Website CV Paper
Fields
Industrial Organization, Applied Micro
Job market paper
Algorithmic Collusion Under Firm-Specific Cost Shocks
The increasing use of algorithms to make pricing decisions has begun to concern antitrust authorities. I analyze the behavior of Q-learning algorithms in a repeated differentiated product Bertrand pricing game where firm-specific marginal cost fluctuates over time. I find that the algorithms autonomously learn to tacitly collude, sustaining high prices through reward-punishment strategies under both privately and publicly observable cost fluctuations. Supra-competitive outcomes remain achievable under incomplete information. The findings are robust to changes in demand and learning parameterization, as well as initial beliefs of the reinforcement learning algorithm.