The Research Of Credit Scoring Model Based On Logistic Regression

With the rapid development of technology and economy,people is changing theirconsumption concept and began to prefer to consume in advance,which has promoted the development of personal consumption credit.Now the form of consumer credit is diversified and concise,and greatly reducing the time for application,but it also means the increase of risk.To effectively control risk,banks and financial companies widely used credit scoring model to score with the customers,implement different credit lines and policies for different levels of customer,which can reduce the risk of personal consumer credit.This paper based on a credit card online product from a financial services company,establishes a credit scoring model which based on logistic regression methods through the sample selection,data cleaning,model establishment,model validation and monitoring process,and evaluates model by the ROC curve and the curve of KS.The credit scoring model in this paper is based on the SAS code independently,where the evaluate result of model is that the AUC value is 0.706,KS value is 0.306,and the time in samples verification result of model is that the AUC value is 0.716,KS value is 0.309,the time outside samples validation result of model is that the AUC value is 0.709,KS value is 0.29.The results show that the model suit for this product.In order to control the default risk of this product,the customers are divided into 10 groups and we calculate the default rate of each group,then cutoff is set according to the default rate and the company's target.Through model establishment and instance analysis,it is learned that the model will be established to ensure sufficient data volume and customer quantity.If the data volume is not enough,the accuracy of the model will be reduced.Moreover,it is proved that logistic regression method is effective in the credit scoring model,and this method is easy to implement and interpreted,so the logistic regression is the main method of setting up credit scoring model.