A Research On Credit Risk Assessment Of P2P Lending Based On Text Analysis

The combination of the Internet and private lending has formed a new financial model-P2 P lending.Without the need to through the traditional financial institutions such as banks,the borrower and the investor can directly trade on the Internet platform.China's P2 P lending market has continued to expand,the gradual establishment and improvement of industry supervision has led to a stable and long-term development mechanism for P2 P lending.However,the fact that there is still a high percentage of problem platforms has made us realize that credit risk is a key issue that needs to be solved urgently for the sustained and healthy development of the entire industry.The borrowing description,as a unique information supplement method,contains more personal information about the borrower,such as the detailed description of the financial status,the family situation,the reason for borrowing,and the willingness to repay the loan.Although these informations cannot be verified,it will,to a certain extent,affect investors' investment behavior.Then,can the borrowing description reflect the borrower's credit risk? Therefore,this study focus on the role of borrowing description playing in the identification of credit risk.This article empirically studies the relationship between borrowing description and borrower's credit risk using the data from renrendai.com.It extracts description variables from borrower's description by the perspective of text features and text content.According to the characteristic that P2 P lending is repayment by installments,starting from the borrower's repayment record,this article defines clearly the definition of the default behavior,and uses default level to indicate the degree of default.Logit regression is used to obtain a default probability assessment model,and sorting multivariate Logit regression is used to explore the effect of text variables on the degree of default.For default borrowers,this article uses default liquidity and default recovery as a measure of their repayment performance,further explores the explanatory role of text variables for follow-on repayment behaviors,filling the gap in P2 P lending credit risk research.The empirical results show that:(1)The credit risk of the borrower cannot be judged by the length of the text and the punctuation mark.The borrower with a higher difficulty of reading the text has a higher probability of default,and the borrower with a higher positive degree of the text has a lower recovery rate after a default.(2)Different borrowing purposes will have different impacts on the probability of default.The probability of default on the use of “business turnover” will be higher and the degree of default will be greater.The probability of liquidation and the recovery rate of defaulted loans will be lower on the purpose of “asset consumption”.(3)The supplement of work information and credit information can significantly reduce the probability of default and the level of default,but the supplementation of income information may be deceptive.(4)Borrowers with guaranteed expression have a higher probability of default and higher level of default,more likely to pass an air commitment.Demand-requested borrowers have strong demand for funds,so their probability of default is higher and probability of liquidation after default is lower.In a word,the description of borrowing has a greater role in identifying the default probability and the degree of default,but it has limited effect on the repayment behavior after default.This article puts forward a new perspective for the research on credit risk identification of P2 P lending,hoping to help investors and the platform further improve risk management capabilities.