Algorithmic and Human Collusion

As self-learning pricing algorithms become popular, there are growing concerns among academics and regulators that could learn to collude tacitly on non-competitive prices thereby harm competition. I study popular reinforcement learning show they develop collusive behavior in a simulated market environment. To derive counterfactual resembles traditional tacit collusion, conduct experiments with human participants the same Across different treatments, vary size number of firms use self-learned algorithm. provide evidence oligopoly markets can more if make decisions instead humans. In two-firm markets, weakly increasing market. three-firm weaken competition most an algorithm sellers inexperienced.