Credit Bond Risk Measurement Based On Improved KMV-XGBoost

At present,the bond market in our country is in the stage of flourishing development,and the bond default event which started in 2014 has attracted increasing attention.The "11 Chaori Bond" was in default in 2014 and failed to pay the interest as schedule,which became the first bond to default in the bond history of our country and broke the zero default record in the bond market.In the following several years,the defaults in the bond market of our country began to increase year by year,mainly presented by increasing default subjects and increasing default amount.In 2014,the accumulative default amount reached to 1.3 billion Yuan with a small scale.In 2015,the default scale and quantity rose to a new level,wherein 23 bonds were in default in total with the default amount of 12.6 billion Yuan.In 2016,the default quantity increased to 64 and involved default amount of 39.3 billion Yuan.In 2017,the default quantity and scale began to decline,but the default amount still reached to 37.6 billion Yuan.In 2018,the default amount and quantity presented a trend of sharp rise,involving 62 bond issuers,134 bonds and 122.3 billion Yuan.Afterwards,the default scale rose to a new level in 2019,involving 178 bonds and 143.528 billion Yuan,wherein 38 newly-increased default subjects were involved,including KDX,TUNGHSU,FOUNDER and XIWANG.It can be seen from the above bond default situations that it is extremely urgent for our country to perfect the credit risk measurement method.In this paper,statistical analysis was conducted on the bond defaults in the financial market of our country,combining with domestic and overseas research literature about credit bond default and considering the similarities and differences between domestic and overseas credit bond defaults,it was found based on periodical context combing that our country still has no method to establish a massive default database owing to short history of bond default and little bond data,and it is disadvantageous to measure default risk only by using foreign classical KMV model in our bond market.Therefore,the bond default risk in our country was measured from multiple aspects combined with the KMV model and XGBoost algorithm.964 Shanghai-Shenzhen A-Stock companies listed from 2017 to 2018 were selected to be research samples,according to related rules,the closing price of stock,the quantity of circulated stocks and non-circulated stocks were taken as the market data,the operation capacity,development ability,profitability,debt paying ability,cash flow indicator and share index were selected as the financial data,the default distance which was calculated based on the KMV model was taken as an indicator and combined with the financial index so as to measure the credit bond default risk of our country more comprehensively.Firstly,the default distance calculated based on the market data of the selected bad credit companies and contrast companies(90 companies listed in the same year)according to the proportion of 1:2 was analyzed,the significant correlation between the default distance and the credit bond default risk was demonstrated,and the default distance was proved to be suitable as a measurement index for credit bond risk.Considering the difference in the default distances of different industries,the financial data of different industries in different years may be different,therefore,standardized processing was conducted on the default distances and financial indexes of 964 listed companies according to different year and industry.Secondly,F test was conducted on the selected 20 indexes to remove those having no significant correlation with labels,the principal component analysis method was used to conduct data dimension reduction based on this,12 indexes was selected and introduced into the KMV-XGBoost model,Bayesian Optimization Algorithm was used to conduct parameter optimization on the model to get the improved KMV-XGBoost model.Finally,the XGBoost model was compared with the traditional LOGIT model and Random Forest model to present the superiority of the model in this paper.