Working Papers


The Winner's Curse in an Online Lending Market [link]

  • 2017 FMA
  • 2017 Toronto Fintech Conference
  • 2018 AFA Poster Session
  • 2018 P2P Financial Systems International Workshop
  • 2018 FMA Doctoral Consortium & Job Market Session

Abstract: Using data from Lending Club and Prosper, the two largest peer-to-peer lenders in the U.S., we provide evidence of the winner's curse in the online personal lending market. Borrowers who were rejected by a competitor are 2 times more likely to default than borrowers who were not rejected by a competitor, conditional on receiving the same contract. Borrowers with canceled loan applications at one lender or more likely to default on loans provided by the other lender, compared to borrowers who did not have canceled applications. Borrowers are also more likely to default when offered higher interest rates by a competitor.


Payment Size and Default: Evidence from Small Changes in the National Mortgage Rate [with Dimuthu Ratnadiwakara and Kevin Roshak]

  • 2017 IBEFA Summer Meeting
  • 2017 FMA (semifinalist for best paper)*
  • 2018 Boulder Summer Conference on Consumer Financial Decision Making*

Abstract: How many mortgage defaults could be prevented by modest reductions in monthly payments? We use monthly fluctuations in the national mortgage rate at loan origination to study the effect of small changes in mortgage payments on default for home purchases made in the same year, in the same area, and which eventually reach similar levels of negative equity. We find that a 50 basis point increase in interest rates causes the 12-month default rate to increase by between 65-85 basis points. The effect is large relative average default rates of 3.78% (5.39%) for homes with 10% (30%) negative equity. The magnitude of the effect is relatively constant across different levels of negative equity, which is consistent with liquidity constraints. Balancing tests show that the instrument is not correlated with poor credit quality. Our results shed light on a hangover effect of tight monetary policy prior to the housing crisis.


Modeling Default for Peer-to-Peer Loans [link, featured by BadCredit.org]

Abstract: I use a discrete-time hazard model to analyze default for peer-to-peer (P2P) loans. My data set is large, publicly available, and includes both extensive credit information and soft information. This combination of features, which is unique to P2P data sets, allows for a more thorough analysis of consumer credit than is possible with data from traditional intermediaries. FICO score, borrower-initiated credit inquiries, income, and loan purpose are the most significant variables for explaining default. Two variables extracted from loan descriptions filled out by borrowers are also significant. Several variables typically thought to be significant predictors of default, including income verification and past bankruptcies, are not significant. My model substantially outperforms Lending Club subgrades for forecasting default.


*presentation by coauthor


Works in Progress

Loan Stacking in Online Lending

Abstract: Loan stacking is a phenomenon wherein borrowers receive multiple loans from different lenders at approximately the same time, without the lenders being aware of each other's loans. We find that stacked loans perform much worse than other loans. They are also larger than other loans, which suggests that borrowers stack loans when the credit available from a single lender is not large enough to meet their needs. We find no evidence that stacked loans are fraudulent--there are zero borrowers in our sample who stack loans and then never make a payment.