The TLDR version:
- A market set Interest rate is better predictor of default over credit score. This is primary because credit scores are backward looking.
- For lower quality borrowers, soft/non standard information is relatively more important (in this case for outcome=Fraction repaid) over credit score (this was very counter intuitive to me)
- Social data (pictures/friend recommendations) did not make a difference either way (not empirically correlated to default). another very interesting result.
- Banks do not look at non standard data especially for smaller loan sizes, not scalable for them
This paper had me thinking on the various implications for a product in this space.
- p2p lending in the smaller loan sizes is a huge opportunity, Banks are not playing in this space.
- Does social even matter?
- Should credit score be a major signal for lower quality borrower in your model? What other soft information can you augment in your model to provide better signal?
We live in interesting times 🙂