Credit Bonds Default : Elastic Network Punishment Hazard Model

As an important part of the bond market,credit bond is an important financing tool for enterprises which mainly issues relied on the issuer's credit issuance.Credit bond has a greater risk of default without endorsement of national credit and mortgaged assets,compared with mortgage bond.Since breaking the ‘rigid payment' in 2014,the default scale of China's credit bond market has been increasing.Under the background of de-leveraging and strong supervision,investors' risk preference declines,the financing channels of the issuers are limited,and the regulatory policy blockades all kinds of off-balance sheet and non-standard channels.Enterprises with high off-balance sheet leverage and generally tightened financing channels are unable to repay their debts on their own cash flows.Therefore,the default risk of credit debt has aroused widespread concern of scholars and industry.This paper mainly estimates default probability of credit bonds by combining elastic network variable selection with discrete-time hazard model to establish single-period and multi-period elastic network penalty hazard models.The hazard model is used to identify the important influencing factors of default probability and predict the default risk of credit bonds which improves the accuracy and explanatory power of model prediction compared with overeliminating important explanatory variables in lasso.In order to test the prediction effect of the model,this paper compares the elastic network penalty hazard model with lasso and ridge regression penalty hazard model in terms of second type of error rate,AUC value and explanatory ability of the model,and verifies the applicability and superiority of elastic network penalty hazard model in estimating the default probability of credit bonds,which provides a theoretical basis for the construction of the risk-alert system.This paper chooses the best one from three single-period penalty hazard model to estimate the default probability of credit bonds based on the default data from 2014 to 31 August,2018,and then estimates the default probability of credit bonds with multi-period elastic network penalty hazard model.The results show that the elastic net penalty hazard model has the best performance in estimating the default probability of credit bonds;Secondly,the risk indicators for estimating the default probability of bonds are different for different prediction time points.Thirdly,the accuracy,true positive rate and AUC value are higher than those of the single-period penalty hazard model,which shows that the risk index system with multi-period lag factors can fully reflects the default risk of credit debt.Fourthly,the financial variables of the issuer play a significant role in evaluating the probability of bond default,especially the profitability and financial leverage of the enterprise.In addition,the impact of macroeconomic variables,the nature of debt issuers and the characteristics of bonds on default probability can not be ignored,and the impact of macroeconomic variables has a certain lag.