Measurement And Empirical Study On Default Risk Of China's Credit Bonds

The rapid development of China's economy has driven the evolution and improvement of China's capital market.What followed was the launch of various financial products.Among them,credit bonds are more and more popular in the market,and gradually become a substitute for bank loans.From 2011 to now,China's credit bond market can be described as an era of spurt development.According to the wind database,as of the end of 2018,the market has issued more than 16,000 various types of credit bonds,with a financing scale of more than 18 trillion yuan.Along with the rapid development of the credit bond market,the frequency of defaults on credit bonds has also increased significantly.As of the end of February 2019,a total of 347 credit bonds were defaulted in China's credit bond market,involving more than 210 billion yuan,involving 115 credit bond issuers.Therefore,in the context of the frequent occurrence of credit default events in China,the analysis and early warning of credit risk of credit bonds is of great significance.This paper will mainly study the measurement and early warning of credit risk of China's credit bonds through literature research,empirical research and comparative analysis.First of all,domestic scholars have been studying corporate bonds or urban investment bonds in the past.Although the government has continuously introduced relevant policies to regulate local government debt problems,these two types of credit bonds actually still have local governments to endorse their credit.When the credit risk of the relevant credit bonds occurs,because there is still no actual default of the first-order corporate bonds in China,local governments often help the relevant entities to pay the principal and interest to avoid the occurrence of substantial defaults because they are not willing to become the “first-order” default.Therefore,these two types of credit bonds do not reflect well the relationship between the company's qualifications and the credit risk of its issued bond bonds.At the same time,although the default of China's credit bonds has occurred more and more,compared with the credit bonds that have not defaulted,the number between the two is still wide,which will be used to measure and predict credit using various models.When the credit risk of a bond is caused,it has a great negative impact and severely degrades the performance of the model.Previous studies have not discussed the non-equilibrium of the distribution of credit bond samples.Based on the above research,this paper attempts to study the credit risk of credit bonds from a more comprehensive and perfect perspective,in order to construct a higher-precision,easier-to-use and more perfect credit risk measurement early warning model.In terms of sample selection,this paper selects corporate bonds,medium-term notes,short-term financing bonds and ultra-short-term financing bonds that were publicly issued between 2015 and 2018 as research samples.The reasons for choosing the above credit bond varieties as the source of the research samples are mainly Several credit bonds are more fully disclosed for information,relevant data are readily available,and there are no related defects in corporate bonds.In terms of data selection,the relevant financial data of the credit bond issuer in the year before or after the credit bond default occurred was selected as the independent variable used in the study.That is,the first time a credit bond entity has a credit bond default event in A,the financial data of the credit entity A-1 is selected as the data source;for the entity that has not defaulted on the credit bond to A year,the credit is also selected.The financial data of the main A-1 year is used as the data source.The choice of financial data as the independent variable of research is mainly because the financial data is relatively easy to obtain and standardize,and the reason why the macroeconomic indicators are not selected as the independent variables used in the research is that the changes in the macroeconomic situation will inevitably affect the company's finances.The data has an impact,so the company's financial data will include macroeconomic factors.Moreover,in the future model application,the samples to be predicted are all in the same macroeconomic environment.In the aspect of model selection,this paper sorts out the relevant models commonly used in the early warning of credit bond default measurement by reading the relevant literatures.By comparing the advantages and disadvantages between the models,the Logistic regression model is selected as the benchmark model of this paper.At the same time,it is found that XGboost is a new generation algorithm with excellent performance.It has not been used to study the credit risk of credit bonds,so it is innovative to apply this algorithm to this paper to better study the credit risk of China's credit bonds.At the same time,in view of the fact that the default of China's credit bond default is still small,resulting in a non-equilibrium sample distribution,the adassyn adaptive comprehensive oversampling algorithm is used to process the unbalanced distribution of credit bond issue subject samples,and the Logistic model and XGboost algorithm are improved.The negative impact of unbalanced samples on model performance is greatly reduced,and the AL-Logistic model and the A-XGboost model are constructed.The Empirical results show that the performance of Logistic model and XGboost algorithm improved by Adasyn algorithm has been greatly improved,and the true-negative ratio of AL-Logistic model and A-XGboost model has been increased by 72% and 45% respectively before using Adasyn algorithm.At the same time,another major purpose of this paper is to compare the results of A-XGboost model with those of AL-Logistic model.In terms of early warning ability,although AL-Logistic model has a high recognition rate for default samples,for non-default samples,AL-Logistic model will misjudge more actual non-default samples as default samples,which will lead to excessive misjudgment in the practical application of the model,which will bring a great burden and cost to the relevant subjects and the supervisors of credit bonds.As for the model,it is not economical in terms of early warning ability.The A-XGboost model has excellent performance in predicting the accuracy of default samples and misjudgement of non-default samples.The overall accuracy is very high,and it has achieved a very good balance in accuracy and economy.In the aspect of variable interpretation,A-XGboost can only output the important index of independent variables because of its own characteristics,and can not explore the quantitative relationship between the relevant independent variables and dependent variables.In this regard,A-XGboost model is at a disadvantage compared with AL-Logistic model.The empirical results of AL-Logistic model show that asset-liability ratio and short-term liabilities/liquid assets indicators are positively correlated with default of credit bonds.The return on total assets,interest-free liabilities/owner's equity,operating income(year-on-year growth rate),sales gross margin and current ratio indicators were negatively correlated with defaults on credit bonds.According to this result,a joint model based on AL-Logistic model and A-XGboost model is constructed to achieve a higher level of explanatory and predictive ability of the overall model,so as to better measure and evaluate the default risk of credit bonds in China.Finally,by setting different P values,this paper constructs the corresponding monitoring and early warning mechanism of credit bonds to meet different needs in the reality.