Algorithmic Pricing and Liquidity in Securities Markets

Abstract

We let ``Algorithmic Market Makers'' (AMs), using Q-learning algorithms, determine prices for a risky asset in a standard market making game with adverse selection and compare these prices to the Nash equilibrium of the game. We observe that AMs effectively adapt to adverse selection, adjusting prices post-trade as anticipated. However, AMs charge a markup over the competitive price and this markup increases when adverse selection costs decrease, in contrast to the predictions of the Nash equilibrium. We attribute this unexpected pattern to the diminished learning capacity of AMs when faced with increased profit variance.

Keywords: Algorithmic pricing, Market Making, Adverse Selection, Market Power, Reinforcement learning

JEL Classification: D43,G10,G14