We provide evidence that accruals quality in the U.S. has generally improved since 2000, following a decade of decline during the 1990s. Our results indicate that both the initial decline in accruals quality and the subsequent reversal can be explained by an inverse relation with operating cash flow volatility. Moreover, even though patterns of cash flow volatility and accruals quality vary in different regions of the world and relative to the U.S., we find evidence of an inverse relation between accruals quality and cash flow volatility globally. We corroborate our main results in a battery of additional tests, which also indicate that our results are not attributable to other explanations. Overall, our evidence challenges conventional wisdom and suggests that concerns about a sustained decline in accruals quality over time are unwarranted.

Researchers typically infer real earnings management when a firm's operating and investing activities differ from industry norms. A significant problem with classifying deviations from industry averages as myopic earnings management is that companies can change their operating and investing decisions for strategic business reasons rather than to mislead stakeholders. Using principal components analysis, we systematically evaluate existing measures and develop a comprehensive real activities measure to better capture earnings manipulation. Our measure reflects (i) deviations from industry averages across multiple activities and (ii) other signals of manipulation. This approach is promising because, although there are many sources of abnormal activities, manipulation is more likely the cause when managers engage in multiple income‐increasing abnormal activities that coincide with other signals that indicate an elevated risk of manipulation. This simple approach results in a metric that associates negatively with future operating performance and earnings persistence, yields high‐power tests, and captures manipulation reasonably well across most life‐cycle stages. Importantly, this approach performs better than the standard real earnings management metrics across all dimensions. Specifically, it generates the expected reduction in future earnings and reduced earnings persistence in 82% of the tests compared to 36% and 46% in common alternatives. Also, because this innovation does not require a long time‐series or rely on future period realizations for classification, it can be useful in more research settings than other recent innovations in the literature.