Estimating Abnormal Accruals for Detection of Earnings Management

This paper addresses certain methodological issues that arise in estimating abnormal (or discretionary) accruals for detection of event-specific earnings management. This paper examines the specification of time-series models as well as cross-sectional models of expected accruals using annual as well as quarterly data. I show that time-series models provide highly imprecise estimates of abnormal accruals due to the small number of observations used in estimating the parameters of the models. I also show that the cross-sectional Jones model, though well specified for randomly chosen firms, is mis-specified for firms with cash flows either above or below their industry median. This paper develops a model for expected accruals which is well specified for all cash flow levels. Further, using simulation analysis, I show that the model developed in this paper is more powerful at detecting earnings management than the cross-sectional Jones model. In addition, this paper examines the impact of audits on the ability of managers to manage reported earnings. Finally, this paper examines the limitations of the models proposed by Dechow, Sloan and Sweeney (Accounting Review, 1995) and by Kang and Sivaramakrishnan (Journal of Accounting Research, 1995)