Screening Peers Softly: Inferring the Quality of Small BorrowersCoauthors
Rajkamal Iyer, MIT; Erzo F.P. Luttmer, Dartmouth University; Kelly Shue, University of Chicago
The recent banking crisis, and a corresponding proliferation of online peer-to-peer lending sites, have generated increased interest in the ways in which creditworthiness is measured. Whereas banks rely almost solely on hard financial information, peer-to-peer lenders incorporate soft, nonstandard variables in their assessment of a borrower’s potential credit risk. In their working paper “Screening Peers Softly: Inferring the Quality of Small Borrowers,” Asim Ijaz Khwaja, Sumitomo-FASID Professor of International Finance and Development, and his coauthors, Rajkamal Iyer, Erzo F.P. Luttmer, and Kelly Shue, assess the accuracy of default prediction in peer-to-peer lending markets and argue that the use of nontraditional variables allows for more-effective screening of borrower quality.
Peer-to-peer lending sites facilitate loans between individual lenders and borrowers and are becoming increasingly popular options for both small and lower-quality borrowers. There are 12 active sites in the United States, and many more throughout Europe and Asia. Prosper.com, the online marketplace that Khwaja and his colleagues used for their study, has more than 1.6 million members and claims more than $500 million in funded loans since its inception, in 2006. Borrowers provide verifiable financial information but can also offer so-called “soft” information they feel might help them, such as pictures and text describing why they need a loan and why they consider themselves a good credit risk.
Borrowers on the Prosper site are assigned to one of seven credit categories, but their exact credit scores are not available to lenders. This allowed researchers to analyze the predictive results of nonstandard screening methods. “The market’s equivalent of assigning a credit score is the final interest rate offered,” says Khwaja. “So if that score is not known, you would assume that the interest rate offered to every borrower within a certain category should not vary at all. But it does, and in a way that moves with the actual credit score; it’s as if the market is able to see the true credit score without actually seeing it.”
He goes on to explain: “You can’t fully predict a credit score. But we found that lenders on Prosper, given the data available to them, were able to extract up to 70% of the information they could possibly extract.” By combining nonstandard information with financial data, the authors say these lenders “are able to predict default with 45% greater accuracy than what is achievable on just the borrower’s credit score.”
Khwaja describes the difference in screening methods between banks and peer-to-peer markets as being on a two-by-two column matrix, with hard and soft information that is both verifiable and nonverifiable. “A bank sits mostly in one box—hard, verifiable information,” he says. “What this market does is open up the three other boxes.” Countering the assumption that nonverifiable data are useless, these markets use soft information—such as the maximum interest rate a borrower states he or she is willing to accept—to determine credibility when assessing a borrower’s risk profile. “If you’re a good risk and you know it,” says Khwaja, “you’re willing to demand a lower interest rate, because you believe you can get funded, whether on this site or elsewhere. That creates what we call the credible signal, and while the borrower maximum rate is a clear example of this, there are also ways of describing why you need the loan that make you more versus less credible to potential lenders.”
Khwaja sees signs of sophistication in both lender and borrower behavior. “Borrowers realize they need to send credible signals, and investors value those signals; the more credible the signal, the more likely lenders respond to it and the more likely it actually predicts creditworthiness. Our sense is that as this type of marketplace matures, these features are going to get more and more salient.”
The broader question asked in this paper is to what extent peer-to-peer markets can complement traditional lenders. After all, in Khwaja’s words, “the bank world is a screening world, not a signaling world.” In reality, though, when we build relationships or create partnerships, we allow for both the screening of hard information and the signaling of soft information. And as the success of these markets demonstrates, effective risk-assessment should involve both.
- by Susannah Ketchum Glass