Research On Risk Evaluation Of Network Lending

Regarding the problem that traditional lending is increasingly difficult to meet the needs of small and micro enterprises and individuals,the State Council has proposed the "Internet +" inclusive finance policy to encourage Internet companies to innovate financial products,and online lending has developed rapidly.However,domestic online lending started late,the country 's level of supervision over the platform is low,and online lending platforms are uneven.Secondly,the platform's inadequate credit reporting system and outdated risk prediction models have created opportunities for “interested people”.In recent years,due to platform rolls,Money runs or lenders deliberately deceived their loans,which caused huge impact on society and attracted great attention from national leaders.The 19 th National Congress of the Communist Party of China has made the "prevention of systemic financial risks" a core element of financial governance in the new era.In accordance with the "three measures and one guideline" proposed by the CBRC to ban illegal online lending platforms and formulate relevant policies to protect the rights and interests of legal platforms,legal protection is not enough for legitimate online platforms.Individual behavior is not fixed.The same,a customer who performed well before the loan may become a high-risk customer in the default process during the loan.Therefore,a classification and prediction model with good performance is constructed to monitor the loan customer in real time and identify default “bad customers”,thereby reducing the investment risks of legitimate online lending platforms will be significant Commercial value and significance.This paper focuses on the construction of a credit risk assessment model for borrowers in a legal online lending platform.Using the loan data provided by the online lending platform Lending Club website,explore the construction of the online lending credit risk assessment model-the Stacking fusion model.At the same time,the recall rate,ACC value,ROC curve,and AUC value are used to compare and analyze the classification effect of Stacking fusion model and random forest,GBDT,XCBoost,and LightGBM from the perspective of different thresholds and reserve ratios.Second,the PSI value is used from different Based on the sample size and the positive and negative proportions of different training sets,the stability of the Stacking fusion model and the other four models are compared.According to the research in this paper,the following suggestions are put forward.First,establish a perfect personal credit information system.By comparing the information of lenders in domestic and foreign online loan platforms,we can see that there is less information about lenders in China and the scope is narrow.Examination of lenders in many aspects will effectively reduce the probability of default.Therefore,a comprehensive personal credit information system is established.It is of great significance to credit risk assessment;secondly,to ensure the accuracy of lender information.According to the research in this paper,the inclusion of outliers,missing values,and erroneous values in the lender's data will affect the classification accuracy of the final classification model.Therefore,it is necessary to ensure the validity of the lender's information as much as possible to reduce the error data on the model.In the end,with the development of technology and the related theories,continue to study the credit risk assessment model,and continue to improve and optimize.