Research On Financial Fraud Identification Of Listed Firms Based On Multi-source Data

Financial fraud of listed firms not only brings huge losses to investors,creditors and other stakeholders,but also has great harm to the allocation of national resources and the development of the securities market.With the financial statements more and more complex means of whitewashing,fraud has become more hidden,relying solely on the financial statements of the original data and ratio indicators have been unable to facilitate,effective identification of corporate financial fraud,financial fraud identification has become an urgent need to solve the problem.On the basis of financial ratio indicators to identify financial fraud,the same time period samples are selected,and with reference to the category of multi-source data,textual indicators,corporate governance indicators and external supervision indicators are selected from the multi-source data set to identify financial fraud of listed firms in turn.First,we select financial ratio data to identify the financial fraud of listed firms in Shanghai and Shenzhen in China.The published financial report data from 2006-2021 are selected,and the441 fraud years of 144 listed firms with financial fraud from 2006-2019 are used as the training set samples,and the non-financial fraud firms are matched with a 1:1 ratio,and the identification accuracy of 249 fraudulent listed firms with fraudulent fraud years in the test set(2020-2021)is used to test the applicability of the model and the indicator’s validity.Based on the accounting practice experience and three tables and inter-table collinearity,44 high-frequency used financial ratio indicators are screened out,and Logistic regression financial fraud identification model is constructed to empirically demonstrate the relationship between financial ratios and listed firm fraud,and the model identification rate is 77.91%.Among them,a total of 10 indicators are significant indicators.Further research,based on the literature containing studies on nonfinancial indicators and three tables and inter-table collinearity,70 financial indicators are sorted out and 36 financial ratios are screened,of which 19 indicators are significant.The model identification rate is 81.11%,which is an increase in the recognition rate from above study.But,from the mean and standard deviation of the indicators,it is not possible to effectively determine the difference between fraudulent and non-fraudulent firms,and such a large number of financial indicators also brings inconvenience to the identification of fraud,multi-dimensional auxiliary perspective of multi-source data for the identification of financial fraud provides a new vision.Second,in order to verify whether the textual information of financial reports helps stakeholders to identify corporate financial fraud,we take 2020 Chinese manufacturing listed companies as a sample and mine MD&A textual information in annual reports from the three dimensions of linguistic structure,linguistic quality,and linguistic expression,and construct an MD&A textual indicator framework that encompasses a relatively comprehensive scope and extends the sources of MD&A textual indicators beyond the previous studies that have only focused on readability or intonation.Then,we utilize text mining techniques to quantify readability,prospective,similarity,correctness,positive sentiment,and negative sentiment metrics of MD&A texts with the help of python tools.The combined disclosure of textual and financial metrics can improve the accuracy of identifying financial fraud compared to using financial metrics alone.From model comparisons,the inclusion of textual metrics improved the identification accuracy of the models in general,with the XGBoost model having the best identification accuracy among the three models(Logistic,XGBoost,and MLP NN models),the MLP NN model having the second highest identification accuracy,and the Logistic model having the lowest identification accuracy.In terms of impact mechanisms,readability and forward-looking are positively associated with financial fraud,and positive emotions are negatively associated with financial fraud.In addition,there is a difference in the significance MD&A text indicators for state-owned and non-state-owned manufacturing listed firms.We need to focus on positive sentiment indicators for state-owned manufacturing listed firms,as well as readability,forward-looking,correctness and positive sentiment indicators for non-stateowned manufacturing listed firms.Again,in order to verify whether internal corporate governance indicators help stakeholders identify financial fraud in listed firms,we take listed firms from 2015-2021 as a sample,and incorporate corporate governance indicators into the XGBoost model to empirically test the role of corporate governance indicators in identifying financial fraud in listed firms.It is found that the recognition accuracy of the XGBoost model is significantly improved after adding corporate governance indicators than when only financial indicators are included in the model.However,when using the XGBoost model,it is necessary to adjust the parameters and search for optimization,which is not only troublesome but also has a certain degree of randomness.Therefore,we optimize the XGBoost model based on Ant Colony Algorithm(ACO)and propose a new model to identify financial frauds of listed firms — ACO-XGBoost.Considering the learning ability of the fraudster,in order to discover the optimal time period for fraud identification,we divide the total time span into six-year time period,five-year time period,fouryear time period,three-year time period,and two-year time period according to the pinch-androll approach.It is found that the fraud recognition rate is higher in the early time period than in the recent time period,and that frauds with a two-year time period have the highest recognition rate.This is because over time,frauds become more sophisticated and complex and difficult to recognize,and when a fraud is recognized,the counterfeiter,by virtue of his or her ability to learn,spends a certain amount of time adapting the fraud to make it concealed and more difficult to identify.The mechanisms by which internal governance indicators influence financial fraud are further analyzed.Fourth,in order to verify whether the company’s external monitoring indicators help stakeholders to identify financial fraud of listed firms,we select listed firms from 2015-2021 as the research sample,and identify financial fraud from the perspectives of media attention,auditor’s attention and political affiliation outside the company.The empirical finding is that media monitoring pressure can identify corporate financial fraud,and the greater the media monitoring pressure is,the more likely financial fraud occurs;newspaper media monitoring pressure is ineffective in identifying corporate financial fraud;online media monitoring pressure can identify corporate financial fraud,and the greater the online media monitoring pressure is,the more likely financial fraud occurs.When firms face media monitoring pressure,newspaper media monitoring pressure and network media monitoring pressure,its interaction term with abnormally high audit fees can be identified on corporate financial fraud,that is,abnormal audit fees as a moderating variable can positively improve the identification of media monitoring pressure on corporate financial fraud.When firms face media monitoring pressure,newspaper media monitoring pressure,and online media monitoring pressure,only the interaction term of their political affiliation with executives’ political affiliation moderates the identification of corporate financial fraud,i.e.,the political affiliation of executives as a moderating variable positively improves the identification of corporate financial fraud by online media monitoring pressure.Further,we use the ACO-XGBoost model to validate the role of external monitoring indicators and the identification of financial fraud,and find that external monitoring indicators can significantly improve the identification accuracy of financial fraud firms.Finally,we interspersed the financial fraud identification analysis of Kangmei Pharmaceuticals behind each part of the study,verified the applicability of multi-source data to the identification of financial fraud of Kangmei Pharmaceuticals from four aspects,namely,financial data,textual information,corporate governance,and external supervision,and supplemented the validation by using Yangzi new material,Jinzhengda and Le Eco as cases.Based on the above research content and main findings,the innovative contributions of this paper are summarized as follows:(1)When using MD&A textual indicators to identify financial fraud,we mined MD&A textual information from the three dimensions of linguistic structure,linguistic quality,and linguistic representation,constructed a textual indicator identification framework,verified the role of textual indicators in assisting in the identification of financial fraud,and compared the Logistic regression model,XGBoost model and MLP NN model recognition accuracy.(2)When using internal corporate governance indicators to identify financial fraud,we use the ACO algorithm to construct a new model,the ACO-XGBoost model,for identifying financial fraud in listed firms,and find that the fraud identification rate in the early period is higher than that in the recent period,and that two-years period have the highest fraud recognition rates.(3)When using the company’s external monitoring indicators to identify financial fraud,we deepen the existing research on media monitoring pressure on financial fraud identification,and find that media monitoring pressure and online media monitoring pressure can effectively identify financial fraud,and abnormal auditing costs and political affiliation as moderating variables can positively improve the identification of corporate financial fraud by media monitoring pressure,and in addition,ACO-XGBoost algorithm improves the accuracy of external surveillance indicators for financial fraud identification.