The Empirical Research On Evaluating The Power Of Discretionary Accruals Models

Earnings management plays an important role in empirical accounting. Since the1980s，many scholars have been working on detecting earnings management， thus many modelsemerged. Most of the models specialize in separating accrued profits into discretionaryaccruals and non-discretionary accruals. As a result， earnings management can be detectedby measuring discretionary accruals through multiple linear regression. This kind of model isnamed discretionary accruals model， which is a widely recognized measure for earningsmanagement in the field of empirical accounting.However this kind of models is controversial for its limitations， which can notcomprehensively and objectively detect earnings management. And few care how much canthe models reflect earnings management and which model is the best and most suitable forChinese enterprises at present.This paper uses empirical accounting methods to study the power of discretionaryaccruals models on detecting earnings management， and the paper is based on the1999-2012financial data of all the listed companies in Shanghai Stock Market and Shenzhen StockMarket.The paper chooses eight representative alternative discretionary accruals models，namely Model1: the Jones model， Model2: the Modified Jones model， Model3: the Jonesmodel with intangible assets， Model4and Model5: the Jones model with ROA (ROA isdifferent in each of them)，Model6and Model7: the Modified Jones model with ROA (ROAis different in each of them)， Model8: the Jones model with cash flow (D-D model). In thispaper， regression analysis is processed with industry-classified cross-section data of eachyear to gain discretionary accruals under different models.Through the descriptive statistics for variates and residues and with the methods of Rlanguage programming and computer simulation， the frequency of the eight discretionaryaccruals models committing Type-I and Type-II errors in detecting earnings management isdetected respectively. The paper also uses the illegal data announced by CSRC (ChinaSecurities Regulatory Commission) to exam the power of the models.This paper provides the following major insights: First， Model2(the Modified Jonesmodel) is a relatively good model to detect earning management， because it seldom commits both Type-I and Type-II errors， and produces comparatively good results on examiningillegal data. Second， Model8(D-D model) performs well in Type-II error test and detectingillegal data， but the possibility of committing Type-I error is high， thus tends to exaggeratethe level of earning management of Chinese listed companies. Third， Mode l performs goodin the Type-I error test， but not as so good in Type-II error test and illegal data detecting.Model3(the Jones model with intangible assets)， Model6and7(the Modified Jones modelwith ROA) perform medium in all the tests， and the poor performance of Model4and5(theJones model with ROA) in the tests proves that the two models are not suitable for earningsmanagement detecting.