Assessing autonomous algorithmic collusion : Q-learning under short-run price commitments

A novel debate within competition policy and regulation circles is whether autonomous machine learning algorithms may learn to collude on prices. We show that when fims face short-run price commitments, independent Q-learning (a simple but well-established self-learning algorithm) learns to profitably coordinate on either a fixed price or on asymmetric price cycles - although convergence to rational and Pareto-optimal collusive behavior is not guaranteed. The general framework used can guide future research into the capacity of more advanced algorithms to collude, also in environments that are less stylized or more case-specific.