036 0.0339 -0.0001 0.0081 0.0007 ** 0.0003 Yes Yes Yes No 0.688 14,582 -0.0864 0.2882 6.4240 *** 2.3290 0.0217 0.0632 0.0428 * 0.0234 -0.1750 0.1326 -0.1156 0.0798 0.0143 0.0199 -0.0087 0.0341 -0.0016 0.0087 0.0006 * 0.0003 Yes Yes Yes Yes 0.718 14,582 BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Topic fixed effects Area under ROC curve N Table 6 Panel B reports the results from re-estimating the regression from Panel A on only the subset of firms that received at least one SEC comment letter relating to any disclosure for the fiscal period. Variable

  45. definitions are provided in the notes to Table 3. Industry fixed effects are measured at the two-digit SIC classification level. Topic indicators are included based on two hundred topics estimated using Latent Dirichlet Allocation over the entire corpus of 10-K Item 1A risk factor disclosures over our sample period. All continuous variables are winsorized at the 1st and 99th percentiles. Robust standard errors are used to assess statistical significance, which is denoted by *, **, and *** for statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.

  46. TABLE 6 (cont’d) Panel C: Alternative Measure of SEC Comment Letter Relating to Item 1A Risk Factors Dependent Variable RF_INADEQUATE_JG Parameter Estimate Standard Error Model 1 0.0118 0.0716 -0.0083 * 0.0044 Model 2 0.0154 0.0687 Model 3 0.0026 0.0701 -0.0044 0.0045 -1.3015 *** 0.4553 Model 4 -0.0779 0.0818 -0.0068 0.0050 -1.0403 ** 0.5188 LENGTH BP_NER -1.4178 *** 0.4348 BP_STANDARD Risk control variables: RET_COR RET_VOL 0.3971 0.5083 9.8993 *** 3.2683 0.0983 0.1014 0.2149 *** 0.0338 -0.3163 0.2108 0.0618 0.1454 -0.0280 0.0315 0.0270 0.0375 -0.0006 0.0009 0.0008 ** 0.0004 Yes Yes Yes No 0.816 33,407 0.4931 0.5113 9.6006 *** 3.2892 0.1044 0.1014 0.1997 *** 0.0338 -0.2925 0.2086 0.0668 0.1451 -0.0259 0.0317 0.0303 0.0362 -0.0006 0.0009 0.0008 * 0.0004 Yes Yes Yes No 0.808 33,407 0.4842 0.5107 9.5229 *** 3.2874 0.1060 0.1016 0.2003 *** 0.0339 -0.2983 0.2091 0.0687 0.1451 -0.0262 0.0317 0.0305 0.0361 -0.0006 0.0009 0.0008 0.0004 Yes Yes Yes No 0.809 33,407 0.4082 0.5449 8.4329 ** 3.7065 0.0741 0.1081 0.2045 *** 0.0403 -0.5306 ** 0.2586 0.1288 0.1553 -0.0333 0.0358 0.0254 0.0381 -0.0004 0.0010 0.0010 0.0011 Yes Yes Yes Yes 0.831 33,407 BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Topic fixed effects Area under ROC curve N

  47. Table 6 Panel C reports the results from re-estimating the regression reported in Panel A after replacing the dependent variable with an alternative indicator of whether the firm received an SEC comment letter relating to Item 1A Risk Factors. The alternative dependent variable is equal to one if the firm received an SEC comment letter that contained the phrase “Item 1A” relating to that period’s filings, and zero otherwise. Variable definitions are provided in the notes to Table 3. Industry fixed effects are measured at the two-digit SIC classification level. Topic indicators are included based on two hundred topics estimated using Latent Dirichlet Allocation over the entire corpus of 10-K Item 1A risk factor disclosures over our sample period. All continuous variables are winsorized at the 1st and 99th percentiles. Robust standard errors are used to assess statistical significance, which is denoted by *, **, and *** for statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.

  48. TABLE 7 Robustness of Results to Alternative Measures of Standardized Risk Factor Disclosure Language Dependent Variable RF_INADEQUATE_SEC Parameter Estimate Standard Error Model 1 -0.0577 0.0418 -0.0029 0.0027 -1.2661 *** 0.4543 Model 2 -0.0562 0.0418 -0.0029 0.0026 -1.8180 *** 0.5751 Model 3 -0.0565 0.0421 -0.0040 0.0027 0.0436 0.3946 LENGTH BP_NER BP_STANDARD Risk control variables: RET_COR RET_VOL -0.0088 0.2751 9.6655 *** 1.9492 0.0616 0.0573 0.1838 *** 0.0218 -0.0553 0.0860 -0.1572 ** 0.0753 -0.0143 0.0159 -0.0141 0.0250 0.0000 0.0022 0.0001 0.0003 Yes Yes Yes Yes 0.740 33,407 -0.0114 0.2750 9.7133 *** 1.9512 0.0633 0.0573 0.1846 *** 0.0217 -0.0553 0.0854 -0.1591 ** 0.0753 -0.0145 0.0159 -0.0139 0.0249 0.0000 0.0022 0.0001 0.0003 Yes Yes Yes Yes 0.741 33,407 -0.0339 0.2742 9.7544 *** 1.9445 0.0604 0.0572 0.1868 *** 0.0218 -0.0562 0.0852 -0.1615 ** 0.0753 -0.0144 0.0159 -0.0169 0.0260 0.0000 0.0023 0.0001 0.0003 Yes Yes Yes Yes 0.740 33,407 BETA SIZE LEV BIG_N SKEW TURNOVER ETR ROE Industry fixed effects Circuit fixed effects Year fixed effects Topic fixed effects Area under ROC curve N Table 7 displays results from running the regression analysis after remeasuring BOILERPLATE_ STANDARDIZED based on alternative cutoffs for identifying boilerplate sentences. In Table 6, this variable defines boilerplate sentences as those with at least 10% of all trigrams in the sentence being boilerplate

  49. trigrams, where boilerplate trigrams are those appearing in at least 10% of all Item 1A disclosures in the same 2-digit SIC industry on average, but not in more than 90% of all industry-peers’ Item 1A disclosures. In Model 1 of Table 7, we change these cutoffs to 20%, 20%, and 80%, respectively. In Model 2, we change these cutoffs to 25%, 25%, and 75%, respectively. In Model 3, we remeasure this variable based on market-wide rather than industry-specific trigrams based on the original 10%, 10%, and 10% cutoffs. Industry fixed effects are measured at the two-digit SIC classification level. Topic indicators are included based on two hundred topics estimated using Latent Dirichlet Allocation over the entire corpus of 10-K Item 1A risk factor disclosures over our sample period. All continuous variables are winsorized at the 1st and 99th percentiles. Robust standard errors are used to assess statistical significance, which is denoted by *, **, and *** for statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.