Improving the Allocation Consumer Credit with Machine Learning

The project will examine the accuracy and fairness of credit scores to improve the allocation of consumer credit and mitigate disparities for disadvantaged groups, such as young and minority borrowers. The research will implement a machine learning approach that is at the frontier of computer science to explore this economic question. The research will investigate the notion that there need not be a trade-off between accuracy of credit scores and a more equitable allocation of loans. The project will further look at factors related to credit conditions and default behavior for disadvantaged groups. The resulting findings will generate direct and actionable implications for policies and regulations pertaining to consumer debt to help reduce inequality and improve welfare.

The project will introduce pioneering advances in constrained machine learning to develop predictions of consumer default that are fair such that the credit allocation does not penalize borrowers in disadvantaged groups, such as young or minority borrowers. The notion of fairness is grounded in economic theory, measurable in the data, and has an intuitive interpretation. Constrained machine learning is at the frontier of computer science, and this project will apply these techniques in examining this economic issue. The project will isolate the most important factors associated with default and how they vary over time for different subpopulations. The results of the project will be useful in designing policies on consumer finance that enables equitable allocation of credit that improves welfare.