Algorithmic Pricing and Liquidity in Securities Markets

Abstract We study “Algorithmic Market Makers” (AMs) that use Q-learning algorithms to set prices for a risky asset. We find that while AMs successfully adapt to adverse selection, they struggle to learn competitive pricing strategies. This failure is driven by limited experimentation and noisy feedback regarding the profitability of undercutting a competitor. Consequently, an increase in AMs’ profit volatility tends to result in less competitive market outcomes. These features leave identifiable patterns: for example, AMs earn higher rents in the absence of adverse selection, and their bid-ask spreads respond asymmetrically to symmetric shocks to their costs.