Prediction And Evaluation Of China's Credit Bond Default Based On Random Forest Algorithm

Bonds are one of the important means for enterprises to conduct direct financing.Although my country's bond market was born relatively late,the scale of bond financing in recent years has risen year by year or even exceeded the scale of equity financing.In 2014,my country's market saw the first bond that had just broken down.Since then,bond defaults have increased year by year,so it is more and more important to control and early warning of default risks.In addition,the rapid development of artificial intelligence and data mining,exploring the model of bond default prediction based on random forests can help to add relevant research theories.First of all,this paper sorts and summarizes the influencing factors of corporate debt default based on the existing literature,forms a logical system of influencing factors of default risk through analysis,and constructs a total of 33 eigenvalues from three levels of macroeconomics,financial factors and bond attributes Feature index system;the second step is to improve the evaluation method of random forest variable importance,and establish an objective and effective method for screening important feature indexes.Under the characteristic index system,the top 11 characteristic indexes ranked by importance are selected.Construct a random forest prediction model based on important feature indicators,and train and test the sample bonds.Finally,the empirical comparison of the random forest model and the logistic model based on important features is conducted;the paper finally uses the "11 Kaidi MTN1" bond that is not in the sample.The default event is taken as a case,the process and cause of default are analyzed,and the default forecast of the bond is made using a random forest model based on important features and compared with the actual situation.The side verifies that the model in this paper has excellent applicability and accuracy.This article can be obtained by building a default prediction model and conducting comparative analysis:(1)Compared with Logistic model,random forest default prediction model based on important features,it has a more accurate performance on the test set;(2)The random forest model has excellent characteristics index evaluation The ability,combined with the stepwise backward evaluation method,can screen out the characteristic indicators that have an important contribution to the default state of the bond,and use the important indicators as the input of the prediction model to improve the efficiency of the model and also make the model more widely applicable;In the bond characteristic index system constructed by the article,the most predictive contributions to the bond default state are bond attribute factors and financial index factors.Random forest has important practical significance in the prediction of credit bond defaults,and has a relatively high accuracy rate for predicting the default probability of the target bond lagging one period behind.