| The “11 CHAORI Bond” was unable to repay the principal and interest within the specified date,becoming the first domestic public offering to default On March 4,2014.With the continued downward pressure of the macro economy,the entities and amounts of credit bond defaults showed a volatile increase in 2018,125 bond entities defaulted and the default amount was 118.278 billion yuan.The bond default incident has eased in 2019,and 145 bonds have defaulted,and the default amount has dropped to 107.311 billion yuan.The default events in the bond market have gradually normalized.Although the default of credit debt entities has declined in 2020,the amount of default has indeed increased substantially.Many companies are affected by Covid-19,the ability to withstand market risks is weaker,the solvency is weakened,and the risk of default is increased,especially small and medium-sized private enterprises.Therefore,for investors Investor risk prevention and stable development of bond market economy,the establishment of a bond default warning model has important practical significance.Companies want to identify bond entities that may default.141 default bonds and 2874non-default bonds were selected as the research samples from the Shanghai Stock Exchange and Shenzhen Stock Exchange,whether the bonds default was set as a two-class problem for identification According to the analysis,the ADMR-AdaBoostSVM classification model based on SVM is constructed to solve this problem.Based on four evaluation factors,20 early warning indicators were selected from the corporate financial statements.Because the bond default data set has the characteristics of sample imbalance,the ADASYN method is used to synthesize new sample points.Compared with the original data,the sample set after oversampling overcomes the problem of sample imbalance and greatly improves the recognition ability of the classification model.In order to solve the redundancy problem caused by high-dimensional data,the feature extraction MRMR method is introduced into the bond default field,and the Ada Boost integrated algorithm is used to optimize the traditional SVM model for risk identification.The research results show that this model overcomes the problem of sample imbalance and significantly improves the classification accuracy.At the same time,the MRMR method is used to solve the high-dimensional data redundancy problem,and the accuracy of identifying default bonds is further improved.The MRMR algorithm screens out four early warning indicators,which can be Taking these four indicators as important factors to measure whether the bond defaults or not,after introducing the integrated algorithm to optimize the SVM,the AUC index of the final ADMR-AdaBoostSVM classification model is 0.97,which is 15% higher than the AUC value of 0.82 of the traditional SVM.The lumped sample accuracy is 92.99%,and the negative sample accuracy is 85.71%.The model accuracy is the best in ADMR-AdaBoostSVM.It has been repeatedly verified that the ADMR-AdaBoostSVM model has high predictive ability and recognition ability,and has strong robustness and effectiveness.The research and application of the bond default risk early warning model should involve the public financial data of each enterprise.In order to make model predictions timely,robust and easy to operate,this article suggests that regulators should strengthen and improve financial market information disclosure mechanisms;core financial institutions should actively introduce advanced risk measurement technologies;rating agencies need to maintain rating agencies The neutral position and rating standards;companies should regularly use models to review their own risk status;investors need to use more scientific tools to detect bond default risks in a timely manner to achieve timely stop losses. |
PDF Full Text Request