A Research On The Prediction Of Default Risk Of Credit Debt Based On XGboost Algorithm

Bond market is an important market that constitutes the main body of the economy of China.In recent years,defaults happen in the bond market frequently,and the default risk of credit bond in our country is increasingly prominent.As a result,the evaluation of default risk is particularly important.At present,most investors rely on the credit rating of bonds,but a large number of data show that the rating of default credit bonds is not low when they are issued.It is not reasonable to determine the investment value of bonds only from the bond rating.Investors need more reasonable methods to evaluate the default risk of credit bonds,and machine learning provides a possible evaluation scheme.Therefore,this paper aims to compare the results of several mainstream classification prediction models,among which choose the best one.First of all,this paper analyzes the default characteristics of China's credit bond market in recent years,combs three typical cases of credit bond default,and introduces eight related technologies of credit risk measurement,such as XGboost,Logical Regression,Support Vector Machine.Secondly,through principal component analysis,10 default risk factors are extracted from 32 variables of bond issuance data,issuer financial data and macro data,and the importance of variables is analyzed by XGboost algorithm.Based on the above eight technologies,the default risk prediction model of credit bonds is established,and classification prediction effects of each model are compared and analyzed by AUC value and other evaluation indexes.Finally,the Booster Parameters of XGboost model are optimized by Grid Search algorithm and K-fold Cross validation,and the final classification prediction model is determined.The empirical results show that:(1)whether in the original data set or in the dimension-reduced data set,the random forest model and the boosting series of models such as XGboost based on decision trees and integrated ideas are better than other models in the credit default risk prediction,and these models have more obvious advantages in the original data set.(2)In the original data set,these four models have more advantages in classification and prediction,and are more suitable for processing high-dimensional data sets.But in the original data set of this paper,the stability of the prediction results of Random Forest model is not as good as that of boosting series model.(3)Taking the AUC value as the evaluation index,the XGboost model optimizedby the parameters has the best prediction effect on the default risk of credit bonds,with the largest AUC value.In a word,under the background of big data age,using XGboost and other machine learning algorithms can accurately predict the default risk of credit debt.This will not only help to ensure the rationality of bond pricing,but also help investors weigh the yield and risk of bonds,so as to promote the healthy and stable development of domestic bond market.