Research On Financial Fraud Identification Of Listed Companies Based On Text Information

In recent years,the financial fraud of listed companies in China is becoming more and more serious,and this kind of financial fraud of listed companies not only hits the confidence of the of investors,but also damages the function of the stock market to allocate of resources.Therefore,it is particularly important to establish an effective financial fraud identification model to identify the risk of financial fraud in listed companies.This paper takes A-share listed companies as research samples and uses the listed company violation information database of CSMAR Database to define the violations involving fictitious profits and fictitious assets in the database as financial fraud.Thus,A total of 258 fraud samples and matchin g samples are selected from the sample database from 2008 to 2020.Through the statistics of the characteristics of listed companies’ financial fraud,the time characteristics and industry characteristics of listed companies’ fraud are summarized.The construction of financial fraud identification model in this paper involves three aspects,one is the construction of financial and non-financial indicators,the other is the construction of text emotion indicators,and the third is the model selection.In terms of financial and non-financial indicators,29 indicators were initially selected from the perspectives of profitability,debt paying ability,development ability,operation ability,cash flow and corporate governance.And then the Mann-Whitney-U test was used to further select 15 indicators as the indicators of the financial and non-financial indicators system from these 29 indicators.This paper fully mining the emotional information contained in MD&A text information,using lexic-based method to extract and quantify the emotional indicators in MD&A text from the emotional perspective,including positive emotion indicator(POS),negativ e emotion indicator(NEG)and net TONE indicator(TONE).In terms of model selection,the paper chooses the decision tre e,support vector machine and Logistic model in the machine learning model to construct the financial fraud identification model.At the same time,Accuracy,Precision,Recall,F1-Score and AUC were used to analyze the classification performance of the mod el.The empirical research results show that the classification performance of the model has been improved to varying degrees after the inclusion of text sentiment indicators,and the classification effect of the support vector machine model is most significantly improved after the inclusion of positive sentiment indicators(POS)and negative sentiment indicators(NEG).It is proved that the text inform ation in the annual report is really helpful to identify the financial fraud of listed companies.At the same time,it is found that the classification performanc e of the decision tree model is the best if the text emotion index is not incorporated.In thi s case,the importance of the indicators is calculated by using the trained decision tree model.It is found that the importance of the three indicators,namely cash flow liability ratio,return on equity and Z index,ranks first.Therefore,we can focus on the abnormality of these three indicators in the study of listed companies’ fraud.