In the beginning there were flies.
You probably wonder what the connection between algorithms competition law and flies is. Think of a biology textbook costing almost $24 million on Amazon. Originally created in 1992, the book was out of print by 2011. Seventeen copies were for sale on Amazon, and fifteen of them started at $35. However, two other copies from two different sellers had its starting value amounting to $1 million. It’s true that these copies were more up to date but doesn’t seem like a fair bargain at all. Day by day the prices of both copies multiplied. The main issue was that the sellers pegged their prices to each other, producing a feedback loop. One day the price went up to $23,698,655.93 only to drop to $106.23 the next day. Whether putting the blame on pricing algorithms on the platform or the sellers’ oversight, the emergence of AI-based pricing algorithms led to a number of debates concerning the existence of algorithmic collusion in relevant markets.
What is algorithmic collusion in the light of antitrust infringements?
Collusions to infringe provisions of competition law are not something new. For example, take a look at cartels as a classic example. Such practices have always been a problem to detect and regulate, yet the rapid growth of technology made things a little bit harder. Algorithmic collusion basically means that prices within digital marketplaces are set by AI rather than humans. Even though it is not illegal to design algorithms to respond to the dynamics of a particular market, viewing the issue as a whole gets us to the question of how it can affect competition. For example, it may introduce liability issues since the ability of AI-based algorithms to learn on their own may give birth to tacit collusion.
According to the OECD, two of the most significant structural features that affect the risk of infringing competition law are the number of companies and barriers to entry. The effect of algorithms on these structural traits can still be deemed as unclear. Industries in which algorithms are used to set out dynamic prices, improve product quality or analyze the consumer base are for example online marketplaces, airlines, social networks, search engines, booking agencies and so on. It is natural that these sectors of industry are characterized by certain barriers to entry such as network effects, economies of scope and economies of scale, enabling many companies to collect broad amounts of data and create more accurate algorithms. On one hand, such algorithms may lower barriers to entry due to enabling entrants to analyze market conditions and opportunities more accurately. In markets associated with a higher degree of transparency, companies can more smoothly monitor and respond to each other’s actions without actual human intervention. It has also been noted that sharing pricing algorithms with other participants on the market could infringe antitrust provisions in a really subtle manner that excludes direct or indirect communication. For example, if market participants outsourced the design of such algorithms to the same developers, it could create a kind of ‘hub and spoke’ scenario.
Different types of algorithms = different types of risks?
Let’s break algorithms in four different categories as done by the OECD and take a look at possible competition law risks and implications.
Monitoring Algorithms
This category executes the most obvious function when looking at antitrust infringements. Monitoring other market participants’ actions may give rise to a collusive agreement through the collection of data on market rivals and screening for potential deviations. According to OECD’s reasoning, monitoring algorithms are able to facilitate illicit agreements, yet they don’t eliminate the need for explicit communication.
Parallel Algorithms
By using parallel algorithms, undertakings can facilitate illegal agreements without the need to explicitly communicate, especially if they use the same pricing algorithm. Parallel algorithms can automate the undertakings’ decision-making processes so that prices react simultaneously to changes in market conditions. This specifically comes in handy within highly dynamic markets characterized by continuous supply and demand changes.
Signaling Algorithms
Signaling algorithms can also be used in highly dynamic markets to facilitate illicit agreements. Such markets produce a need to design more complex strategies of illegal cooperation among participants. One of these strategies could be the use of signaling algorithms through, for example, unilateral price announcements in the expectation that rivals will follow such a lead.
Self-Learning Algorithms
Last but not least, the fourth category is about using machine and deep learning techniques that could easily enable monopoly-associated results without the need to explicitly communicate or explicitly design algorithms to do so. Self-learning algorithms could represent a real type of ‘digital collusion’ since infringements would be even harder to prevent using traditional tools of antitrust investigation.
Current legal and practical implications.
In the context of international commerce, there is a broad consensus on antitrust policy in the area of collusion and infringements, as seen in Article 101 of the European TFEU, and section 1 of the Sherman Act in the United States. Noteworthy, tacit collusion per se is not illegal under EU competition law. Namely, Article 101 TFEU bans coordination between undertakings by laying down three categories of illicit conduct, namely agreements, decisions by associations of undertakings and concerted practices. The debated point may include the question of whether these categories involve tacit collusion as an act of independent pricing achieved through the means of an AI-based pricing algorithm. While we currently lack empirical studies documenting the frequency of this occurrence in real-world scenarios, we can reflect on two cases, namely the 2016 European Court of Justice’s judgment in the Eturas case and the Commission’s imposing fines on electronic goods manufacturers in 2018.
In the Eturas case we can see the Court’s reasoning on the conduct of concerted practice. The case involves a Lithuanian online travel booking system used by more than 30 travel agencies. Through its internal messaging system, a system notice was sent that announced that discounts were capped at 3 percent, and that higher rebates would be automatically reduced by the software to 3 percent unless a travel agency took additional technical steps. The message was available to everyone in the system’s section under ‘information messages. Due to only two agencies accessing the message, no one replying or taking public distance from it, the Lithuanian competition watchdog found an infringement. The ECJ held that travel agencies that were aware of the message could be presumed to have participated in concerted practice unless they had distanced themselves, yet absent further evidence and taking into account the presumption of innocence, the mere sending of the message couldn’t justify the notion that all travel agencies had to be aware of the message’s content.
Back in 2018, the European Commission fined a number of electronic goods manufacturers such as Asus, Philips, Denon & Marantz and Pioneer for targeting online retailers that offered their products at low prices and threatening them with subsequent sanctions such as blocking bonuses and supplies. The interesting part refers to achieving compliance with the manufacturers’ wishes by the use of algorithms that monitored the price setting in resales and adjusted prices that decreased. The wide impact of such coordination through a pricing algorithm can be seen in the fact that these manufacturers had seats in different parts of the world such as Taiwan, Japan and Netherlands, but affected consumers in EU Member States. However, due to finding physical evidence of written communication between manufacturers and resellers, it was not hard to determine that the infringement happened via pricing algorithms.
Concluding Remarks
Even though we have been riding the bandwagon of technological innovations for a while now, the issue of using algorithms to infringe competition law is still considered as a relatively distant and novel subject. On the other hand, the matter has also been referred to as ‘an old type of wine in new bottles’, meaning that it basically comprises a more elegant manner of implementing the same practices known for centuries. It has also been highlighted in modern literature that we should focus more on blockchain-based collusion and the use of smart contracts in relation to possible antitrust infringements. Nevertheless, the use of algorithms to infringe provisions of competition law may get us to thinking whether there is a need to revisit the concept of agreement as the ‘meeting of minds’ along with the scope of antitrust liability.
Sources:
Books/Journal Articles/Publications:
Beneke, F. & Mackenrodt M. O. (2020) Remedies for Algorithmic Tacit Collusion. Journal of Antitrust Enforcement, Vol. 9, Issue 1, pp. 152-176.
Kokkoris, I. (2020) A Few Reflections on the Recent Case Law on Algorithmic Collusion. Competition Policy International, Antitrust Chronicle – July 2020.
Mazundar, A. (2022) Algorithmic Collusion: Reviving Section 5 of the FTC Act. Columbia Law Review, Vol. 122, pp. 449-488.
OECD (2022) OECD Handbook on Competition Policy in the Digital Age. OECD.
Schrepel, T. (2019) Collusion by Blockchain and Smart Contracts. 33 Harv J.L. & Tech. 117.
Legislation:
Consolidated versions of the Treaty on European Union and the Treaty on the Functioning of the European Union (2016) OJ C 202/01.
Sherman Antitrust Act of 1890, 15 U.S.C. §§ 1-38,
Case Law:
Commission decision of 24.07.2018. Relating to proceedings under Article 101 of the TFEU (2018) AT.40465 – ASUS.
Commission decision of 24.07.2018. Relating to proceedings under Article 101 of the TFEU (2018) AT.40469 – Denon & Marantz.
Commission decision of 24.07.2018. Relating to proceedings under Article 101 of the TFEU (2018) AT.40181 – Philips.
Commission decision of 24.07.2018. Relating to proceedings under Article 101 of the TFEU (2018) AT.40182 – Pioneer.
Eturas UAB and Others v Lietuvos Respublikos konkurencijos taryba (2016) C-74/14.
