Research On Corporate Bond Credit Risk Assessment Scheme Based On Combination Algorithm

In March 2014,"11 Chaori Bonds" announced that it could not pay interest on time,which became the first public bond bond default event in China,which meant that the rigid payment was broken.Since then,the default situation in China's bond market has gradually worsened,especially in 2018 and 2019,the number of bond defaults and default funds have increased significantly compared with previous years.In 2020,the default of the bond market will show new characteristics,the default subject will change from private enterprises to state-owned enterprises,and the default object will change to high-grade bonds.Successive default events have caused a great impact on the stability of the credit bond market and the whole financial market,which is not conducive to the protection of investors' rights and interests.Therefore,using big data algorithm to identify the credit risk of bonds has very important theoretical and practical significance.Based on the review of relevant theories and combing the credit debt,on the basis of related literature,combed the credit debt default risk transmission path,selection of the macro level,industry level and profit ability,debt paying ability,operation ability,growth ability,and corporate governance,debt levels,the influence factors of 29 credit debt default risk as characteristic factor input model,The machine learning algorithm selects Light GBM algorithm,which is relatively new and has not been used in bond credit risk assessment before,and uses XGBoost algorithm,Logistic algorithm and SVM algorithm to supplement the comparison.The data in this paper take the availability of financial statements into consideration.The sample of the training set is the listed companies that have materiality defaulted from January 2017 to December 2019.Then,the non-default sample is selected according to the industry by the ratio of the defaulting group subject and the non-defaulting group subject 1:4.The sample of this test set is selected for all listed corporate bonds that are not due until at least the end of 2019.The sample label refers to whether the sample bonds within the forecast interval have materially defaulted.The empirical results show that the model built based on Light GBM algorithm can predict bond default well,which is slightly worse than XGBoost algorithm,but obviously better than the traditional Logistic algorithm and SVM algorithm.Therefore,this paper believes that Light GBM algorithm is applicable to the field of bond credit risk assessment,and the introduction of this algorithm is helpful for further research to improve the prediction ability of defaulted bonds.After proving the rationality of the model established based on Light GBM algorithm,the combined optimization of Light GBM model and the control model was carried out,and it was found that the prediction effect of the improved model was better than that of the single model which was the worse of the two.In addition,the XGB-LGBM model formed by the fusion of the better XGBoost model and Light GBM model has the best prediction effect on the test set.Finally,in this paper,the model in each of the sample has been done for the output value of meaning interpretation,according to the numerical size,will be divided four test set bond risk level,and presents a model to judge the positive and negative cases of threshold setting guidelines,based on the successful identification of defaulted bonds,the purpose of this paper holds that the critical value of model judge default should only make small changes to near 0.5.