| Online Peer-to-Peer(P2P)lending has been developed rapidly around the world due to its high yield,low cost,and high speed,since its birth in the UK in 2005.In 2014,the P2 P platform Lending Club founded in the USA in 2006 was listed successfully and became the first going public company in the industry.In 2007,the first P2 P platform Paipaidai was established in China and the online P2 P lending in China has had a rapid development.However,behind the illusion of prosperity of P2 P online lending,there are a series of problems,including credit risk,technical risk,and compliance risk.Among them,credit risk is the most important risk and challenge faced by P2 P lending platforms.Borrowers’ frequent default behaviors have caused serious economic loss to platforms and investors,which has a negative impact on the normal operation of platforms,even leads to the bankruptcy declaration of P2 P lending businesses.Therefore,it is of great significance to build an accurate P2 P credit scoring model for the healthy development of the P2 P lending industry.Current researches on P2 P lending focus on the default behaviors.Various studies have applied different techniques to build credit scoring models to predict the possibility of defaulting.These models can help personal investors judge borrowers’ behaviors,manage risk,make correct decisions and reduce investment loss.However,prepayment behaviors,which also breach the contract,can also affect the profitability of a loan,increase the risk of reinvestment.It is necessary to consider the prepayment behaviors and help investors better understand the risk of P2 P lending.In this paper,we consider the default behavior and the prepayment behavior,the two main causes reducing the return of a loan,at the same time.Building the rules extraction and integration model based multi-class classification for P2 P risk assessment predicts borrower’s actual behavior and finds loans with high default or prepayment rate.Besides,very simple rules are extracted based on single features and selected based on the macro-F,so that we can get the meaningful and easy-tounderstand rules,make a deep understanding of characteristics of different borrowers,and help platforms and investors predict repayment behaviors more accurately.In this work,the empirical study was conducted on two datasets collected from domestic platform-Paipaidai and foreign platform-Lending Club.The experiments showed that the proposed approach is able to outperform 9 other common credit scoring models in terms of assessment performance and interpretability.In addition,the differences between extracted rules of the two platforms are analyzed and the differences of P2 P platforms in China and the USA are summarized in terms of collecting borrowers’ information.According to these differences,we give suggestions about how to control P2 P credit risk and provide a reference for investors’ return and the healthy development of the industry. |
