Antitrust & Competition Policy Blog

Editor: D. Daniel Sokol
University of Florida
Levin College of Law

Monday, December 3, 2018

Assessing Autonomous Algorithmic Collusion: Q-Learning Under Sequential Pricing

Timo Klein (University of Amsterdam) is Assessing Autonomous Algorithmic Collusion: Q-Learning Under Sequential Pricing.

ABSTRACT: A novel debate within competition policy and regulation circles is whether autonomous machine learning algorithms are able to tacitly collude on prices. Using a general framework, we show how autonomous Q-learning -- a simple but well-established machine learning algorithm -- is able to achieve supracompetitive profits in a stylized oligopoly environment with sequential price competition. This occurs without any communication or explicit instructions to collude, suggesting tacit collusion. The intuition is that the algorithm is able to learn and exploit the dynamics of Edgeworth price cycles, where periodic price increases reset a gradual downward spiral of price competition. The general framework used can guide future research into the capacity of various algorithms to collude in environments that are less stylized or more case-specific.

| Permalink


Post a comment