An Empirical Study On Bank Credit Evaluation Based On Detailed Loan Data

Personal credit is the foundation of market economy.It widely exists in many fields of economic life like the commodity trading,capital market investment and financing,band loans and etc.A well-formed personal credit system and a mature operating mechanism could secure and facilitate the development of personal credit economy.However,the loss comes with personal credit could not be ignored as well.Chinese credit collection system is still on the stage of developing.Banks are not willing to take the risk with inadequate information about accrediting party.Thus,the threshold of personal credit has to be increased,along with more complicating credit processes to safeguard the interest of bank itself,resulting in a narrow benefit to credit loans.It is of great importance for the development of personal credit to build a practical indicators and modeling system in respect of evaluation of personal credit.This paper applies the detailed personal credit data of a bank from 2011-2016,employing three methods as Logistic Regression,Random Forestand Naive Bayes to empirically study the bank personal credit valuation based on this comprehensive sample.The context is divided into the construction of credit rating model and credit grade division according to the modeling process.In the aspect of building the credit risk assessment model,we compare the evaluation results of three kinds of evaluation models.We find that the Logistic regression model has the best prediction performance on the data set in this paper,and the prediction accuracy reaches over 85%.At the same time,this paper sorts out the variables of credit risk assessment based on the importance,and finally finds that four variables,such as loan category,loan usage,lender registration type and annual interest rate,have a significant impact on credit risk assessment,which enriches the extant evaluation indicators.In the construction of credit rating classification model,this paper analyzes the credit rating classification based on the results of the credit rating results.As a result,we find that credit default Pyramid principle can guarantee the construction that the credit rating increases with the decreasing Loss Given Default(LGD),but the inhomogeneous interval distribution of credit scoring model.There exists the very short credit scoring intervals.Therefore,this paper improves the credit rating classification model based on the default pyramid principle through the local optimization scheme,the optimization of global objective function and piecewise optimization measures.Finally,we find the credit rating model which meets the default pyramid principle that the credit rating increases with the decreasing LGD as well as the homogeneous interval distribution of credit scoring model.The empirical results indicate the results are improved.