Tree-based Optimization Methods For Predicting Changes In Credit Rating Of Corporate Bond Entities

Due to external factors such as the Sino-US trade war,the central bank's "new rules of asset management" and other factors such as fierce market competition and rising financing costs,China's bond market credit risk incidents intensively broke out in 2018.Bond credit rating is an important indicator of bond credit risk.How to accurately predict and evaluate it is a very practical issue.This thesis uses classification tree,bagging method and random forest three tree-based models,to predict the credit rating of corporate bonds by using r language,and conducts four optimized sampling to balance the data set and improve the forecasting effect.The main issues discussed include:factors affecting bond credit risk,building the models,parameter selection,accuracy test of the model,and classification effect test and indicator importance assessment.The study found that random forests and bagging methods have more accurate predictions than single classification tree,while undersampling optimization methods have the best effect on the prediction of the decline in bond ratings that investors are most concerned about.The random forest model under the sampled sample has the best results.In the assessment of the importance of indicators,the profitability and solvency of bond-issuing companies are important indicators for measuring the credit risk of their bonds.Combined with the research conclusions,the thesis believes that from the perspective of supervision,it should improve the information disclosure degree and market transparency of its bond issuing enterprises,and optimize credit risk assessment technology from the perspective of rating agencies to improve timeliness.