Scienter pleading and Rule 10b-5: empirical analysis and behavioral implications.

Author:Donelson, Dain C.
Position::II. An Empirical Analysis of Scienter Case Law Related to Auditor Defendants through Conclusion, with footnotes, p. 477-509
 
FREE EXCERPT
  1. AN EMPRICAL ANALYSIS OF SCIENTER CASE LAW RELATED TO AUDITOR DEFENDANTS

    1. Introduction and Sample Description

      Previous empirical studies have generally supported the conclusion reached in Part I, that there is great uncertainty in the law regarding scienter pleading under rule 10b-5. Grundfest and Pritchard, in studying the early years of post-PSLRA pleading, found little predictability in court decisions:

      Judges can also value ambiguity to the extent that it allows them greater latitude to exercise discretion, more room within which to compromise with colleagues, and increased opportunity to avoid resolutions that they view as unjust or incorrect by whatever metric they might apply. Judge Posner, for example, suggests that judges often vote "their policy preferences and personal convictions," within the confines of the "rules" of judging, as part of the judging "game." (137) Does an empirical study of rule 10b-5 scienter cases involving auditors support or undermine this picture? To examine this issue, we collect data from securities class action complaints where the auditor is named. Our initial sample comes from the RiskMetrics Securities Class Actions Services database and includes cases naming auditors as defendants where the original case was filed between 1996 and 2005 and the auditor portion of the case was resolved by December 2011. We utilize this time period because all cases are under the same legal framework under the PSLRA, and cases against auditors often take numerous years to resolve. We then match cases involving auditors to complaints from the Stanford Securities Class Action Clearinghouse.

      In order to keep data collection manageable and to ensure that the defendant firms have reasonably similar resources, we limit our sample to cases against the Big 4 (formerly, Big 5 and Big 6) audit firms that involve claims under rule 10b-5. The sample includes a total of 144 cases involving the major audit firms.

    2. Variable Definitions

      Due to our limited sample size, we cannot code every possible allegation by plaintiffs and still maintain sufficient degrees of freedom to conduct analyses. We do, however, define numerous commonly used allegations and utilize common themes (Notice, Risk, and Independence) to group certain variables and define other variables individually.

      The Notice group variable includes allegations that the auditor was informed or had notice of the alleged fraud, from either the press (Press) or a firm employee (Employee), such as a whistleblower. The Risk group includes several variables that commonly imply that a client's audit is higher risk than normal, including poor internal controls (Controls), financial distress (Distress), unusual or questionable transactions, particularly at the end of financial reporting periods (Transactions), and executive turnover (Turnover). The final group is Independence, which includes allegations that the auditor was "too close" to the client, including the provision of nonaudit services (NAS), the employment of alumni from the audit firm at the client, and allegations of unlimited or unusually high auditor access to the client (Access). (138)

      The individual variables are broken out separately for two reasons. First, the "Benchmark" variable, defined as the client firm displaying an unusual propensity to hit earnings or revenue benchmarks or debt covenants, does not fit well within another group. Second, other individual variables may have implications beyond simply being a red flag. For instance, in addition to providing the auditor notice that the client may have an incentive to commit fraud, the "Offering" variable indicates that section 11 liability is likely also present. (139) "SEC" and "Restatement" can be viewed as providing relatively "hard evidence" that something was wrong in the client's financial statements. (140) In particular, the Restatement variable provides an admission of an accounting misstatement, although courts generally hold that a restatement by itself is not indicative of scienter. (141) "High audit fees" may provide evidence of scienter, but the inference behind this variable is substantially crowded due to its mechanical relation with litigation risk. A vast accounting literature finds that auditing firms price litigation risk into engagements. (142) Thus, relatively high audit fees could indicate not only the idea that the client "bought off" the auditor or that the auditor could not afford to lose the client but also that the auditor was aware of potential litigation risk. The "GAAS" variable indicates that the plaintiff alleges that the audit process itself was deficient.

      These variables are defined in detail in Table 1. Each variable is an indicator variable, equal to one if the definition is met, zero otherwise. The group heading variables are equal to one if any subvariable is equal to one, zero otherwise. (143) Table 1 also contains the relative frequency with which each type of allegation appears in the sample.

    3. Univariate Correlations

      Table 2 contains univariate (Pearson) correlations between the primary independent variables (as defined above) and two outcome variables. The first outcome variable is "Settle," indicating the case survived the motion to dismiss. The second outcome variable is "Merit," indicating the auditor settled the case for at least $5 million, the approximate cost of taking an average case to trial. (144) On a univariate basis, 118 (81.9%) of the cases survive the motion to dismiss and settle (Settle), while only 52 (36.1%) of the cases reach a settlement of at least $5 million (Merit). Prior studies use similar definitions of meritorious outcomes with respect to the primary defendants in securities class actions. (145)

      The correlations in Table 2 reveal several interesting patterns. First, even on a univariate basis, the Settle variable is uncorrelated with many of the most common plaintiff allegations: Notice, Risk, Independence, Offering, and GAAS. The only independent variables significantly (at the 10% level, as noted by bold type) correlated with Settle are Benchmark, SEC, Restatement, and High Audit Fee. Second, the Merit variable is significantly correlated with more variables, indicating that settlement negotiations may take into account factors that do not appear to influence courts. While the Benchmark variable loses significance, the Notice and Risk variables become significant when correlated with Merit.

      Among the independent variables, the correlations are as one would expect. For instance, SEC and Restatement are highly correlated (0.29), and all of the variables that relate to the auditor's conduct (Independence, High Audit Fee and GAAS) are highly correlated.

    4. Multivariate Analysis

      We now move to a multivariate setting to investigate which factors are most relevant to the court's decision. We utilize logistic regression because we have a binary dependent variable. Our first model is as follows:

      Prob(Settle = 1) = F([alpha] + [[beta].sub.1] Notice + [[beta].sub.2] Risk + [[beta].sub.3] Independence+ [[beta].sub.4] Benchmark + [[beta].sub.5] Offering + [[beta].sub.6] SEC+ [[beta].sub.7] Restatement + [[beta].sub.8] HighAuditFee + [[beta].sub.9] GAAS) (1)

      "Settle" is a dichotomous variable that is set to one if a lawsuit withstands the motion to dismiss, and to zero otherwise. F is the cumulative distribution function of the logistic distribution, and other variables are as defined above.

      Model (2) is identical to model (1), but the dependent variable is "Merit," which is set to one when the auditor settles for at least $5 million, zero otherwise. This model is provided to provide corroborating evidence regarding whether the factors that the court takes into account also affect the settlement negotiation. The second model is as follows:

      Prob(Merit = 1) =F([alpha] + [[beta].sub.1] Notice + [[beta].sub.2] Risk + [[beta].sub.3] Independence+ [[beta].sub.4] Benchmark + [[beta].sub.5] Offering + [[beta].sub.6] SEC+ [[beta].sub.7] Restatement + [[beta].sub.8] HighAuditFee + [[beta].sub.9] GAAS) (2)

      Models (3) and (4) are very similar to models (1) and (2), respectively, but break the Notice, Risk, and Independence variables into their sub-components. Models (3) and (4) are as follows:

      Prob(Settle = 1) = F([alpha] + [[beta].sub.1] Press + [[beta].sub.2] Employee + [[beta].sub.3] Controls+ [[beta].sub.4] Distress + [[beta].sub.5] Transactions + [[beta].sub.6] Turnover+ [[beta].sub.7] NAS + [[beta].sub.8] Alumni + [[beta].sub.9] Access + [[beta].sub.10] Benchmark+ [[beta].sub.11] Offering + [[beta].sub..12] SEC + [[beta].sub.13] Restatement+ [[beta].sub.14] HighAuditFee + [[beta].sub.15] GAAS) (3)

      Prob(Merit = 1) =F([alpha] + [[beta].sub.1] Press + [[beta].sub.2] Employee + [[beta].sub..3] Controls+ [[beta].sub.4] Distress + [[beta].sub.5] Transactions + [[beta].sub.6] Turnover+ [[beta].sub.7] NAS + [[beta].sub.8] Alumni + [[beta].sub.9] Access + [[beta].sub..10] Benchmark+ [[beta].sub.11] Offering + [[beta].sub.12]SEC + [[beta].sub.13] Restatement+ [[beta].sub.14] HighAuditFee + [[beta].sub.15] GAAS) (4)

      Table 3 presents results from estimating these regressions:

      TABLE 3: LOGISTIC REGRESSION RESULTS Panel A: Regressions utilizing group variables Model (1) Model (2) Dependent Variable Settle t-stat Settle t-stat Intercept 1.30 1.34 -2.32 ** 2.49 Notice -0.33 0.62 0.84 * 1.94 Risk 0.38 0.44 1.03 1.29 Independence -0.79 1.20 -1.19 * 1.67 Benchmark 1.15 * 1.85 0.08 0.14 Offering -1.12 1.45 -0.14 0.26 SEC 0.51 0.75 1.11 ** 2.47 Restatement 1.39 ** 2.56 1.11 ** 2.32 High Audit Fee 1.75 ** 2.38 1.00 * 1.76 GAAS -1.09 1.40 -0.44 0.64 n 144 144 Pseudo [R.sup.2] 0.295 0.272 *, **, and *** Bindicate statistical significance at the 10%, 5%, and 1% levels, respectively, in two-tailed tests Panel B: Regressions utilizing all individual variables Model (3) Model (4) Dependent Variable Settle t-stat Merit t-stat Intercept 1.41 1.48 -1.96 ** 2.27 Press 0.20 0.33 1.20 ** 2.43...

To continue reading

FREE SIGN UP