Algorithmic Cartels: A New Frontier for Antitrust Enforcement

The antitrust suits against Amazon and RealPage have shed light on a concerning phenomenon: pricing algorithms that facilitate cartel-like behavior without human agreement or even explicit collusion. The concept of a “ghost cartel” – where algorithms independently arrive at the same, anticompetitive outcome – is both fascinating and frightening.

The case against Amazon’s Project Nessie highlights the more deliberate version of this problem, where companies design systems to anticipate rivals’ moves and adjust prices accordingly. This strategy may seem like a clever way to outmaneuver competitors, but it can lead to higher prices sustained over time, without the overt conduct that antitrust law typically targets.

A study on German gas stations in 2017 found that when automated pricing software was adopted by multiple companies, market-level margins increased by about 38%. This suggests that even when algorithms are not explicitly designed to collude, they can still learn to set prices against each other, leading to a cartel-like outcome.

The Subtlety of Algorithmic Cartels

The difficulty in detecting and addressing algorithmic cartels lies in their subtlety. Unlike traditional cartels, where companies explicitly agree on prices or strategies, algorithmic cartels arise from the independent actions of multiple systems. These algorithms may not even be designed to collude; they simply optimize margin through repeated encounters.

This is the “ghost” cartel – a phenomenon that can occur without anyone designing it, and with no data exchange between competitors. It’s a sign that competition has quietly stopped because algorithms have learned that leaving each other alone pays better than fighting.

The Three Faces of Algorithmic Cartels

Algorithmic cartels can manifest in different ways: independently deployed algorithms learn to stop undercutting one another, leading to a cartel-like outcome without human agreement or data exchange. Companies use software to anticipate how rivals will react and adjust prices accordingly, creating a unilateral strategy rather than an explicit pact. Competitors also feed their data into a common provider whose algorithm guides them all, making it easier for regulators to challenge.

The Challenge for Antitrust Enforcement

Antitrust enforcement was designed around human agreement and evidence of meetings or understanding between competitors. However, coordination that arises from machine learning provides none of this, making it harder for regulators to detect and address algorithmic cartels.

The RealPage case highlights the challenges in addressing this issue. Despite the Department of Justice’s suit, the company paid no penalty and admitted no wrongdoing, with the settlement mainly restricting data usage and installing a court-appointed monitor. This outcome raises questions about the effectiveness of antitrust enforcement in the face of algorithmic cartels.

What Does it Mean for Companies?

Boards and executives should take note: even if algorithms are not explicitly designed to collude, they can still lead to cartel-like behavior. The key is subtle – an algorithm that sets an obviously wrong price may be easy to catch, but one that optimizes margin through repeated encounters can be much harder to detect.

As researchers continue to study and debate the implications of algorithmic cartels, companies would do well to watch their behavior closely. If left unchecked, these algorithms could lead to higher prices sustained over time, undermining competition and innovation in markets.

The phenomenon of algorithmic cartels represents a new frontier for antitrust enforcement. As we navigate this uncharted territory, it’s essential that regulators, companies, and researchers work together to understand and address the subtleties of these complex systems. The future of competition depends on it.