Research On Default Prediction Of Chinese Bonds And Its Driving Factors Based On Feature Importance Selection

In the context of the Party Central Committee’s comprehensive implementation and insistence on preventing the occurrence of high systemic financial risks and stabilizing employment and economy,it is necessary to timely understand and predict the default risk of Chinese bonds and find the driving factors affecting their default.In this paper,I take the national bond default samples as the research object,and establishe eight machine learning models,including logical regression algorithm,K nearest neighbor algorithm,support vector machine algorithm,random forest algorithm,XGBOOST algorithm,LightGBM algorithm,CatBoost algorithm,and the fusion model of these four models.Grid search,random search and Bayesian search are used to optimize hyperparameter,and default prediction is directly made for Chinese bond default issuers in recent nine years.The macro and meso economic variables,bond characteristics variables,corporate governance and corporate characteristics variables,and internal financial variables of the company are selected as variables from the macro,meso,and micro aspects.Based on the importance of characteristics,the variables that are more important to bond default risk are selected.After systematic research,I found that:① Four ensemble learning models were used and all 43 features were sorted and selected based on their importance scores.Finally,the top 25 features were selected based on their importance level.I apply these 25 features to eight machine learning models combined with three hyperparameter parameter adjustment techniques,and find that most of the machine learning models have better prediction effects,among which XGBoost model is better for predicting the default of Chinese bonds,AUC of that model reached 98.4285%,higher than the result of the scholar Dai Yarong et al.(2022)using the random forest model to predict bond default,and its AUC was 94.84%.②Finally,based on the 25 characteristics selected and logical regression,16 driving factors that have a significant impact on China’s bond default were screened,involving seven factors(macro factor,meso factor,bond factor,company’s own characteristics factor,profitability factor,solvency factor,operation factor)and sixteen major factors(M2 currency and quasi currency growth rate,whether it is a local government financing platform,provincial GDP,number of employees,nature of the enterprise,total assets,issuance period,nominal interest rate,net return on assets,return on total assets,return on invested capital,net sales interest rate,asset liability ratio,cash ratio,total asset turnover).Specifically,the faster the growth rate of M2 currency and quasi currency,with implicit government guarantees,the higher the GDP of the province,the more employees,the non private enterprise,the larger the asset size,the longer the issuance period,the lower the coupon rate,the higher the return on assets,the higher the return on total assets,the higher the return on invested capital,the higher the net interest rate on sales,the lower the asset liability ratio,the higher the cash ratio,and the higher the total asset turnover,made the lower the risk of default.③By combining the 16 major driving factors with the latest bond case analysis,it was found that Rongsheng Development,as a bond default entity,did not perform well in these driving factors.In summary,in this article,I use four ensemble learning models to rank the importance of 43 features and selects the top 25 features,based on this,eight machine learning models and three hyperparameter optimization techniques are used to directly predict the default issuers of Chinese bonds and analyze the 16 major driving factors of bond default.In the end,it was found that most machine learning models had better predictive performance,with the XGBoost model having the best predictive performance,with an AUC value of 98.4285%.Compared to the 94.84%result of Dai Yarong et al.(2022),the AUC value of this article has increased by 3.58 percentage points.Finally,based on the analysis of the bond default case of Rongsheng Development,the first defaulting entity on January 23,2023,it was found that it performed poorly in the 16 major driving factors two years before the default,and the bond default event had already laid the groundwork.The research of this article is of great significance for individuals and institutions to identify the default risk of Chinese bonds and improve the risk warning mechanism.