Research On Herding Effect And Credit Evaluation Of Online Peer To Peer Lending

In recent years,internet finance has developed rapidly.Online P2P lending becomes one of the typical modes of internet finance.It is of great importance to both borrower and lender,which can effectively promote the development of inclusive finance.However,high degree of information asymmetry may lead to the problems of irrational herding effect and inaccurate credit evaluation.There are many P2P companies,their automatic bidding mechanism greatly improves the transaction efficiency and saves transaction time,but there is relative little research available on considering the impact of the automatic bidding mechanism.What's more,online P2P lending is able to alleviate the contradiction between supply and demand of capital.However,the risk of loan default is highly enlarged due to the asymmetric information.How to evaluate credit effectively has become a hot research topic.To this end,this thesis focuses on the above two points.On the one hand,we study the impact of automatic bidding mechanism on herding effect in online P2P lending.First we collect trading information of online P2P lending from PaiPaiDai and analyze behavior mode of online P2P lending market.Then we study herding effect of online P2P lending including its influence mechanism,and research hypothesis is proposed according to the theory of information asymmetry and herding effect.Finally,we study the herding effect under the environment of automatic bidding mechanism through panel data regression.We find that herding effect exists in online P2P lending.In addition,the automatic bidding mechanism can weaken herding effect.Furthermore,the automatic bidding mechanism presents a rational herding behavior similar to loan amount and loan term factors.On the other hand,we study credit evaluation of online P2P lending.First we describe a new model XGBoost,which can deal with the classification problems in the complex data environment and identify important variables.Second,we compare the XGBoost model with some traditional model including logistic regression,lasso-logistic regression,decision tree and random forest,on the open large scale data of credit card defaults.Finally,we apply it to online P2P lending credit evaluation through the empirical analysis of real Pai PaiDai data.The empirical results show that XGBoost model outperforms the other methods under the large-scale and unbalance data.It can not only significantly improve the AUC value,but also select key factors which impose great impacts on the default behavior.These empirical findings are not only helpful to improve the transaction mechanism for promoting the development of online P2P lending,but also useful for default prediction and control of online P2P lending.