Our new paper “Banking debt collection aided with a Markov Decision Process optimization engine”, authored by Leonidas Lymberopoulos and Manos Margaritis, has been accepted for presentation at the 27th Annual Conference on Operational Research (EURO 2015), 12 July to 15 July 2015, Glasgow, http://www.euro2015.org
The paper introduces an experimental debt collection optimisation engine, whose goal is to find which collection actions should be applied at each collection stage in order to optimize the revenue to the financial institution. To solve this optimisation problem, we model the collection system as a Markov Decision Process (MDP) and verify that the time homogeneity Markov assumption is met and that our devised MDP is first order. The parameters of the MDP (transition probabilities and rewards) are calculated from a statistical analysis of real customer transactions, which were provided by the debt collection department of a private bank. Simulation results using the R toolkit indicate that when the collection actions are those decided by our optimisation engine, the revenue of the collection procedure is estimated to be increased by ~14% within a four months forecasting period, compared to the actual revenue that the bank gained in the same time period.
More information will be provided after the EURO 2015 conference.