| With the decline of the market interest rate and the adjustment of the economic structure,microfinance companies are facing a more competitive environment and operational risks.From the national perspective,there have been a number of failures of small loan companies,and small loan companies are facing greater credit risk problems.Therefore,in order to ensure the normal operation of small loan companies and the steady development of small loan industry,it is necessary to study the credit risk of small loan companies.This paper studies the credit risk prediction of personal loans of small loan companies in Sichuan Province.First of all,analyze the relevant theories and literature of personal loan credit risk prediction.Secondly,it analyzes the current situation and existing problems of personal loan credit risk prediction of small loan companies in Sichuan Province.Thirdly,based on the deep belief network and the extreme learning machine,a personal loan credit risk prediction model is constructed.Fourth,collect the personal loan data of small loan companies in Sichuan Province for empirical analysis,and verify the model effect through comparative analysis with different models.Finally,it puts forward suggestions on improving the credit risk control of personal loans of small loan companies.The research conclusions of this paper are as follows:(1)The overall non-performing loan rate of the small loan industry in Sichuan Province was 9.56%,an increase of 2.24 percentage points over the beginning of the year.Although under the regulatory requirements,all small loan companies further strengthened their control measures on credit risk,but the control of credit risk has not shown good results.At present,small loan companies mainly have problems in controlling credit risk,such as indicators and scoring rules reflecting customer credit risk,and scoring methods are difficult to distinguish credit risk.(2)Collect personal loan data of small loan companies in Sichuan Province,and conduct empirical analysis on DBN-ELM model.The DBN-ELM model performs relatively well on the randomly divided test set.Compared with the Network evaluation method currently used by microfinance companies,the AUC is improved by 0.088.The DBN-ELM model performs relatively well in the K-fold cross validation.Compared with the evaluation method currently used by micro credit companies,the AUC is improved by 0.061,which is relatively robust.(3)The cash liability ratio and individual income tax payment flow have a higher impact.The cash flow ability of customers to repay loans every month can be obtained from the cash liability ratio.For the loss of individual income tax payment,it is a better way to calculate the customer’s income information based on a more objective standard than the customer’s income certificate or bank statement.For judicial litigation,judicial litigation and credit investigation have certain reference to judge credit risk. |
