Research On Company Default Risk Prediction Based On Twin-SVR Model

In recent years,with the increasing development of the economy and society,the company,as an indispensable part of the modern society,has become an important driving force for the development of the national economy.In order to increase the company's own liquidity and expand its business scale,the corporate bonds have become the most important channel for the company financing.However,faced with more and more problems,the company continues to finance for a further development.The company default caused by the issuingcorporate bonds has become a problem of company default risk that needs to be solved urgently,especially after the financial crisis.If the company default arises,itwill certainly break the established order of the bond market and the financial market,bring a large negative sentiment to the financial market,and even cause systemic risks.These negative results will have a serious impact on the stable development of the economy and society.Therefore,how to construct a scientific and effective model to accurately predict the company's default risk has important and far-reaching implications for regulators,company operators and investors to prevent and resolve risks.Based on the analysis and understanding above,this paper takes the company that issues short-term financing bills as the research object.The sample size is 4040 after setting criteria to deal withthe sample.This paperselects 25 characteristic indicators and uses the correlation analysis technique to filter the financial characteristics.After filtering,the 19 characteristic indicators are included in the model construction system.At the same time,based on the credit spread,the paper makes a reasonable calculation for the company's default risk,and finally constructs the company's default risk variable.In addition,this paper selects RMSE,MAPE and MAE as the evaluatingcriteria,and uses the method of 5-fold cross-validation to investigate the predictive model.On this basis,considering the traditional statistical model and the difficulties that the neural network model is difficult to deal with nonlinear relations and easy to fall into the local optimal solution,and therefore a nonlinear SVR prediction model is introducedand the prediction research analysis is carried out.At the same time,because the SVR model predicts the biased majority of the company's default risk,it is difficult to overcome the shortcomings of accurate prediction of the company's default risk abnormal variables.This paper introduces Twin-SVR to solve the problems thatthe traditional SVR model faces.Twin-SVR model of default risk is constructed through a large number of empirical studies.The predictive performance on the model is comprehensively analyzed.The empirical results show that:(1)This paper combines RBF,Polynomial,Linear,and Sigmoid kernel functions with SVR model,respectively.The SVRmodel using RBF kernel function can achieve superior prediction accuracy to other kernel functions for company default risk.At the same time,the paired sample T test of the prediction accuracy obtained by the RBF kernel function and other kernel functions is significant at the 5% significance level,which indicates that the optimal SVR company default risk prediction model can be obtained by using the RBF kernel function.(2)Some conclusions can be obtained from the analysis of Twin-SVR's default risk prediction model.Compared with Polynomial,Linear and Sigmoid,the Twin-SVR model using RBF kernel function has the better performance than other kernel functionsnot only in the default risk prediction,but also in the paired sample T test.These resultsindicate that the Twin-SVR model withRBF kernel function is the prior choice for the prediction research of the corporate default risk.(3)Comparing the Twin-SVR with SVR,BPNN and Logistic models,this paperfound that the Twin-SVR model has the best prediction accuracy and outperform other models in the process of paired sample T test.More specifically,the Twin-SVR model is significantly superior to SVR?BPNN and Logistic models.(4)Consideringwhether different industries affect the robust predictive performance of the Twin-SVR model forthe default risk,this paper compares the models(the Twin-SVR,SVR,BPNN and Logistic)of prediction accuracy and the paired sample T test analysis in different industries.The empirical results show that theTwin-SVR modeling the default risk can obtain the optimal prediction accuracy,and the obtained prediction results are significantly superior to other prediction models.(5)Research on the ability to interpret the characteristics of Twin-SVR's default risk prediction performance.The empirical results show that credit rating is the primary characteristic indicator affecting the default risk forecast of Twin-SVR model companies,whether it is state-owned enterprises or net assets,whether it is the overall industry or sub-industry for the explanatory ability of the Twin-SVR model.The rate of return,return on total assets,and turnover of accounts payable should be classified according to the characteristics of the industry(coal,steel,and non-ferrous metals industries).Based on the above analysis,this paper suggests that Twin-SVR model predicting default risk is an operational tool for regulators,company operators and investors to prevent risks.For regulators,the Twin-SVR model can be used to predict and supervise companies that issue short-term securities lending in the future,so as to immediately formulate macro policies to maintain market stability,smooth economic development and risk prevention.For the company's operators,they can use the Twin-SVR model to better analyze their financial situation to maintain the corporate reputation and development.For investors,the model can also be used to capture the corporate default signals in advance,so as to achieve the asset allocation well,and to achieve the capital preservation and appreciation.