The Research On Financial Risk Warning Mechanism Of Listed Companies By Introducing Fraud Identification Model

Nowadays,with the rapid and changeable economic development,changes in the capital market or problems in the company's operation and management may lead to financial risks of listed companies,which will further damage the interests of investors and other stakeholders.Stakeholders in order to reduce the risk of listed company,therefore,need a can predict the listed company financial risk early warning mechanism,makes the managers of listed companies can learn that the company's financial situation whether deterioration as soon as possible,so as to timely implement the corresponding means,such as adjust the direction of operation,management strategy to reduce the loss.However,when sorting out the research results of financial risk prediction at home and abroad,it is found that the current research is mainly focused on the quantitative analysis of financial data of listed companies to predict the possibility of financial crisis of listed companies.However,no matter in the domestic or foreign capital markets,there are listed companies punished by the regulators every year for financial fraud.Financial fraud is also a financial risk,and once found,it will bring serious losses to investors and other stakeholders.In addition,due to the existence of financial fraud,the financial information of these companies is not reliable,and the distortion of data quality leads to the reduced reliability of the traditional financial crisis warning model.Therefore,innovation of this article is through the introduction of financial fraud identification mechanism,build a two-phase model discrimination of financial risk early warning mechanism,in identifying whether has the financial fraud of listed companies at the same time,also can improve the accuracy of the financial crisis early warning model,so as to really reduce the risk of enterprises and investors.Based on the research results of financial fraud identification and financial crisis prediction of listed companies,this paper proposes a financial risk warning mechanism that introduces the financial fraud identification mechanism,andconducts an empirical analysis on the financial data and non-financial data of listed companies from 2015 to 2017.The empirical analysis of this paper mainly consists of several parts: the selection of research sample companies,the determination and screening of research index characteristics,and the comparison of the effect of establishing risk warning mechanism based on different algorithms.On the selection of the sample company,the financial fraud identification and of sample selection on financial crisis prediction model established by the choice of specific conditions,ultimately to financial fraud identification model according to the proportion of 1:1 to 616 samples from 2007-2017 by the company as the training data,the prediction model of financial crisis from 2015-200 samples from2017 companies as the training data.In terms of test set construction,in order to be more close to the real a-share market,this paper adopts the ratio of risk enterprise sample and normal enterprise sample 1:10 to construct the test set,and selects 60 risk enterprises and 600 low-risk enterprises as samples of the test set.In the determination and screening of the characteristics of the research indicators,the indicators of the fraud identification model and the financial crisis prediction model were preliminarily selected according to the existing literature,and then the indicators were screened by the method of quantitative analysis.First,the Mann-Whitney U non-parametric test was used to test whether the index was significantly different between positive and negative samples.If it was significantly different,it was retained,otherwise it was eliminated.Then,the IV values of these indicators with significant differences were calculated,and the indicators with strong contribution to the model were selected and retained.Among the research indicators of financial fraud model,non-parametric test method was used to screen out 21 research indicators with significant differences.Among them,the IV value of 15 indicators was relatively high,making a great contribution to the model.Among the research indicators of financial crisis prediction model,non-parametric test method was used to screen out 14 research indicators,and the IV values of the 14 research indicators were all greater than the threshold value of0.3,which had a great impact on the prediction ability of the model,and was reserved for the construction of the model.In the stage of empirical analysis,the validity of the model is evaluated by using different algorithms and comparing with traditional methods.The results ofthe empirical analysis show that the risk warning mechanism established by using random forest is better than the risk warning mechanism established by Xgboost,and performs better in AUC value,model accuracy,and analysis of high-risk individual samples.In addition,the two-stage risk warning system proposed in this paper is more capable of identifying risk enterprises than the traditional financial risk crisis prediction model.