Leveraging Payment Data for Greater Gender Equity in Credit Scoring

Context

Gender gaps in access to and usage of digital credit exist around the world. Low-income women disproportionately lack credit histories (sometimes referred to as “having a thin file”), property rights, and formal earnings, exaggerating this gap in financial inclusion. Digital financial services, including digital credit, have the potential to remove barriers impeding women’s access, particularly by leveraging non-traditional data in lieu of formal credit histories. Prior research demonstrated this approach to gender-differentiated credit scoring may increase women’s access to formal credit. However, it is still unknown whether and to what extent women rejected by standard credit scoring models would benefit from credit access, nor how, if at all, the potential benefits differ from those experienced by individuals approved by existing models.

Study Design

Researchers are partnering with RappiCard Mexico, the fintech arm of a leading delivery platform in Latin America, to develop novel gender-based credit scoring algorithms and evaluate the impact of using them to allocate credit. Using detailed data from roughly 300,000 active RappiCard users (40% of which are women), the team will train a machine learning model to jointly optimize on subpopulations of men and women. They will then identify a sample of women who would be rejected by standard models but approved by the new stratified model and evaluate the impact of expanded credit access on their economic and psychological outcomes, as well as profitability for RappiCard. This approach will be compared to a model based on credit bureau records alone and a model that also uses digital footprints, but is not gender differentiated.

The projects’ second phase will pilot a randomized controlled trial (RCT) with 200 women to evaluate the impact of access to credit on individuals who would be rejected with a pooled model but approved with the stratified model. A local partner organization, Que Funciona para el Desarrollo (QFPD), will survey study participants two months after they were awarded Rappi credit cards, measuring psychological and economic outcomes such as risk coping and consumption smoothing, access to other sources of credit, and subjective well-being. Researchers will also test the profitability of the new algorithm by looking at credit card use, and default rates of women approved with the different models.

After completion of the pilot, researchers intend to scale up with a full-scale randomized control trial (RCT) and accompanying endline survey.

Results and Policy Lessons

Results from this project are forthcoming.