Personal Credit Risk Analysis Of The Survival Model Under The Regularization Method

Credit risk is a key area of banking,It is a common concern of stakeholders,such as organizations,consumers and regulators.Credit risk research has been a hot topic in finance,and has attracted the attention of statistical researchers in recent years.Wikipedia defines credit risk as the risk of default on a debt that may arise from a borrower failing to make required payments in 2017.The core of credit risk is a breach of contract,if the debtor can not pay the relevant debt or fulfill the statutory obligations based on the debt contract,there has been a breach of contract.In the analysis of the bank customers' credit risk research,it is not accurate to evaluate the customer's credit simply by judging whether a customer defaults or not.Because most of the customers will not breach of contract during the study period,that is,during the study period we can not observe the survival time of most individuals,so that many censer data produces that is common for survival analysis.In recent years,some studies have applied the method of survival analysis to the credit risk analysis model.Survival analysis is a dynamic analysis method,it can not only predict the probability of the occurrence of an event,but also can predict the time at which an event occurs.Survival analysis can process censored data and truncated data compared to other credit risk analysis models,and we can use the estimation of survival probability to reflect the relationship between risk and characteristic factors.At the same time,the survival analysis model introduces the time variable,which can better reflect the living state of the object.This paper is based on the microfinance desensitization data of 420 high-dimensional characteristic variables of 60508 sample bank customers during the three-year(36)study period.At first,for the high-dimensional data,we compare the regularization algorithm of today's hotspots and try the algorithm when the traditional variable selection methods are faced with challenging.We consider the default time span of the default in the model creatively,taking the period of customer defaults firstly during the analysis period into the survival analysis model,process the data as a fixed format for the survival data,establishing a Cox Proportional Hazards Model based on the LASSO-MCP regularization method and Additive Hazards Model based on LASSO-SCAD regularization method.At the same time,we use the product of the important parameter estimation and the value of the important variables as the credit score,and establish the classification rule,and then evaluate the credit risk of each customer comprehensively.We compare the results with experience,and give the economic significance of some important characteristic variables based on the survival model.Finally,we compare the two models of survival analysis from the two aspects of the results of the important variables and the predictive effect of the model.It is found that the Cox Proportional Hazards Model based on the LASSO-MCP regularization method has a relatively good classification effect with fewer variables.Finally,we use the two survival analysis models,to compare with the traditional Logistic Model and the modern Decision Tree Model.Based on the theoretical analysis and the model results,we compare the four models from the ROC curve which explains the accuracy of the model and the KS statistic which represents the distinguishing ability of the models.It is found that the survival Cox Model based on the regularization method is superior to the other three models.From the two aspects of the accuracy and differentiation of the model,it is concluded that the Cox Proportional Hazards Model based on the LASSO-MCP regularization method has the highest accuracy and the largest distinguish for the three-year loan data.