Earnings Manipulation and Expected Returns

An accounting-based earnings manipulation detection model has strong out-of-sample power to predict cross-sectional returns. Companies with a higher probability of manipulation (M-score) earn lower returns on every decile portfolio sorted by size, book-to-market, momentum, accruals, and short interest. The predictive power of M-score stems from its ability to forecast changes in accruals and is most pronounced among low-accrual (ostensibly “high-earnings-quality”) stocks. These findings support the investment value of careful fundamental and forensic analyses of public companies.In our study, we investigated the investment value of a particular form of financial analysis associated with the detection of earnings manipulation. The statistical model we examined (the Beneish model) represents a systematic distillation of forensic accounting principles described in the practitioner literature. Specifically, we investigated a potential link between the probability of manipulation (M-score) generated by the Beneish model and subsequent returns.Although relatively few companies are indicted for accounting fraud, the incidence of earnings manipulation among public companies is likely much higher. We posited that the M-score is informative about a company’s expected returns because the “typical earnings manipulator” is a company that is growing quickly, experiencing deteriorating fundamentals, and adopting aggressive accounting practices.Our main hypothesis was that companies that share traits with past earnings manipulators (i.e., those that “look like manipulators”) represent a particularly vulnerable type of growth stock. Although the accounting games they engage in might not be serious enough to warrant regulatory action, we posited that their earnings trajectory is more likely to disappoint investors (i.e., they have lower “earnings quality”). To the extent that the pricing implications of these accounting-based indicators are not fully transparent to investors, companies that “look like” past earnings manipulators will also earn lower future returns.We found that companies with a higher probability of manipulation (M-score) earn lower returns in every decile portfolio sorted by size, book-to-market, momentum, accruals, and short-interest ratio. These returns are economically significant (averaging just below 1% a month on a risk-adjusted basis) and survive a host of risk controls. We further found that a large proportion of the abnormal return is earned in the short three-day windows centered on the next four quarterly earnings releases, suggesting that our results are due to a delayed reaction to earnings-related news rather than risk-based factors. The robustness of these results, even among highly liquid companies, implies that they are unlikely to be fully explained by transaction costs.We performed three sets of analyses to better understand the nature of the information conveyed by M-score. First, we conducted detailed tests on the joint ability of accruals and M-score to predict returns. We found that the dominance of M-score over accruals is evident in both independent sorts and nested sorts. When companies are sorted on these two variables independently, M-score is particularly effective in predicting returns among low-accrual companies (i.e., companies that have “high earnings quality” according to their accruals ranking). For example, in the lowest-accrual quintile—companies typically viewed as “buys”—the spread in size-adjusted returns between high- M-score companies and low- M-score companies is –19.8% over the next 12 months.Second, we used a difference-in-difference test to examine which individual components of the model contributed the most to its incremental predictive power. Our results show that variables related to a predisposition to commit fraud (sales growth, asset quality index, and leverage) are more important than variables associated with the level of aggressive accounting (accruals, days in receivables, and depreciation expense).Third, we found that the Beneish model’s efficacy is associated with its ability to predict the directional change in current-year accruals (i.e., whether the accruals component of current-year earnings will continue into next year or disappear). Specifically, we found that high- M-score companies have income-increasing (-decreasing) accruals that are more likely to disappear (persist) next year; we observed the exact opposite among low- M-score companies. In other words, M-score provides useful information about the future persistence of current-year accruals.Our study adds to the literature on the effective use of financial information by documenting the usefulness of earnings manipulation detection techniques for earnings quality assessment and return prediction. Our evidence on how and why such techniques work suggests new directions for earnings quality analysis and should enhance future efforts to identify potential over- and undervaluations. Overall, our analyses provide substantial support for the use of forensic accounting in equity investing.