This repository contains the formulation and results of the research project conducted in 2021-2022 with Rappi Company information; this work was done as the first project by Ph.D. candidate Ms. Sherly Alfonso-Sanchez, under the supervision of Dr. Cristián Bravo Roman and Dr. Kristina Sendova, and with colaboration of Dr. Alejandro Correa Bahnsen and MSc. Jesus Solano. The work has been published at EJOR . Please cite this work as follows:
Alfonso-Sánchez, S., Solano, J., Correa-Bahnsen, A., Sendova, K. P., & Bravo, C. (2024). Optimizing credit limit adjustments under adversarial goals using reinforcement learning. European Journal of Operational Research, 315(2), 802-817.
In this research, we aim to find and automatize an optimal policy to credit limit adjustment, a fundamental problem for banking and fintech companies that offer credit card products, using Reinforcement learning techniques. This problem requires the optimization of the expected profit given the performed actions for the portfolio of credit card customers; with this purpose, we have to balance two different objectives, first, maximize the portfolio's renewal and minimize the portfolio's provisions.
To apply reinforcement learning techniques, we use an offline learning strategy that needs the simulation of the impact of the action based on historical data provided by a Supper-App company in Latin America. Our work gives insights into the impact of alternative data used to make decisions about credit adjustment limits in credit card portfolios, as well as a new objective technique to make these decisions that are primarily based on expert-driven systems instead of data-driven methods.