Estimation of earnings manipulation in U.S. listed companies based on weighted discriminative model

Abstract

The paper profiles sample of earnings manipulation in U.S. listed companies, identifies their distinguishing characteristics, and estimates a model for detecting manipulation. Compared with whole sample firm, there are small amount of firm engaging in earnings management and data are uneven for analysis, Weighted Discriminative Model (support vector machine) have been selected to solve this problem. SFS and several feature selection methods have been adopted to select proper feature sets for Weighted Discriminative Model. After feature selection and training, the trained Weighted Discriminative Model is suitable for supporting users such as investor and auditor to detect earnings manipulation. It is also helpful to make correct decision on earning judgment when anglicizing listed company's financial report.