The Study Of Small And Micro Enterprise Credit Scoring Model Based On Softmax Regression

At present,China's economy is entering the shift slowdown,improve the efficiency of the new normal,as China's economic half of the small micro-enterprises(SMEs),have become the backbone of China's national economic.However,SMEs's poor credibility,opaque corporate financial situation,unstandardized financial system,information asymmetry,lack of security and other factors make it difficult for SMEs to obtain financing.The shortage of funds has become the biggest stumbling block in the development of small and micro enterprises.This paper focuses on the SMEs's credit scoring,analyzes how commercial banks use small and micro enterprise credit scoring model,reduce loan costs and the risk of information asymmetry.Compared with the results of the SMEs's credit scoring in the existing literature,our uniqueness lies in the fact that when we carry out credit rating modeling on SMEs,in order to fully reveal the SMEs in the credit score model In the information asymmetry,we support the modeling of the sample were more refined classification,that is,in addition to marking the customer status as"normal",we put the bad sample is"concerned",and the corresponding bad customers of the other three A delicate state"subordinate","suspicious",and"loss".In this way,the results of this model are more refined than the corresponding results of the current literature.The research of this paper provides a new idea for the practice of small business credit score.Firstly,this paper introduces the related concepts,theoretical basis and main methods of credit rating of SMEs;Secondly,according to the specific characteristics of SMEs,initially constructs SMEs scoring index system;Finally,based on the preliminary establishment of the index system,slects a real loan enterprise as the research object,on the basis of the initial established index system,utilizes the IV test,the AR test and the?~2 test to screen the indicators,and conducts empirical research using the loan application data and business owners personal credit data.For the First time,the Softmax regression method is applied to the small business credit rating model.The results showed that Softmax regression model has a strong ability to distinguish between good and bad customers,the prediction accuracy of the test group reach 85%,recall rate is 85%and AUC is91%;The model also has a good distinction between normal,attention,secondary and suspicious capacity,.The accuracy of the test group is 75%,recall rate is 77%,the average AUC is 80%.Logistic regression was used to test the robustness of the model.Logistic results were consistent with those of Softmax regression model,and the robustness of the model was better.