Research On Personal Consumption Credit Risk Assessment Model

With the development of Internet consumer credit market,the application of big data and statistics have become an important strategy for banking industry.Increasing number of customers who couldn’t get access to credit funds can get convenient online credit services,because quantitative credit evaluation becomes possible in the era of Internet Finance.Focusing on the personal credit scoring problem in consumption credit business,based on the official credit data of the People’s Bank of China,this paper constructs more than 100 independent variables which are related to personal credit risk.By using logistic regression,BP neural network and XGBoost modeling methods,this paper establishes individual credit evaluation models with different principles,and makes a comparative analysis of the prediction effects,advantages and disadvantages of different models.Finally,logistic regression is selected as the optimal modeling method to solve the practical problems of this case.In recent years,the credit risk level of credit customers has changed,thus the effect of the credit scoring model has generally declined.By using the real customer data of a commercial bank in recent two years and comparing traditional regression model with machine learning model,this paper creates a more market-oriented credit scoring model.In the construction of characteristic variables,all the characteristic variables in this paper are highly correlated with credit risk and have passed the cross-test of importance.It indicates eight independent variables have significant impact on the dependent variable:the number of loan and credit card inquiry institutions in the past 3 months,the last overdue months,the approval rate of credit card in the past 12 months,the average loan amount,the approval rate of loan in the past 24 months,the outstanding loan balance,the number of overused credit cards and the credit history.These variables are fitted into a logistic regression expressions,on this basis,the personal credit standard score card is established.In conclusion,logistic regression model does show good prediction efficiency in empirical analysis.The model has good generalization ability in the validation set and the over fitting problem is reduced.Moreover,the logistic regression model can control the occurrence of type 1 errors at a low level,which means it is fully capable of identifying bad applicant.In solving the problem of personal credit scoring with limited sample size and short observation period,logistic regression can be a practical modeling method because of its good prediction accuracy,low training cost,strong business interpretability and less sample size requirements.The analysis result proves that the personal credit scoring model can be used for bank credit risk assessment.