Michael Mark Finalist of the 2020 INFORMS Student Paper Competition
Michael Mark © EPFL
The paper “A Reinforcement-Learning Approach to Credit Collections,” co-authored by Michael Mark, a PhD Student at the Chair of Operations, Economics and Strategy, Prof. Naveed Chehrazi (Olin Business School, Washington University), and Prof. Thomas Weber, has been selected as a finalist of this year’s INFORMS Student Paper Competition in the Section on Finance. Michael is scheduled to present the paper at the upcoming INFORMS Annual Meeting, on November 9.
Abstract:
This paper develops a dynamic reinforcement-learning agent capable of finding high-quality policies for the practice of debt collections. At its core, the agent effectively learns how to control a stochastic self-exciting point process in order to maximize an asynchronously obtained reward. Because we use the general formulation of the problem as an agent-environment interaction our results are readily extensible beyond the presented application. Furthermore, with the growing need for interpretable machine-learning models we provide a policy regularization technique which makes learned policies intuitively understandable for human decision makers. Finally, our implementation features a novel “guided exploration" mechanism which improves the agent's performance in hard-to-reach states.