The Study On Financial Fraud Detection Model Of Listed Companies Based On Partial Least Squares-support Vector Machine

In recent years, with the propulsion of global economic integration course, the world is more connected every day in every way. However, financial fraud cases in each country exposed gradually. For example, Enron Corp, Global Crossing, MCI WorldCom in the United States and Qiongmingyuan, Zhengbaiwen, Lam Tin shares in China. These financial fraud issues brought investors great loss, seriously hampered the development of economic and stability of society. Financial fraud of listed companies has become the global focus problem and need to be solved urgently. Our rapidly developing socialist market economy and immature capital market require auditors to discover financial statement fraud timely, in order to decrease the economic loss of stakeholder of listed companies. Thereby, our aim is to combine Partial Least Squares with Support Vector Machine, and apply them to build a financial fraud detection model, willing to build an accurate fraud detection model, based on financial report information and suit to Chinese securities markets.This paper mainly applies empirical research method, combined with normative research method. First of all, this paper reviews the relevant literatures of financial fraud detecting of listed companies, expounds existing research results from the aspects of motives, signs and detecting methods. We appraise these results according to national conditions and characteristics, analyze the definition of financial fraud, compare the signs of financial fraud mentioned in SAS NO.82 of the USA and ATW.NO.1 of China, and we also introduce theories and advantages of Partial Least Squares and Support Vector Machine.This paper chooses the financial data of listed companies in Shenzhen and Shanghai between 2004 and 2007 as the base of study, and acquires 223 samples of fraud and 4704 samples of nonfraud. We choose thirteen independent variables, from the aspects of profitability, debt paying ability, special treatment, governance mechanisms, earnings management, size of accounting firms and audit opinions. Through statistical analysis, we discover that earnings per share, frequency of board meetings, share of largest stockholder, liquidity ratio, ST, loss, change of managers and audit opinions are significantly related with financial fraud, while return on equity, ratio of independent directors, receivables/prime operating revenue, inventory/ prime operating revenue and size of accounting firms are not significantly related with company's fraud.We respectively use traditional support vector machine model and partial least squares—support vector machine model to train and test data samples. The result shows that the recognition effect of PLS—SVM model is superior to traditional SVM model, and identification accuracy of fraud samples as well as nonfraud samples exceed 80 percent. Partial least squares are able to reduce dimension effectively, acquire nonlinear factor matrix, and support vector machine has many advantages, such as high imitation degree, effective classification and strong robustness. The model which combines PLS and SVM has great recognition effect. We draw ROC curve to prove the high value of the model.Ultimately, this paper makes some conclusions and proposals, according to the research above:The auditors should observe if the audited entities have fraud signs, in order to guide next steps in auditing; should observe the governance mechanisms of entities and find out whether the entity manage earnings; make use of data mining methods to improve audit approaches, thereby to detect financial fraud of listed companies effectively.