Invisible Primes: Fintech Lending with Alternative Data

Abstract

We study the impact of alternative data on credit access and borrower outcomes using anonymized data from a major fintech platform that uses alternative data and artificial intelligence in its underwriting models. Comparing actual outcomes of the fintech platform's model to counterfactual outcomes based on a "traditional model" used for regulatory reporting purposes, we find that the fintech platform's model approves 15-30% of low credit score applicants rejected by the traditional model and offers substantial reductions in interest rates. The borrowers most positively affected are the "invisible primes"—borrowers with low credit scores and short credit histories, but also a low propensity to default. About two-thirds of the effects are due to the inclusion of additional data, while the remainder is due to a more sophisticated underwriting model. Leveraging exogenous variations in credit access, we show that funding loans to invisible primes leads to better economic outcomes for borrowers.

Keywords: Fintech Lending, Alternative Data, Machine Learning, Algorithm Bias

JEL Classification: D14, H52, H81, J24, I23