AI-Powered Trading, Algorithmic Collusion, and Price Efficiency

Abstract

The integration of algorithmic trading with reinforcement learning, known as AI-powered trading, has significantly impacted capital markets. This study employs a theoretical laboratory characterized by information asymmetry and imperfect competition, where informed AI speculators serve as the subjects of our simulation experiments. It explores how AI technology impacts market power, information rents, price informativeness, market liquidity, and mispricing. Our findings show that informed AI speculators can autonomously learn to sustain collusive supra-competitive profits without any form of agreement, communication, intention, or any interactions that might violate traditional antitrust regulations. AI collusion robustly emerges from two distinct mechanisms: one through price-trigger strategies ("artificial intelligence") when price efficiency and noise trading risk are both low, and the other through self-confirming bias in learning ("artificial stupidity") under other conditions.

Keywords: Reinforcement learning, AI collusion, Homogenization, Experience-based and self-confirming equilibrium, Asymmetric information, Price informativeness, Market liquidity. (JEL Classification: D43, G10, G14, L13)

JEL Classification: D43, G10, G14, L13.