Research On Default Factors And Prediction Of Private Enterprise Bonds With Bond Performance Information

The General Secretary clearly put forward "promoting the development and growth of private economy" and "supporting the development of small,medium and micro enterprises" for the first time in the 20th National Congress.This important statement clearly defined the importance of private economy to China’s economic development and pointed out the direction for the development of private enterprises.However,since 2020,under the repeated impact of the COVID-19,China’s private enterprises have faced unprecedented development difficulties.In 2021 alone,100 of the 133 enterprises that have defaulted are private enterprises,making them the hardest hit area for bond market defaults.Therefore,studying the causes of bond default of private enterprises,building bond default factor indicators,and establishing bond default risk prediction models will help enterprises identify default risks,adjust enterprise structure,maintain their own development,and help improve the development of China’s bond market in the long run.This paper focuses on studying the default factors of private enterprise bonds and building a bond default risk prediction model,mainly from the following parts:First of all,it explains the concept of bond default,the theoretical knowledge involved and the literature.The second part describes the current situation,mainly from the total size of the bond market,the development trend,the characteristics of private enterprise bond issuance and the distribution of bond defaults.In the third part,bond performance indicators are added on the basis of macroeconomic indicators and financial indicators of bond issuing enterprises frequently used by previous scholars and Lasso is used for feature screening to build an indicator system suitable for bond default risk measurement in this paper.In the fifth part,four machine learning base models and Stacking fusion models are constructed to predict and analyze the bond default risk with and without bond performance indicators,and the prediction effects of the single model and the fusion model are compared.Through the research on the influencing factors and prediction of bond default,the following conclusions are finally drawn:(1)There are obvious differences between the financial conditions of non-defaulting private enterprise bonds and defaulting bonds.(2)Bond performance index is an important factor affecting bond default,and incorporation of bond performance factors can significantly improve the prediction ability of the model.After feature screening with Lasso,19 significant variables were obtained,including 6bond performance indicators alone,including total bond issuance,issuance method,whether the issuing company is listed,whether there is guarantee,coupon rate and interest rate type.After the risk indicators of the model are integrated into the bond performance indicators,as for the single model,LR,RF,Light GBM and XGBoost models have significantly higher F1 value,AUC value and precision than the prediction results when the bond information is not integrated.(3)Whether bond performance indicators are added or not,the fusion model can significantly improve the prediction ability of the model,and the prediction accuracy and prediction accuracy are better than the single model in the corresponding situation.After using Stacking model to fuse the four base models with bond information,the model has the best prediction effect.Among them,the AUC value reached the highest,0.756,the precision rate of the model prediction was0.826,the accuracy rate was 0.842,and the model differentiation was 0.721,indicating that the model can effectively distinguish the defaulted and non defaulted bonds of private enterprises.That is,compared with a single model,the fusion model can significantly improve the accuracy and accuracy of prediction.