UC Berkeley

I re-examine the tests of real earnings management to avoid losses developed in Roychowdhury (2006). I find that small profit firms do not contain a higher proportion of observations with real activities manipulation than other firms in nearby earnings intervals. In addition, the real earnings management detected by the models is highly persistent, while the likelihood of staying in the small profit zone is not, suggesting the presence of omitted variables. I confirm this interpretation by demonstrating that the appearance of real earnings management for small profit firms is driven by persistently abnormal values for firms in extreme earnings intervals. Finally, a set of newly designed tests is unable to confirm the use of real earnings management to avoid losses in Roychowdhury's setting.