Research On Default Risk Of Corporate Bond Issuers Based On Xgboost Algorithm

In the whole year of 2018,there were 13 listed bond issuers with defaulted event in China’s securities market.However,the credit rating company only gave only two of them lower ratings than AA at the end of 2017.The credit rating agencies have certain limitations on the early warning of credit risk.This paper studies the default risk of debt of all non-financial listed companies that issue bonds publicly in China,and tries to give an early warning of the default risk before the company defaults.We collecte data on the capital structure and stock prices of listed issuers at the end of 2017,as well as the credit rating status of issuers around the end of 2017.Firstly,We use Moody’s classic KMV model to study the data,and modify the four parameters in the KMV model.Specifically,the net asset pricing method is used to optimize the measurement of the value of non-tradable shares,the benchmark loan rate is used to correct the default point setting,the volatility is calculated by Garch(1,1),and the Gram-Charlier series is used to correct the stock price distribution.The revised KMV model can improve the calculation accuracy of the default distance.As the modified KMV model is not stable for the recognition of individual companies’ credit qualifications,this paper selects the xgboost algorithm to train and fit the enterprise default risk indicators and establish a credit downgrade model.Since the number of listed default samples does not meet the number of labeled samples required by the classification algorithm,this paper selects the implicit rating down event as a high-risk label to mark the listed company,and the default is the extreme case of the implicit rating down event.In this paper,Boruta method is used to reduce the dimensionality of important variables related to bond credit risk.We use xgboost algorithm for classification training,and obtain the predicted value of all the issuers through repeated random sampling.The credit downgrade model possess a high identification of credit risk.Based on the empirical analysis of non-financial listed issuers in accordance with the credit downgrade model,this paper finds that six of the top 10 companies with the highest credit downgrade model score have an implied rating downgrade event in 2018,and three of them actually defaulted in 2018.Therefore,the company with high default risk predicted by the model can be used as a negative list of bond investment,so that investors can effectively avoid the risk of default of the target position as early as possible.Among the top 5% companies with highest chance of credit downgrade,we find 10 companies default in 2018.Getting such a negative list before the rating agency has issued a warning will help investors reduce the probability of severe loss.