Algorithms and Collusive Agreements. When computer algorithms rather than humans fix the price of goods or services, do the antitrust laws apply? This question is currently vexing antitrust agencies across the globe. Naturally, agencies say yes. However, it is difficult to see legally how algorithmic price fixing without human intervention could facilitate culpable collusion between competitors. The answer, we believe, is not a definitive yes, but rather is determined by the nature of the algorithm and the circumstances in which it is employed. It is a trite but nonetheless relevant observation that if the algorithm facilitates:
(a) explicit collusion, in which an agreement between competitors can be identified, then the antitrust laws will apply;
(b) tacit collusion or conscious parallelism, in which competitors coordinating their behavior without explicit agreement, the antitrust laws would not apply; and
(c) intermediate collusion, in other words collusion that is less than explicit collusion but more than tacit collusion, the antitrust laws may apply. In these circumstances, an antitrust infringement is found where there is evidence of tacit collusion accompanied by other plus factors, such as correspondence or information exchanges between competitors exposing collusive intent. In the European Union, this practice is known as a “concerted practice.”
Broadly, there are three forms of algorithms that are relevant.
The Tactic Collusion Algorithm. This algorithm replaces human pricing decisions with a computer algorithm. It dynamically changes prices according to market demand and without the need for firms to engage in communications with their rivals, thereby replicating tacit collusion. Under current antitrust laws, this type of algorithm is not inherently unlawful and is in fact ubiquitous in the hotel and transportation sectors; it adjusts prices according to demand to minimize the likelihood of empty rooms/seats. However, antitrust violations could arise where firms share their respective tacit collusion algorithms with one another or agree to use a single algorithm. At worst, this would result in unlawful explicit collusion; at best unlawful intermediate collusion. An example of the former is the 2015 U.S. Department of Justice (DOJ) matter of David Topkins. Topkins, a seller of posters, agreed with other firms engaged in the sale of posters to fix prices of posters sold through Amazon Marketplace in the United States. To implement the agreement, Topkins and his co-conspirators agreed to adopt a specific pricing algorithm for the sale of the posters with the goal of coordinating changes to their respective prices based on market conditions. Of course, it was the existence of the explicit agreement between Topkins and his rivals to jointly implement the tacit collusion algorithm that allowed the DOJ secure a conviction.
Less clear is the situation where rival firms employ the same IT vendor to create a tacit collusion algorithm and the IT vendor uses the same algorithms for everyone. In these circumstances, absent an explicit agreement, the only possible current antitrust violation would be intermediate collusion, and the agency would have to identify relevant plus factors revealing an intention to collude.
The Monitoring Algorithm. This algorithm is created by firms in order to monitor rivals’ market behavior in furtherance to an existing collusive agreement. The monitoring algorithm dispenses with the need for regular meetings, communications and reporting between competitors to ensure compliance with the collusive agreement. Not only may it automatically gather information regarding pricing and other agreed to market terms but also search for deviations and, at its most sophisticated, retaliate against cheats. Of course, explicit collusion such as this is an unambiguous violation of the antitrust rules and the monitoring algorithm is merely a tool created to efficiently invigilate the unlawful agreement. This behavior is clearly caught by existing antitrust laws.
The Machine Learning Algorithm. The machine learning algorithm employs big data, including rivals’ data and decisions, to not only instantly react to price and/or market adjustments, but also to constantly learn and improve. Without human intervention, this algorithm is capable of computing an infinite number of scenarios with the singular aim of maximizing the profits of the firm employing it. This may include learning that interdependency with competitors means that joint profit maximization is in fact the optimal outcome i.e. tacit collusion. It is hard to see this algorithmic collusion falling within the ambit of current antitrust laws.
We anticipate antitrust interest in this area to increase exponentially across all major jurisdictions. Antitrust enforcement against firms that use algorithms to implement their collusive agreement is relatively straightforward. However, it is doubtful that without amending the current antitrust rules, algorithms that facilitate tacit collusion could be prosecuted. If, as is expected, the frequency of tacit collusive outcomes increases because of the ease with which algorithms facilitate collusion without the need for contact with rivals, governments and antitrust agencies are likely to become more willing to adopt expansive, controversial interpretations of the antitrust rules to bring clearly within their ambit collusive algorithms. They may also seek to amend their antitrust laws to achieve the same aim.
We are already seeing a growing propensity of the antitrust agencies to examine these issues. On June 19, 2018, the French and German antitrust agencies launched a joint project to analyse the above antitrust issues raised by algorithms and identify approaches to address them. At the end of the project, the authorities intend to publish a joint working paper. On July 24, 2018, the European Commission fined, in four separate decisions, consumer electronics manufacturers Asus, Denon & Marantz, Philips, and Pioneer for imposing fixed or minimum resale prices on their online retailers in breach of EU competition rules. The fines were in excess of €111 million. Of interest, each manufacturer appears – the decision has yet to be published – to have deployed algorithmic monitoring tools that not only allowed them to track resale price setting in the distribution network but also to intervene rapidly in case of price decreases. The Commission did not in these cases target the algorithms or their programmers but rather the manufacturers that employed them. Additionally, each manufacturer individually deployed its own algorithm and there was no suggestion that the algorithms facilitated collusion. However, these cases exemplify the growing trend of antitrust agencies to examine the employment of these pricing algorithms and their readiness to challenge companies who use them.