Optimal Credit Scores Under Adverse Selection
33 Pages Posted: 20 May 2022
Date Written: May 18, 2022
Abstract
The increasing availability of data in credit markets may appear to make adverse selection concerns less relevant. However, when there is adverse selection, more information does not necessarily increase welfare. We provide tools for making better use of the data that is collected from potential borrowers, formulating and solving the optimal disclosure problem of an intermediary with commitment that seeks to maximize the probability of successful transactions, weighted by the size of the gains of these transactions. We show that any optimal disclosure policy needs to satisfy some simple conditions in terms of local sufficient statistics. These conditions relate prices to the price elasticities of the expected value of the loans for the investors. Empirically, we apply our method to the data from the Townsend Thai Project, which is a long panel dataset with rich information on credit histories, balance sheets, and income statements, to evaluate whether it can help develop the particularly thin formal rural credit markets in Thailand, finding economically meaningful gains from adopting limited information disclosure policies.
Keywords: information design, applied machine learning, credit scores, rural credit markets
JEL Classification: C58, D82, G21, Q14
Suggested Citation: Suggested Citation