The Research On The Financial Report Violation Of Listed Companies Based On Data Mining

Financial report violations of the listed companies is a worldwide problem, it even exist in some large enterprises in developed countries with mature market economy. Various stakeholders made their investment and lending decisions based more on the results of operations and financial condition of the listed companies than before. Financial report of listed companies has become an important tool to judge the operating conditions of the listed companies and predict the development prospects of the companies. However, the listed companies take advantage of some financial means within the allowable range of laws, regulations and accounting standards to whitewash the financial report have occurred frequently. Between March to June in2011, the auditors of the24Chinese listed companies in the U.S. submitted their resignation or exposure the audit object’ financial problems. There are19companies suffered suspension or delisting.It is a complex area to administer the financial violation of listed companies. The form of violations changed with each passing day. Given the severity of the harm and the extensive destruction of financial reporting violations, it is meaningful to explore the identification method of violations of the listed companies for the stability of the capital market.Data mining technology has unique advantages in identifying financial report violations. It can excavate potentially useful information from the vast amounts of data of the financial statements. Compared to the experience of professionals, data mining techniques can reduce subjective judgment, shorten the identification time and improve the accuracy of identification. In recent years, data mining techniques become more sophisticated, and its application in the field of economic management has become increasingly widespread. However, the way to identify financial report violations in China is mainly concentrated on the logistic regression model and neural network model. The research on financial report violation with joint identification method is relatively few.This paper research on186punished manufacturing listed companies in Shanghai and Shenzhen from2006to2011due to the financial report violations. And I select186manufacturing listed companies in the same accounting year as a control sample. The empirical study on whether the violation occurred on the financial reports of listed companies is using data mining techniques with the statistical analysis tools of R and SPSS20.0. The main model in this paper is RPART-AdaBoost. And I compared it with the traditional model to explore the efficiency of the integrated model on recognition accuracy.First, by reviewing the findings of the domestic and foreign scholars and analyzing the main motivation to procure financial report violations and the means of financial report fraud, we can determine the financial and non-financial indicators needed to establish the model.Secondly, by classifying the listed companies with financial report violations, I found that they mainly come from the manufacturing industry. So this paper examines violations features limited to listed companies in the manufacturing industry. And then select the listed companies without financial report violations based on some principles. There are series data preprocessing should be done for financial indicators and non-financial indicators.Again, I will build a Boosting model with decision tree as based classifier, and compared it with traditional financial report recognition model to explore the efficiency of the integrated model on recognition accuracy. In the comparison of model effects, I evaluate the effect of recognition of the five models through comparing the wrong size of class Ⅱ, the shape of the ROC curve and the size of AUC values, so that enhance the convincing of the contrast results.Finally, through the theoretical review, empirical research and comparative analysis, we can extract the following conclusions:Motive analysis shows that financial report violations is mainly based on the interests of the drive. By classifying the listed companies with financial report violations,I found that they mainly come from the manufacturing industry.31financial indicators have significant difference between the anomalistic companies and normal companies according to the Mann-Whitney U test. Non-financial indicators have no significant difference. Neural network model, support vector machine model, decision tree model, RPART-AdaBoost model have well recognition results, while the logistic regression model has less effect. Given the complexity of the financial report violations identified, we should analyze financial information from different point of view.The main contribution of this paper are:(1)The variables in this paper includes not only financial indicators, but also non-financial indicators.(2)Combination of the two types of data mining technology. By integrating Boosting model with decision tree as based classifier, we can find it significantly improve the recognition accuracy of the decision tree. The results show that the integrated Boosting algorithm with other classifier can produce good results.(3)The comparative analysis of a variety of data mining model. Comparing RPART-AdaBoost with traditional methods help we find the differences in the recognition error rate on the model. On the other hand, we can use different model to identify key indicators to focus on the inspection and get a more comprehensive analysis of the financial reports.(4)The introduction of a variety of methods for comparing the models. I evaluate the effect of recognition of the five models through comparing the wrong size of class Ⅱ, the shape of the ROC curve and the size of AUC values, so that enhance the convincing of the contrast results.This study provide the tools and means which assist in identifying financial report violations, improve recognition accuracy, and reduce the possibility of misjudgment.