Guides

Reference pages on algorithmic pricing and competition law: how collusion screens work, what the reinforcement learning evidence establishes, how US and EU law treat parallel pricing, and how to audit a pricing algorithm.

Can Pricing Algorithms Collude? What the Evidence Actually Establishes

Three different things get called algorithmic collusion: an algorithm used to carry out a human cartel, a hub-and-spoke arrangement running through a shared vendor or data pool, and supracompetitive prices learned autonomously by self-interested agents that never communicate. The legal exposure and the evidentiary burd

Do Pricing Algorithms Learn to Collude? What the Q-Learning Experiments Actually Show

In the most cited experiment in this literature, pairs of Q-learning agents in a simulated logit-demand duopoly converged on prices capturing 85 percent of the gap between static Bertrand-Nash profit and monopoly profit, and they punished forced deviations before drifting back up. Convergence took roughly 850,000 perio

Collusion Screens Explained: How Economists Detect Cartels in Price and Bid Data

A collusion screen is a statistical test that flags markets whose price or bid patterns fit collusion better than competition. This guide sets out the five main screen families, variance and dispersion, structural breaks, asymmetric price transmission, parallelism, and procurement bid rigging screens, with the publishe

Algorithmic Pricing and the Agreement Requirement: US and EU Law Compared

Sherman Act section 1 and Article 101 TFEU both require an agreement, and conscious parallelism on its own is generally lawful under each. This piece sets out, from the statutes and judgments themselves, what turns algorithmic parallel pricing into an agreement or a concerted practice, why EU information exchange rules

How to Audit Your Pricing Algorithm for Antitrust Risk

A working audit procedure for pricing, revenue management and compliance teams. It covers the records to freeze before anyone asks for them, the inputs that create the sharpest exposure under Section 1 and Article 101, the tests to run against a live model, and why the choice of who runs the audit decides whether the f