A Conceptual and Empirical Evaluation of Accrual Prediction Models

Abstract

Using the data from 1,267 firms suspected of managing earnings to avoid losses, this study evaluates the statistical power of the Jones (1991) model and the Kang and Sivaramakrishnan (1995) model. Both models offer strong evidence suggesting that these firms have inflated earnings by the order of 2.0% to 2.4% of beginning total assets. This finding reinforces the result of Burgstahler and Dichev (1997) which relies solely on the empirical distribution of earnings. More importantly, the evidence indicates that the statistical power of the Kang-Sivaramakrishnan model using GMM or IV procedure (both available in SAS statistical package) is greater than that of the Jones model for both large and small samples, and for using the firm-by-firm version or the pooled cross-sectional version of the Jones model. Such results using real data are consistent with those using simulated data reported originally by Kang and Sivaramakrishnan (1995). The conceptual advantages of each approach -- in particular, the use of levels versus changes in current accounts when estimating unmanaged accruals--are also discussed.

JEL Classification: M41, M43