Learning How to Borrow in a FinTech World

Abstract

We examine more than 750,000 loan applications made during 2019-2024 by about 200,000 unique individuals on a large fintech lending platform in Finland. We complement this dataset of loan applications, loan offers, and loans disbursed with credit registry information regarding the credit score of applicants as well as defaults post-loan application.

On the platform there are typically 15-20 financial institutions which can offer loans to applicants within minutes, using proprietary algorithms to determine if an offer should be made, and if so, what loan amount, interest rate and maturity to offer to an individual. In this setting, loan applicants are able to learn with minimal cost information about the supply of credit available to them. Lenders also are able to infer how competitive their loan offers are in each segment of the applicant pool. Although loan applications involve verifiable information, including via lenders' access to the applicant's credit record, this is a setting with significant asymmetric information regarding an individual's type, i.e., their ability to repay the loan. At the same time, this setting allows applicants to learn about the supply curve by seeing offers (or lack thereof) from many lenders at the same time, and by applying multiple times and asking for different terms, with close to zero costs.

Empirically, we document four main results. First, there are significant benefits to individuals from learning about the credit supply curve, and hence high benefits to searching, as there exists high dispersion in terms offered by lenders to the same applicant. Second, fintech innovation (automatically tagging the best loan offer among those received by an applicant) can help people learn in this complex setting, with lower income individuals being most impacted by the information provided by the platform. Third, loan applicants search significantly, by applying multiple times, and asking for loans with different terms, while rejecting a majority of offers. Search intensity and offer rejection probability vary with the applicant's credit score in ways that suggest that individuals understand their own type. Fourth, suggesting that this is a dynamic adverse selection setting, we find the lenders are less likely to offer loans to repeat applicants, which are inferred to be worse types. Default outcomes of all applicants on the platform after the time of applying for loans are consistent with these patterns.