Dynamic Early Warning Of Bond Default Based On Optimal Features

Bond default refers to the behavior that the issuer of the bond fails to perform its obligations in accordance with the previously agreed bond agreement within a specified time.As a hot topic,bond default risk prediction,acts as an important role in decision-making of various stakeholders,including:bond investors,commercial banks,and company managers.This study consists of five parts:The first chapter is the introduction.The second chapter is the basic principle of the default prediction model based on ptimal features.The third chapter is the construction of default prediction model.The fourth chapter is an empirical study of Chinese bonds.The fifth chapter is the conclusion.The research focus of this study includes:the first is the selection of critical points for model default judgment.When the critical points of the model are different,the number of default bonds and non-default bonds determined by the prediction of the same set of samples will be different.Furthermore,the prediction accuracy of the model is different.There should be an optimal critical point that can distinguish the default bond from the non-default bond to the greatest extent.The second is the determination of the optimal feature combination.In today's big data environment,the number of indicators is very huge.Choose different feature combinations to predict the same bond,the default probability will be different,which will affect the prediction of default.There is bound to be an optimal combination that can maximize the accuracy of the prediction model.The characteristics and innovations of this research include:First,by changing the ratio of type I errors to type ii errors(we tried five different ratios:1:1,1:2,1:3,1:4,1:5),the optimal critical point of logistic regression is obtained by taking the weighted sum of Error-II and Error-I as the objective function.Second,this paper adopts the random forest algorithm,traverses the default prediction accuracy of model with the number of decision trees in the random forest from 20 to 500.Takes the minimum Type ? error and the maximum AUC as the target to invert the optimal number of decision trees,and obtains the optimal feature combination at the same time.This paper adopts the data of matured bonds in China's bond market from 2014 to 2018 for empirical research.The results show that "Monetary Funds/Short-Term Debt","Net Profit","Number of Bonds Issued by Issuers","Industry Prosperity Index" and "Industry Entrepreneurs Confidence Index" have important effects on bond default prediction in each period.Comparative analysis shows that the accuracy of the default prediction model constructed in this paper is better than classical models such as Support Vector Machine(SVM)and Linear Discriminate Analysis(LDA).This paper uses the K-W test to analyze the industry characteristics of China's bond market and the characteristics of bond issuers:according to industry classification,"education" and "manufacturing" have the best credit qualifications,and "lease and business services" have the worst;according to The classification of debt-issuing enterprises is attributed to the highest credit qualification of individual enterprises,followed by state-owned enterprises and private enterprises,and the worst credit qualification of public enterprises.