Research On Risk Identification Of P2P Network Loan In China

P2P lending is an Internet financial mode which relies on online lending platform for transactions.It effectively complements the traditional financial sector.The first of online lending platform was established in 2007 in China,From then on,the P2 P online lending industry has expanded rapidly.However,the development situation has taken a sharp decline for the worse,and P2 P online lending is entering into the end of the road,leaving a large number of bad debts and fees,which greatly persecuted the interests of investors and destroyed the order of Internet finance after 2017.Most P2 P lending platforms have to run away,shut down or transform to find a new way out.Therefore,the main body of a book works over the risks of P2 P lending,for finding the deficiencies of China's P2 P online lending platforms in business model,risk control method and borrower identification method compared with foreign platforms,and establish a more scientific risk identification model of default online lending.On the one hand,it provides suggestions for the self-optimization of domestic online loan companies,so as to reduce the shutdown caused by unfavorable operation,uplift their level of ability to hit back risks,and bring down the effect of cities to environments such as cash withdrawal difficulty and running away.On the other hand,the online lending platform can use the default risk model to screen high-quality borrowers,thus formulating lending rules,reducing investor risks and maintaining the order of the Internet financial market.Firstly,this thesis works over the actual situation of industry and abnormal online lending platforms,and finds that China's online lending market is shrinking seriously,with the number of negotiators,turnover and number of platforms continuously decreasing,and the problems of online lending platforms frequently occur.The network loan platform has poor ability to resist risks.The number of abnormal operation platforms is increasing,and the operation is lack of norms,and they are mainly located in the eastern coastal areas of China and other economically developed areas.The number of platforms that go out of business due to their own operation problems or increased liquidity risks is far higher than the number of platforms that have transformed,and the transformation of P2 P online lending platforms is difficult.Then it makes a systematic analysis of the management model and risk control methods of online lending platforms at home and abroad,and lists the representative companies of various platforms.It is found that the business model of domestic platforms is not scientific,and the application scope of risk control methods imitated abroad is small,and the effect is weak.In addition,it is found in the study that the commercial credit investigation environment in China is not yet mature,the information is unbalanced,the platform is difficult to define some information of borrowers,and the audit method is mostly manual audit,which rely on subjective influence markedly.Finally,the real transaction data were used for empirical research.After missing values and outliers were processed,the correlation test was conducted between borrower information and default,indicators that failed the correlation test were eliminated,and the importance of features was determined by using the Random Forest.It is found that the historical successful loan amount is the most important,followed by principal that all hasn't paid,the loan amount,the number of historical normal repayment periods,the number of loans,etc.The historical transaction data of the borrower can significantly reflect its default risk.The ten indicators with greater importance are used in the final model construction.The use of the same data for 5times the construction of the BP neural network,found a way to use default risk identification model to build the average prediction error is 0.2409.GA-BP prediction model was optimized by combining genetic algorithm and neural network,and the prediction error was found to be 0.2088.,the error was reduced by 3% or so,Moreover,the accuracy and F1 score of GA-BP prediction model are higher than that of BP neural network prediction model,indicating that the BP neural network model optimized by genetic algorithm has better performance and better prediction effect.