UC Santa Cruz

This dissertation has three chapters, with emphasis on the efficiency of online lending using multiple identification strategies. The first chapter presents a theoretical model (1) to help illustrate the behavior of lenders and borrowers in a P2P market and (2) to derive a reduced form model for empirical analysis. In the model, borrowers decide whether or not to default. The benefit of default is that a borrower gets to increase consumption if he chooses not to repay. On the other hand, I assume that default imposes a direct utility loss on the borrower. This loss is assumed to be random and unobservable by lenders or anyone other than the borrower. Ex-ante, when lenders choose whether or not fund a loan, they compare the expected return of the loan vs. their opportunity costs. The opportunity cost is also assumed to be random and unobservable by others. Also, the expected return of the loan depends on lender’s subjective expectations of default risk. The subjective expectations are formulated based on the observed borrower characteristics and loan information. I present methods of testing whether the lenders’ subjective expectations of default is the same as the objective ones.

In the second chapter, using a unique dataset of peer-to-peer (P2P) lending with detailed loan and borrower information, I study which borrower characteristics lenders value when choosing loans to fund, and whether lenders value characteristics that minimize the probability of default. In this online context, the researcher observes everything that the lender does, enabling unbiased estimation of borrower characteristics that lenders favor. However, estimating characteristics that predict loan default is problematic due to selection at the funding stage. I implement three strategies to address this issue: (1) restricting attention to borrower characteristics for which there is no evidence of selection in the first stage; (2) exploiting variation in the probability of funding caused by contemporaneous competition on the platform; and (3) bounding the default estimates in the style of Lee (2009). The results imply that P2P lenders consider employment borrower characteristics as most important when making funding decisions, but disregard several characteristics that are valuable predictors of default risk. Specifically, lenders overestimate the importance of verified employment information, and underestimate the importance of verified education level and marital status.

The third chapter estimates the effect of credit insurance in the peer-to-peer (P2P) lending market. Online lending is a fast growing area of finance. However, it is often plagued by asymmetric information problems and high credit risk, especially in developing countries that do not have well-established financial and credit rating systems. To address this issue, some Chinese P2P marketplaces have incorporated loan insurance into their online platforms. We estimate the treatment effects of credit insurance on the P2P market by exploiting a unique quasi-experiment. Specifically, loan guarantees gradually became available on a top Chinese P2P lending platform to borrowers from 31 major cities between 2012 and 2014 through 12 waves of business expansion. Our empirical results suggest that the availability of credit insurance resulted in significant, strong, and persistent treatment effects on the market demand and supply. The adoption of credit insurance was associated with dramatic increases in the number of loan listings, funding probability per loan, and bidding amount per lender. The average funding time per loan decreased by about 170 hours.