Research On Risk Alerting Model Of Bond Default Based On Machine Learning

The "11 Chaori Bond" default in 2014 was the first default bond in China's public placement bond market.Since the "11 Chaori Bond" default occurred,the default situation in China's bond market has intensified,and the default issuers have also spread from private enterprises to local SOEs and central SOEs.In 2018,the amount of default entities and credit bonds experienced explosive growth,a total of 125 bonds with a sum of 120.061 billion defaulted throughout the whole year,this data increased further in 2019,with a total of 179 bonds and 144.408 billion default throughout the year,default events in the bond market gradually became normal..Affected by the decline in market risk appetite and aggravation of aversion sentiment,private enterprises,particularly small and medium-size operated enterprises,have a relatively poor ability to withstand risks,and are most injured in the concentrated default cycle.Except for private companies,listed companies have also experienced a large number of defaults since 2018.In the current economic environment,the frequent occurrence of corporate bond defaults has aroused widespread concern.The normalization of bond defaults means that the identification and management of default risks will be one of the major problems of bond market.The measurement and early warning of bond default risks of listed companies will provide important reference for the future management of default risk in the bond market.Credit bond defaults usually have multiple causes,including external factors such as the macroeconomic environment,industry development trends,and internal factors such as the company's own financial and non-financial issues.In terms of the selection of risk warning indicators,this article further introduces the company's public opinion indicators on the basis of financial indicators,rating information and corporate governance indicators,so as to fully characterize the credit status of the issuing entity.With the rapid development of Internet technology,the frequency of using network is increasing,and the information people receive is growing at a geometric level.The amount and transmission speed of public opinion on the Internet is increasing,and it has also had an increasingly significant impact on the capital market.In the capital market,the Internet media plays the role of information intermediary by publishing online public opinion,and it can also play a certain role in monitoring various subjects.The impact of negative events in the company's public opinion will adversely affect the company's credit qualifications,reduce the company's ability to raise funds in the capital market,and cause bond default events.This paper uses support vector machines to build risk alerting models of bond default.With the continuous development of computer technology,machine learning has been widely used in many fields by virtue of good feature extraction and prediction effects on data.The final decision of the support vector machine model on the classification problem only depends on small number of support vectors selected during the learning process,which reduces the complexity of the model calculation and can also avoid the high degree of sample dimension.At the same time,kernel function is introduced into the model to solve the classification problem in the non-linear case,and the original sample is mapped to a high-dimensional space for classification,which can better solve the non-linear classification problem.For the credit debt default samples of the past two years,the constructed risk alerting model shows high prediction accuracy,the accuracy rate of the test set using the radial basis kernel function reaches 96.875%,and the prediction is correct for all default bonds,thus the model can well achieve the purpose of risk alerting.At the same time,from the empirical results,it can be seen that the company's public opinion indicators introduced in this article have a strong ability to distinguish default samples.The introduction of company public opinion information has a good complement to traditional financial indicators,improving the predictive power of model for defaulted bonds.Compared with traditional methods such as Logit model and credit rating,the machine learning model constructed in this paper also shows a certain performance improvement.On the basis of empirical research,this article further puts forward policy suggestions such as using public opinion monitoring,big data and other new methods to enhance regulatory capabilities,unify the bond market regulatory system,and strengthen credit rating supervision.