消费行为在个人信用风险识别中的信息含量研究

Informational Content of Consumption Behavior in Consumer Credit Risk Evaluation

本文使用某现金贷平台的借款人数据,研究消费行为信息在个人信用风险评估中的信息含量。研究发现,对于现金贷借款人这类信用记录不足的群体,传统征信信息往往不足以识别借款人的信用风险,而通过大数据技术引入更加高频的消费行为信息能够有效补充额外的信息含量,提高对信用信息薄弱人群的风险识别效率。传统征信信息和消费行为信息提供的信息含量互不相同,不能相互替代。因此金融科技创新对传统征信系统以外信息的利用有助于降低消费信贷市场的信息不对称。本文创新性地利用真实的微观信贷数据对不同信息的信息含量进行了比较,对消费金融的理论和实践都具有重要价值。

With the recent development of Fintech,massive amounts of financial market information can be processed at much higher speed than ever before.New information sources and advanced information processing methods have brought salient changes to the finance industry,especially the consumer credit market,in which increasingly diverse information is used to evaluate consumer credit risk.In this paper,we study whether borrowers consumption behavior predicts personal credit risk.By using individual-level data from a Chinese cash loan platform,we innovatively explore the informational content of consumption behavior in consumer credit risk evaluation,which is important to both the theory and practice of consumer finance.Consumers borrowing history and past repayment performance is always at the core of mainstream credit scoring systems,represented by FICO scores,because historical performance reveals the borrower s repayment capability and willingness,which strongly predict credit risk(Barron,2003).However,the effectiveness of such credit scoring systems varies among groups.For consumers with limited borrowing history,credit risk cannot be assessed based on a traditional credit scoring system.There is growing literature showing severe information asymmetry in the high-cost lending market(Karlan&Zinman,2009;Dobbie&Skiba,2013).Recent studies on online lending products have shown that various information outside of the traditional credit scoring system is highly correlated with borrower credit risk(Berg et al.,2019;Duarte et al.,2012;Li et al.,2014;Liao et al.,2015).In this paper,we focus on borrowers consumption behavior.Different from past borrowing and repayment records,consumption data are of higher frequency and easier to collect in the mobile Internet era.However,the relation between consumption behavior and repayment performance is not straightforward.Therefore,we add to the literature by exploring the informational content provided by consumer behavior information in credit risk evaluation.We obtain loan-level data from a large Chinese cash loan platform,and we observe loan performance to measure borrower credit risk.Cash loans are a high-cost,short-term online lending product of small size.Cash loan borrowers are usually subprime borrowers subject to tight financial constraints,and they tend to apply for multiple consumer credit products,such as credit cards from banks and Ant Credit Pay(ACP)from Ant Financial,one of the largest Fintech companies in China.It is noteworthy that credit cards and ACP are almost identical in providing consumer credit services but differ in the information collected to evaluate credit risk and process applications.Banks use traditional credit scoring methods and collect information regarding applicants employment and income,which directly prove repayment capability;whereas ACP mainly uses consumption data from the e-commerce platform Alibaba.The cash loan platform obtains borrowers permission to check their personal shopping and payment accounts,which makes it possible to observe credit cards and ACP held.We use credit cards and ACP to measure borrower predicted credit risk based on traditional credit scoring information and consumer behavior information,respectively.We use a representative sample of 3,814 loans on the platform from May 2015 to April 2017 for a regression analysis comparing the predictive power of different information-based credit risk evaluation systems,and we find that both credit cards issued by large banks and ACP can significantly predict the default of cash loan borrowers.When credit card holding is controlled,the holding and credit limit of APC is still significantly negatively related with the probability of default,which means that consumption behavior provides additional power to traditional credit scoring for predicting consumer credit risk.We conclude that consumption data provide different information from the data used by traditional credit scoring methods.Our study sheds light on the importance of utilizing new information beyond traditional credit scoring systems to reduce information asymmetry in the consumer credit market.For consumers with insufficient credit records,such as cash loan borrowers,incorporating consumption behavior data provides additional information and improves risk evaluation results,which benefits consumers who struggle to obtain low-cost financing services and makes it possible for them to establish personal credit.