Semiparametric estimation of default probability: Evidence from the Prosper online credit market

- We examine the effects of a person's past financial characteristics on his likelihood to default.
- Borrowers with higher credit score rankings usually have lower probability to default.
- A borrower with score ranking B is less likely to default than a ranking A borrower.
- The semiparametric estimator outperforms the Probit estimator.
- A model specification test rejects the null hypothesis of the Probit specification at 5% level.

This paper examines the effects of a person's past financial characteristics on his likelihood to default in ex-post loan performance using both Probit and a semiparametric single-index estimator proposed by Klein and Spady (1993). The data used in the paper are a sample of individual loans generated on Prosper, a large US online lending market. The out of sample predictions and the model specification test suggest a misspecification of the Probit model due to the violation of the normality assumption. Estimation results suggest that a borrower's past financial credit score is a reasonably good indicator of one's loan performance. In general, the higher one's credit score ranking, the lower the probability that one would default. One exceptional finding is that a borrower with score ranking B is less likely to default than a borrower with score ranking A. Such a finding suggests that individuals who are in the middle range of credit grades may be more financially credit-dependent than those with higher rankings. As a result, they are more willing to keep their loans in good standings.