A Quantificational Research Of Identifying Fraudulent Accounting Information Of Listed Company

Fraudulent accounting information has become a focus of social concern, but it is difficult to be identified especially when using the method of quantification. Taken the fraudulent accounting information of China's listed companies as the object, this paper applies the quantificational research of identifying fraud to China's securities business. According to the audit opinion of annual financial reporting of listed companies, we identify two kinds of empirical samples and choose twenty eight financial indices. Making use of Multivariate Discriminant Analysis and Artificial Neural Network based on L-M algorithm, we build up three models of identifying fraudulent accounting information of listed companies in China. In the research, the MDA function and 5-indices-ANN model illustrate a great power in identifying the fraud. This article is divided into four chapters.Chapter 1 introduces the background and literature survey of this research, clarifies the purpose, process and structure of this paper. With the purpose of providing foundation of choosing empirical samples according to audit opinion, Chapter 2 studies the theoretical basis and concept of fraudulent accounting information, expatiates the relationship between audit opinion and fraudulent accounting information. Chapter 3 introduces the process of sample choosing, indices selecting and filtering, and model constructing of quantificational research of identifying fraud. Based on the descriptive analysis, Chapter 4 builds up MDA function and ANN models of identifying fraudulent accounting information of China's listed companies, and verifies the identifying ability of these function and models. Conclusion summarizes the result, probes into its academic and practical significance, and points out the limitation and direction of the future empirical research.The results of research show that the MDA function and 5-indices-ANN model can successfully identify the fraudulent accounting information. Particularly, 70 percent accuracy can be achieved using the latter model in identifying fraud of listed companies other than the original samples. The outcome of the research will provide academic and practical support to accounting firm, investors and government departments.