On November 6, 2019, the FCO and the French Competition Authority (“ADLC”) presented a joint study on “Algorithms and Competition.” The study focuses on algorithms used for dynamic price setting and their potential effects on competition, particularly in the form of collusion, and contains important insights for companies utilizing third- party algorithms.
Differentiating Between Algorithms
Algorithms utilized today in private industry are extremely varied, differing in terms of their purpose, data sources, methods of operation, and source. From a competition law perspective, algorithms that involve prices are particularly relevant, whether they pertain solely to the collection of price inputs or are directly used to set prices. For this reason, the study focuses in large part on algorithms used for dynamic pricing: adjusting prices, potentially in real-time, in response to changes in input costs, supply and demand, and competitor pricing. Moreover, for competition purposes it is particular relevant whether the algorithm was developed internally or was provided by third-parties, which might sell the same or similar algorithms (also) to competitors. The latter may create avenues for coordination among competitors, perhaps unwittingly.
Theoretical Impact Of Algorithms Remains Unclear
The study discusses the theory of how dynamic pricing algorithms could impact, or even inadvertently cause, horizontal collusion. The authors, however, acknowledge that the effects on competition are ambiguous and will depend on market conditions.
Inter alia, the study analyzes whether a competition law violation could “organically” arise from the use of machine-learning algorithms unilaterally designed and implemented by each competitor in the absence of any human agreement. Apart from questions about the technical feasibility of this scenario, the study concludes that any form of convergence that arises in this way would likely be categorized as permissible parallel behavior.
Facilitating Traditional Anticompetitive Practices
The study further explores two different scenarios in which algorithms may be used for anticompetitive purposes. The first such scenario involves the use of algorithms to support or ease the implementation of anticompetitive practices that have already been established. Two illustrative examples are provided from regulatory decisions issued in the United Kingdom. In the first, two companies selling posters used an algorithm from third-party software to implement their agreement to not undercut each other on price on an online marketplace. In the second, two energy suppliers that had agreed not to recruit the other’s customers used an algorithm to share customer details and avoid actively targeting each other’s customers.
As these examples make clear, algorithms could greatly facilitate the enforcement of anticompetitive agreements by providing a means of monitoring prices or competitor activities, as well as automatically correcting prices or “punishing” deviations from the anticompetitive agreement. These applications are equally applicable to both horizontal collusion and vertical agreements. While the algorithms are not necessary in order to establish that an anticompetitive agreement exists, they are key to understanding the scope of an agreement’s negative effects—and thus relevant for fine calculations.
Third-Party Algorithms As A “Hub”
The second scenario involves a situation in which a third party provides the same algorithm, or algorithms that are somehow coordinated or connected, to multiple competitors that themselves have no direct (human) communication or contact. Such algorithms could, for example, function on the basis of common principles, such as the formula for setting prices, or share data in a way that allows competitors to track each other’s pricing and sales. The study notes that even a well-intentioned third party attempting to calculate prices for each individual competitor could potentially draw on confidential data from multiple competitors, thereby introducing convergence on the market.
The above described behavior does not pose novel legal issues, but likely falls under established precedents regarding “hub-and-spoke” arrangements and third-party cartel facilitators. Liability for competitors would not arise unless at least two of the competitors were aware of or could have reasonably foreseen the third party’s anticompetitive acts. Nonetheless, the study suggests that companies need to exercise great caution when relying on third-party algorithms that may be used by competitors or have been developed (or “trained”) with competitor data, particularly in the sensitive area of price or price inputs.
The study makes clear that the FCO and ADLC are highly alert to the fact that algorithms have the potential to make collusive arrangements more effective and thus increase their harm to competition. Aside from theoretical considerations of “self-colluding” algorithms, the study suggests that existing legal concepts are well-adapted to deal with the anticompetitive use of algorithms. Companies should thus apply particular scrutiny before utilizing data sets or third-party algorithms also used by competitors or developed with their input.
 The study references a number of cases in which the European Commission (“Commission”) and the CMA competition authorities have identified the use of monitoring algorithms in vertical agreements, see Philips (Case AT.40181), Commission decision of July 24, 2018, available in English here; Pioneer (Case AT.40182), Commission decision of July 24, 2018, available in English here; Asus (Case AT.40465), Commission decision of July 24, 2018, available in English here; Denon & Marantz (Case AT.40469), Commission decision of July 24, 2018, available in English here; and Digital piano and digital keyboard sector (Case 50565-2), CMA decision of August 1, 2019, paras. 3.97 et seq., available in English here.