An Evaluation of Alternative Accrual-Based Models for DetectingEarnings Management

This paper evaluates nine alternative accrual-based models for detecting earnings management. TheEvaluation compares the two type errors (type I and II). The empirical analysis is conducted by testingfor earnings management using three distinct samples of firm-years:A) Randomly selected sample of 1000 firm-years,B) Sample of 1000 randomly selected firm-years in which a fixed and known amount of accrualsmanipulation has been artificially introduced; andc) A sample of firms that we have strong evidence to that their earnings have been managed.Sample (A) was designed to investigate the type I error. Type I error arise when the null hypothesis,that earnings are not systematically managed, is rejected when the null is true. Samples (B) and(C)were designed to test the type II error. Type II error arises when the null hypothesis, that earnings arenot systematically managed, is not rejected when it is false. In this paper, Type II error is generated intwo ways. First, we use simulations in which earnings have been artificially manipulated by added afixed and known amount of accruals. Second, we use another set of firm-years, for which we havestrong priors that earnings have been managed.The empirical analysis generates the following major evidences. First, all of nine models appear wellwhen applied to investigation the type I error. Second, regression-based models (Jones1991, ModifiedJones 1995, Dechow et al 2002, and Two deflated models) to detect earnings management, introducesthe lower type II error than naïve models(Healy1985, De Angelo 1986, Modified De Anglo 1994).Third, the modified Jones model 1995, and Dechow et al 2002, have lower type II error than the otherregression-based models.