Research On Financial Fraud Identification Of Listed Companies

The financial fraud of listed companies exists with the establishment of the capital market,and this behavior harms the whole society seriously.But financial fraud has the characteristics of universality,concealment and complexity,even experienced auditors,not to mention ordinary investors,have failed to spot flaws in the carefully crafted financial statements.The number of listed companies in China is increasing every year,but the current financial statement audit of the company is still a manual inspection method.This time-consuming and laborious method still has great subjectivity,and can not meet the needs of urgent financial fraud in China.Therefore,finding a simple and efficient method to identify financial fraud has become the focus of research.Massive amounts of data will be generate in the age of informationization.We use data mining techniques to find the hidden value behind massive data.The company-related information published by l isted companies each year involves a lot of data,and data mining technology can be used to identify these data for financial fraud.This paper first studies the literature on financial fraud,from the reasons of financial fraud,the characteristics of financial fraud companies and the identification of financial fraud companies,and then explains the definition of financial fraud and related terms.Then,from the CSMAR database,we select the companies that were publicly punished by the regulatory authorities for the six kinds of violations involving fictitious profits,false assets,false records,postponement of disclosure,major omissions and false disclosures from 2008-2017.According to the screening results,we obtained a total of 1081 listed companies,which had a total of 2,851 frauds during the sample period.We analyzed the samples from the perspectives of fraud years,types of violations,provinces and industries of fraud companies,determined the preliminary indicator system based on previous studies,and obtained 170 samples of financial fraud companies and 170 samples of non-financial fraud companies based on our matching rules.After standardization,significance test and correlation test on the preliminary indicator system,we use the manual dimension elimination,principal component analysis,and Boruta algorithm to obtain three types of final indicator data sets.Finally,these three types of indicator data sets are identified by Logistic regression,random forest,support vector machine and artificial neural network for financial fraud,and evaluate the model based on the evaluation indicators set by the model.We have the following conclusions:(1)The number of companies that experienced financial fraud in 2008 and 2017 was less than the other eight years.The possible reasons are that the change in accounting calibre in 2007 caused the regulators to be insensitive to the company's financial data and the lag of fraud.The financial fraud often used by listed companies is fictitious profit,followed by disclosure.During the sample period,the number of companies with financial fraud in only one year was the largest.However,there are still some companies that carry out financial fraud in a number of years,so the current financial fraud situation in China needs to be improved.Listed companies often use only one type of violation to conduct financial fraud,but the use of two or more violations to conduct financial fraud companies accounts for about half of the total sample.This shows that it is very simple for a listed company to commit financial fraud.When it uses one kind of irregularities to commit financial fraud,if it is not found in time,it will use other irregularities to keep the financial report normal.In other respects,Guangdong Province and the manufacturing sector are the provinces and industries with the highest frequency of financial fraud.(2)According to the significance test,financial fraud companies and non-financial fraud companies have significant differences in some financial indicators,especially in terms of development capability and operational capability.There is a significant difference between the two indexes of price-to-book ratio and price-to-cash ratio,therefore,people should focus on these two aspects when judging whether a company is doing financial fraud.On the non-financial indicators,except for the size of the board of supervisors and the logarithm of the domestic audit fees,the other non-financial indicators have not passed the test.This is different from the previous research results,which may be due to the difference in sample selection or non-financial indicators have little difference between the two types of samples.(3)It is obtained from the results of the model that the combination of principal component and neural network is the best to identify financial fraud.However,the identification of the indicators selected by Boruta is not superior to the indicators of manual screening.This may be because the Boruta algorithm only selects the index based on the importance of the indicator to the forecast category.The indicators it screens are also highly correlated,resulting in redundant information in the indicator and loss of information of other indicators.The combination of the indicators selected by the algorithm and the four classification models is not good.The extracted 12 principal components can represent 82%of the information of the 36 indicators,and the principal components are not related to each other,so the principal component is better than the other two indicator datasets.Among the four classification models.the recognition effects of random forest.support vector machine and artificial neural network on these three types of indicator data sets are better than Logistic regression.This shows that the classification model in data mining identifies the effect of financial fraud better than the traditional classification model.Finally,according to the analysis of the current situation of domestic financial fraud,it is proposed to optimize the characteristics of senior management team and strengthen team building to help enterprises avoid internal financial fraud.It is proposed to increase the intensity of financial fraud and strengthen information disclosure to help the government improve the current situation.Based on the results of the study,it is important to focus on the differences in indicator statistics and volatility recommendations to help investors make decisions.