Dealer markets: A reinforcement learning mean field game approach

We study the problem of finding an equilibrium strategy in an Over The Counter (OTC) market populated by multiple strategic dealers quoting the bid–ask prices. The need for an equilibrium strategy comes from the assumption that each dealer adapts their behavior by learning the optimal quoting strategy. Hence, we model the market as a game between many agents competing for resources. Based on this framework, we propose an efficient numerical procedure using Reinforcement Learning and Mean Field Games which learns an approximate equilibrium. Through an experimental campaign, we validate the proposed method in a realistic market-making scenario against strategic dealers trained to exploit our weaknesses and evaluate their performance against non-strategic agents.