Applying reinforcement learning to estimating apartment reference rents

Abstract

In apartment revenue management, rental rates for new and renewal leases are often optimized around a reference rent, which is defined as the “economic value” of an apartment unit. In practice, reference rents are usually estimated using some rules-based approaches. These rules are mostly intuitive to understand and easy to implement, but they suffer from the problems of being subjective, static, and lacking self-learning capability. In this study, we propose a reinforcement learning (RL) approach to estimating reference rents. Our intent is to find the optimal reference rent estimates via maximizing the average of RevPAUs over an infinite time horizon, where RevPAU (Revenue per Available Unit) is one of leading indicators that many apartments adapt. The proposed RL model is trained and tested against real-world datasets of reference rents that are estimated with the use of one rules-based approach by two leading apartment management companies. Empirical results show that this RL-based approach outperforms the rules-based approach with a 19% increase in RevPAU on average.