Reseach On Credit Rating Model And Application Of P2P Lending Borrowers

As an important part of the Internet financial market,P2P network lending has been recognized and developed rapidly at home and abroad.It has played an important role in alleviating the financing difficulties of small and micro enterprises,and injected new vitality into economic growth.However,the current P2P online lending platform is confusing to the credit risk management of borrowers,which leads to the explosive growth of the P2P industry.At the same time,it is accompanied by the collapse of a large number of platforms,and the credit risk problem is becoming increasingly serious.Through the credit rating of P2P online loan platform borrowers,the risk management level of P2P online lending platform can be effectively strengthened,and the key issue of borrower credit rating is the study of borrower's credit rating.Therefore,solving the credit rating of the borrower is of great significance to the current P2P online loan market credit risk management.Credit rating is generally based on information such as credit score or default loss rate to distinguish the credit rating of the borrower.The determination of the credit score matching the default loss rate is the key to credit rating.The focus of the credit score is to address the combination of indicator screening and retention indicators.Indicator screening will focus on the discrimination of default discrimination of indicators under unbalanced samples.In the study of the combination problem of the retention index,according to the default judgment practice of the borrower,it is found that the default judgment of the credit score determined by the nonlinear combination of the indicators in the machine learning method is better than the credit score determined by the linear combination of the indicators.In addition,experts have considered the different effects of financial indicators(hard information)and non-financial indicators(soft information)on the score results.Based on this,this thesis divides the evaluation indicators into discrete indicators and continuous indicators based on the characteristics of indicators in the P2P platform,and studies the different effects of different types of indicators on default judgment.Then based on the relationship between the discrete index score and the default loss rate,the relationship between the continuous index score and the default loss rate,the problem of the credit rating of the borrower matching the default loss rate is solved.This thesis studies the credit rating of P2P loan borrowers.The main tasks are as follows:(1)Based on lasso-logistic screening indicators under unbalanced samples,the credit score model of the borrower was constructed.First,The Wald statistic based on binary Logistic regression was used to screen the single index with significant ability of default discrimination.And then the non-default samples are randomly divided into equal proportions according to the ratio of the default samples,and recombined with the default samples to obtain multiple balanced sample data.Then,On each equilibrium sample,lasso-logistic regression was used to screen the index group with significant ability of default discrimination.Retain the indicators that are filtered out more than half,and build the credit score indicator system of the borrower.Next,the credit score index is divided into two types of discrete and continuous indicators.According to lasso-logistic regression coefficient,the index is weighted,and the influence of data changes on the index weight is reduced by increasing the weight constraint,and the weight after constraint is normalized.The index weight space is determined according to the normalized weight.Finally,Under the two types of indicators,the size of the index weight is used as the radius of the domain,and the weight space is constructed.The Stochastic Multi-objectives Acceptability Analysis(SMAA)is used to randomly calculate the index weights in the index weight space to calculate the borrower's credit score.(2)Based on the improved default pyramid theory,the threshold value determination model of credit rating is constructed.Based on the matching relationship between credit score and default loss rate,the principle of default pyramid with "default loss rate decreases with the increase of credit rating" is selected.Under the discrete and continuous two index types,the optimization of the credit level threshold of borrowers is constructed.Under the two types of discrete and continuous indicators,an optimization model is constructed to determine the threshold points of borrowers' credit ratings to determine the threshold points of credit ratings under the two types of indicators.(3)Based on the integration of the credit rating information under the two types of indicators based on the smaa-evidence theory,the credit rating classification model of new borrowers was constructed.Firstly,the credit scores of the SMAA simulation calculation are compared with the credit rating threshold points of the corresponding indicator types,respectively,and the probability that the borrower belongs to each credit rating under the two indicator types is obtained.Then,using the evidence theory(Dempster-Shafe,D-S)to synthesize the probability that the borrower belongs to the same credit rating under the two indicator types to obtain the probability distribution of the borrower at each credit level.The maximum probability level is used as the ultimate credit rating of the new borrower,and the validity of the credit classification model is verified by comparative analysis of the credit division results.This thesis constructs a credit rating classification model for P2P borrowers.This model provides theoretical and practical reference for relevant researches on credit rating classification,and lays a foundation for the expansion and application of relevant researches.