Financial Reporting Fraud Research Based On The Method Of The SVM

This article selects from1994~2013, which criticized by public、censure by public andadministrative penalties by the China Securities Regulatory Commission、 Shanghai StockExchange, Shenzhen Stock Exchange and Ministry of Finance. A total of96fraud companyselect, with the proportion of1:1matched company96selected to analysis. This paperbases on the theory of fraud GONE, we analyze the relationship between of the greed、chance、needs and exposure and financial reporting fraud. We set index respectively, and useway of support vector machine (SVM) to construct the model of financial reporting fraud andnon financial reporting fraud company. And we draw the conclusions:(1) when thecompany’s managers are risk preference, the managers own significant stakes, and themanagers join together of two position, the managers will abuse power for personal gains, tryto use the right to meet the interests of the individual, purse the maximization individualutility, which means the managers own higher level of greed, the greater the chance of fraud.(2) The chairman change or not, related transactions more, The greater effect of the relatedtransaction, than the greater the possibility of financial reporting fraud. The less the times ofshareholders meeting, the smaller the size of the board of supervisors, than the greater thepossibility of financial reporting fraud.(3) The company will provide false informationandfraulent financial reporting to match the purpose of SEOs, right offerings, relieve the debtpaying ability, maintain good operating performance.(4)When the auditor is responsible, theaudit of financial reports are more strictly, The audit opinion is more cautious, than the lessthe possibility of financial reporting fraudThis article analyzes192companies,and decided to adopt the model by SVM,this articleincludes two models which we named the Adance waring model and the Investigation model:the Adance waring model is based on substitution variable of the greed、chance、needs andexposure; the Investigation model is based on financial indicators which has significantdifference. And than we draw a conclusion that the precision of the two models are73.26%and71.43%. And the first misjudgment rate is under the second misjudgment rate, which means the recognition effect of the models we established is good.