Measuring Accounting Fraud and Irregularities Using Public and Private Enforcement

Corporate accounting fraud has a significant negative impact on the economy and investors, so academic research on factors that make accounting fraud more or less likely to occur has substantial real-world and public policy implications. However, conducting such research is difficult because researchers cannot observe the incidence of fraud for most firms, corporate admissions of fraud are rare, and trials proving fraud are almost nonexistent. Thus, researchers are forced to rely on proxies for fraud to conduct empirical analysis. Our recent paper examines the use of both public and private accounting enforcement with appropriate screening to proxy for accounting fraud and demonstrates how this combined proxy improves research inferences.

The current dominant proxy for accounting fraud in research is public enforcement through the Securities and Exchange Commission (SEC). In contrast, relatively few papers, particularly in the accounting literature, use private enforcement via securities class actions (SCAs). However, we argue that the use of only public or private enforcement excludes credible fraud firm observations as the SEC lacks a sufficient budget to pursue all possible fraud and private litigants lack the incentive to pursue such cases if their expected costs exceed expected recoveries. Critically, the use of either enforcement regime does not only reduce statistical power but can also bias regression estimates.

Our paper is organized into three main parts. First, we review the relevant legal standards and procedural process for both public and private enforcement and make two main contentions (1) there is no legal or procedural reason to rely on only SEC cases, as is commonly done, or only SCAs in accounting fraud research but conversely (2) there are numerous instances of public and private enforcement that should not be included in such research because (a) both enforcement regimes bring cases that involve non-accounting issues, (b) the SEC in particular often does not allege fraud in its enforcement cases, and (c) SCAs are often dismissed.

Second, to empirically validate these contentions, we construct a large sample of SEC cases and SCAs from 1998 to 2014. Notably, we observe a significant decrease in both types of enforcement over time, which makes the reduced statistical power from using cases from only one enforcement regime even more problematic. Contrary to common claims, we show that SCAs are rare even after significant stock price drops, and we highlight the importance of including SCAs in fraud samples by noting several high-profile cases, such as the Lehman Brothers “Repo-105” case, which would be excluded from a sample using only SEC cases. Emphasizing the importance of screening both types of enforcement, we show that settled SCAs have significantly higher merits than dismissed cases, consistent with a robust process for screening out less meritorious cases. We also show that SEC cases that allege fraud have significantly higher merits than those that do not allege fraud. Ultimately, we show that both SEC cases and SCAs with appropriate screening “look” like fraud (and each other) as both target firms with similar financial (mis)reporting and are associated with adverse consequences after the revelation of the fraud (e.g., higher executive turnover and lower institutional ownership).

Third, we examine how the use of our recommended proxy for accounting fraud affects research inferences. The use of only public or private enforcement data in isolation results in fraud “false negatives,” while the failure to screen out dismissed cases or those that do not allege fraud results in fraud “false positives.” If such misclassifications are random, they only introduce measurement error, increasing the risk of Type II errors (see Hausman et al. 1998). However, non-random misclassifications can result in biased coefficient estimates, increasing the risk of Type I errors (see Meyer and Mittag 2017).

We examine two research settings using only public enforcement or both public and private enforcement to illustrate these issues. First, we show that prior research inferences from Lennox and Pittman (2010) that the Big 4 auditors reduce the occurrence of accounting fraud may rather be due to the SEC being less likely to investigate the clients of Big 4 auditors (i.e., Type I error). Second, we show that a viable hypothetical research project – developing an abnormal inventory model for misreporting fraud – may have been abandoned by researchers due to a lack of statistical power (i.e., Type II error). In both settings, researchers using our recommended proxy for accounting fraud are less likely to obtain erroneous research inferences.

We conduct two additional analyses to complement these findings. First, we examine restatements, which have been used in their entirety or in subsets (e.g., focusing only on severe restatements) as an alternative proxy for accounting fraud. We find that many restatements, including those labeled as “fraudulent” in Audit Analytics, a commonly used research database, appear to be false positives that lack credible fraud allegations. Additionally, numerous instances of fraud in our sample lack restatements and are thus false negatives in a sample of restatements. Second, we create a fraud prediction model that demonstrates cross-enforcement regime predictability (i.e., existing known determinants of public enforcement also identify private enforcement and vice versa) and has increased explanatory power for accounting fraud that reduces both false positives and false negatives relative to models used in existing research.

Our paper makes five primary contributions. First, we provide a framework to consider when evaluating fraud proxies for construct validity. Second, we provide descriptive evidence on both public and private enforcement over a long sample period. Third, we show that the use of a fraud proxy based on both public and private enforcement can beneficially affect research inferences. Fourth, we demonstrate the problems of using restatements to proxy for accounting fraud. Fifth, our fraud prediction model can be used by future research, particularly by researchers who desire to hold the fraud probability constant across firms, which is important for numerous research questions.

The complete paper is available for download here.