Research On Risk Measurement Model Of P2P Lending Platform

The P2 P network lending platform has developed rapidly through the east wind of the "Internet +" era.According to statistics,as of June 1,2019,there are 6,617 P2 P network lending platforms in China,and the cumulative transaction amount of the national P2 P online lending industry is conservatively estimated at 6.07 trillion yuan.However,due to the rapid development of the Internet financial market and the lack of supervision,according to the statistical website statistics,the number of problems due to various problems has been reduced to 5,703,accounting for 86% of the total number of platforms.Under the background of fast-moving online lending platform and national policy support to encourage the steady development of Internet finance,investors' safe investment choices and regulatory policies of the regulatory authorities have an urgent need for the risk measurement of online lending platforms.This paper aims to use a variety of historical machine learning algorithms to use all historical online platform data to fit a reliable online loan platform risk measurement model.In this paper,firstly,after processing the multi-dimensional and large-capacity data obtained by the crawler,the crawled "dirty data" is converted into standardized data for modeling.Finally,through the data processing process,the following nine variables are available for modeling: average rate of return,platform duration(year),platform registered capital(ten thousand yuan),withdrawal point,station position,service point,experience point,investment Whether it is recommended,whether the platform supports automatic bidding.The paper takes the "credit scorecard model construction process" as the research basis,combines the theory of machine learning algorithm to deal with the two-category problem,and uses Python to use Logit,CART decision tree and random forest model to fit the platform risk and platform index items.The characteristic variables(indicators)of the platform risk are analyzed,and the following conclusions are drawn: the platform investment return rate,the user's recommendation,and the platform withdrawal score are all in the forefront of the importance level of each of the three characteristic variables in each model,which is the impact on the platform risk.The most important factors;the duration of the platform and the registered capital of the platform have certain influence on the platform risk,and the model results show that the duration of the platform and the registered capital of the platform are inversely proportional to the risk of the platform;whether the platform supports automatic bidding or not does not constitute a platform risk.Significantly affected.The empirical research results of this paper have important reference value and reference significance for investors to choose investment platform and supervision department to evaluate platform risk.Investors and regulatory authorities should fully understand the basic information of the platform,focusing on platform yield and public opinion information,ie user evaluation keywords.(whether recommended),platform cash flow situation(withdrawal points)indicators.The classical model performance evaluation index ROC-AUC curve and PR curve are used to detect the model prediction results,and the model has good predictive performance.The constructed model has certain practical value.