Algorithmic collusion in digital markets and AI: science fiction or reality?
As a result of the digitization of the economy, companies are more often using algorithms to solve a variety of issues. The application of artificial intelligence (“AI“) has made these algorithms more sophisticated. The growing influence of algorithms is accompanied by legal issues, such as potential competition law infringements through price algorithms. In recent years, competition authorities have been actively looking into scenarios in which algorithms are used to engage in anti-competitive practices. This blog focuses on algorithmic collision in digital markets.
What are (price) algorithms?
An algorithm is a finite set of unambiguous instructions to be followed in a computation, thereby systematically converting input into output. Algorithms have been used for centuries and are in itself not new. However, the advent of AI, Big Data, Data Analytics and The Internet of Things (“IoT”) casts algorithms in a new light. For more information on IoT, we recommend our earlier blog. Due to these technologies, present-day algorithms are often more complex than their predecessors.
Many companies use algorithms for price-setting. They generally use dynamic or personalised pricing-methods. Dynamic pricing determines the price based on fluctuations in supply and demand. This is fundamentally different from pricing algorithms using personalised data, where the price is determined per individual consumer and depends on the consumer’s personal characteristics. Depending on the intelligence of the price algorithm, pricing is based on historical data or real time data. Unlike algorithms using historical data, algorithms using real-time data can quickly respond to changes in markets and thus determine prices accurately.
Competition authorities’ growing interest in algorithms and AI
Algorithms are clearly on the political agenda. This is evident from the Rutte IV Coalition Agreement, presented on December 15 2021, which announces the introduction of an algorithm supervisor. This supervisory role has been assigned to the Dutch Data Protection of Authority (“DPA“). The DPA will mainly monitor algorithmic applications for transparency, arbitrariness and discrimination. The use of algorithmic applications also affects competition. In 2020, the Netherlands Authority for Consumers and Markets (the “ACM”) published Guidelines on the Protection of Online Consumer, followed by a Position Paper: Oversight on Algorithms, in which the ACM clarifies its position concerning the monitoring of algorithms. Last month, Martijn Snoep, chairman of the ACM, stated at a Blockchain event that algorithmic collusion is currently the biggest concern for the ACM. Especially because algorithmic collusion is extremely difficult to detect. During his speech, he indicated that the ACM is lagging behind developments. However, in order to change this, the ACM has set up a special technology-focused department. Competition authorities in other countries also focus on the growing influence of (price) algorithms on competition. For example, in January 2021, the UK Competition and Markets Authority published a market study that examined the extent to which algorithms are anti-competitive and harmful to consumers. The Norwegian Competition Authority also published a market study of a similar scope last year.
At European level, the European Commission (the “Commission“) is also examining the compatibility of algorithmic applications with European competition law. Earlier this month, the Commission published a draft Horizontal Block Exemption Regulation (“HBER“) and accompanying Guidelines on Horizontal Agreements (“Horizontal Guidelines“). The reason for these drafts is the expiration of the current HBER and Guidelines, later this year. An evaluation of the current Horizontal Guidelines showed that there is lack of clarity regarding the use of price algorithms. The draft Horizontal Guidelines show that the Commission has somewhat taken this criticism to heart. For example, the draft Horizontal Guidelines make a distinction between algorithmic collusion and collusion by code. The latter refers to intentional coordination between competitors, by means of a common algorithm. This is typical cartel behaviour and constitutes a restriction of competition by object. Algorithmic collusion, on the other hand, may be unintentional. The draft Horizontal Guidelines further note that for algorithmic collusion to take place, a number of structural market conditions are required. For instance, the presence of homogeneous products/services. The draft Horizontal Guidelines further state that the introduction of a price rule in a shared algorithmic instrument is likely to fall under the cartel prohibition. It is not required that such pricing is explicitly agreed upon.
In addition to the draft Horizontal Guidelines, in April 2021 the Commission proposed a Regulation laying down harmonised rules on artificial intelligence and amending certain sections of European Union law (“AI Regulation“). According to this proposal, each Member State must appoint a national supervisory authority to monitor the application and implementation of the AI Regulation. If the designated supervisory authority encounters competition law issues, it must inform the national competition authority. In addition, the Commission proposed a Digital Markets Act (“DMA“) and a Digital Services Act (“DSA“). We have previously devoted a blog to the DMA. Since then, the DMA has undergone several changes. On 24 March 2022, the European Parliament and the Council agreed on the content of the DMA. The text will soon be technically finalised. After that, formal adoption by the Council and the European Parliament is required. For a complete overview of the current status of the DMA, consult this webpage of the European Parliament.
All these proposals show a clear trend that the Commission wants to regulate the use of algorithms and AI in digital markets.
Algorithmic collusion in digital markets
From a competition law perspective, collusion by price algorithms may cause infringements of Article 101 of the Treaty on the Functioning of the European Union (“TFEU“) and Article 6 of the Dutch Competition Act (“Mw“). According to competition authorities and literature on the matter, two main types of collusion can be distinguished: explicit collusion and tacit collusion. In the case of explicit collusion, a price algorithm is used to implement and/or monitor an already existing cartel agreement. This offers a significant advantage to the cartelists as it stabilises the cartel agreement and reduces the likelihood of deviations. In the case of tacit collusion, coordination is not explicitly agreed between the undertakings concerned. Three different variations of tacit collusion can be identified:
- In a hub-and-spoke scenario, tacit collusion arises when rival companies – also known as spokes – choose not to develop their own price algorithm, but to use a third-party algorithm. If competing companies purchase a price algorithm from the same supplier – a so-called hub – the hub may have an incentive to raise prices above the competitive level. In that case, the hub facilitates a cartel between the spokes through a price algorithm. This form of tacit collusion is also explicitly mentioned in the draft Horizontal Guidelines.
- A predictable agent is a price algorithm that reacts in a predictable way to external factors. The algorithm is instructed to monitor market prices, to follow price leadership and to punish deviations from tacit cooperation. These instructions may lead to coordinated pricing. This is particularly likely if the price algorithm is programmed to set prices in a simple, transparent and predictable manner.
- An autonomous machine algorithm aims to develop an optimal strategy. Subsequently, it receives certain input, such as the instruction to maximise profit. Next, the algorithm learns how it can best achieve that goal. The possibility arises that an algorithm gradually learns that collusion with algorithms of competitors is the optimal strategy for achieving profit maximisation. As a result, without human intervention, collusion can occur between the algorithms of competing companies.
Assessment of algorithmic collusion under the cartel prohibition
The detection and assessment of algorithmic collusion constitutes a major challenge for competition authorities. For that reason, there have been only a few algorithmic collusion cases. A well-known case is the Eturas case, in which an operator of an online booking platform for travel agents had sent – via the platform’s mailbox – a message to travel agents, informing them that the discounts for products sold via the platform were capped. Subsequently, the platform operator implemented the change in the platform’s system. The Court of Justice of the European Union (the “ECJ“) examined whether the travel agents were liable. According to the ECJ, no participation in the alignment could be established as long as it was not proven that the travel agents were aware of the changes in the platform’s system. This shows that the ECJ applies the elements for a concerted practice in a flexible manner. The bar for coordination is set relatively low, as the travel agents would have been guilty of coordination if they had read the message and/or had been aware of the change in the system. In which case, the travel agents should have publicly distanced themselves from the message. It is therefore clear that companies – in this case the travel agents – can coordinate with each other without knowing it, because the coordination takes place via a platform. The Eturas case has also been included in the draft Horizontal Guidelines.
Another case in which algorithms facilitated a cartel agreement concerns the 2018 Consumer Electronics case, in which the Commission imposed fines on electronics manufacturers Philips, Pioneer, Asus, Denon and Marantz. They were fined for restricting retailers from independently setting sales prices, thereby keeping sales prices high. The electronics manufacturers did this by using price algorithms to monitor sales prices and put pressure on retailers to align their prices with those of competitors. This form of algorithmic price coordination was proved by the physical evidence, which mainly consisted of written communications from the electronics manufacturers to retailers, informing them of the policy and the consequences if they did not comply.
The benefits of price algorithms
Price algorithms also have advantages. They enable companies to respond faster and better to changes in markets, allowing for a better match between supply and demand. In addition, price algorithms may lead to a significant reduction in production and transaction costs. Because price algorithms can generate efficiency gains, an exemption from the cartel prohibition may be desirable in such cases. For example, in the Webtaxi case, an exemption was granted by the Luxembourg Competition Authority, because the joint use of a price algorithm by competing taxi companies led to greater cost efficiency and more favourable prices for consumers.
Future regulation of digital markets
Currently, competition authorities are working on various legislative proposals concerning the regulation of digital markets. The Commission has already made proposals for new Horizontal Guidelines, the AI regulation, the DMA, and it is likely that it will not stop there. Also on a national level, the ACM focusses on the supervision of the use of algorithms and AI in digital markets. As Martijn Snoep, chairman of the ACM, recently pointed out, algorithmic collusion is currently the biggest concern for the ACM. Especially since detecting algorithmic collusion is extremely difficult. To solve this, competition authorities are employing more programmers and data experts. This shows that future enforcement in respect of algorithmic collusion is to be expected.