ANALYSING THE IMPLICATIONS OF ALGORITHMIC COLLUSION: THE WAY FORWARD FOR THE CCI?
Updated: Feb 27
This post is authored by Kunal Singh, 2nd-year B.A.LL.B (Hons.) student at Vivekananda Institute of Professional Studies, GGSIPU.
Enterprises in this modern world are heavily reliant on algorithms to track and amass data for delivering efficient and targeted services to consumers. Although not defined clearly, an algorithm is an application of reasoning by Artificial Intelligence ("AI") for finding a probable solution to a problem. This reliance on algorithms can be attributed to their cost and efficiency benefits. Enterprises such as Netflix and Spotify use algorithms to deliver a personalized user experience by looking into the search history and activity of the user. These algorithms contribute to an enterprise in numerous ways to facilitate the conduct of business, such as by controlling production, collecting market data, and self-learning through regular monitoring of market conditions.
However, they can also raise antitrust concerns depending upon their functionality. For instance, the personalized pricing used by enterprises to curate a user-friendly experience often dodges the antitrust regulators’ eyes. It causes significant consumer harm by subjecting them to higher prices based on their purchasing thresholds. This article seeks to critically analyze the implications of algorithmic collusion and the approach of the Competition Commission of India ("CCI") towards curbing it.
2. ALGORITHMIC COLLUSION
2.1 Dynamic Pricing
AI allows an enterprise to check and monitor market conditions and rival competitors’ activities and enables them to deliver targeted offerings. It mostly functions in two-sided markets where both buyers and sellers meet to exchange a product or service. Enterprises use AI facilitated dynamic pricing, often called ‘surge pricing’ (for cab aggregators), which is different from traditional pricing. It helps to find the right balance between price and availability that offers the maximum profit and modulates price as per the demand. It facilitates continuous prediction, adjusts prices within minutes by reviewing market numbers, and generates revenue by reducing speculation.
Cab aggregators such as Ola and Uber, use surge pricing to determine the price for a particular ride. When someone books a cab on their platform, AI quickly reviews driver availability, traffic in the route, consumer’s history, and frequency of using the platform before displaying a price for that ride. This is the major reason behind the variability of prices during different times. Cab aggregators function in an oligopolistic market whereby an enterprise’s performance depends on the rival firm’s policies and pricing strategies. Easy availability of information of rival competitors allows AI to monitor their pricing strategies and determine their prices, causing indirect price fluctuations. Such fluctuations do not allow prices to change quickly, reducing substitutability and causing harm to consumer welfare. Such fluctuations in prices, however, are not anticompetitive in themselves.
2.2 Price Personalisation
Likewise, many enterprises use AI to look into consumers’ preferences, search history, and the purchasing threshold, to display prices accordingly. This is known as price personalization or more precisely, price discrimination. It is a form of pricing that involves charging different prices to consumers according to their willingness to pay and primarily focuses on B2C relations. A great example of price personalization is the hotel website Orbitz, which uses AI to collect data such as browser type, zip code, and even the type of device consumer uses, and determine prices accordingly. For example, consumers using Mac devices may expect higher prices. Here, AI predicts that consumers using Mac devices are able to pay more than other consumers, and likewise displays higher prices. Even though this form of pricing fosters consumer loyalty and a user-friendly experience, it also entails significant consumer harm.
The absence of any explicit agreement indicating coordination between/among enterprises is a major regulatory challenge for the CCI and gives rise to Tacit Price Collusion ("TPC"). This is different from the type of collusion covered under section 3(3) of The Competition Act, 2002 ("the Act"), which governs collusion through an explicit agreement between enterprises/group of enterprises. TPC arises where enterprises/groups of enterprises collude informally (without explicitly agreeing) with their rivals to determine their strategies.
According to Ariel Ezrachi and Maurice Stucke, AI can collude in four possible ways:
1. Hub and Spoke Cartels: In cases of hub and spoke cartels, there is an informal coordination between people operating at the same level of the production chain (horizontal price coordination) and the persons/enterprises operating at a different level of the production chain (vertical level), in order to determine prices. In the Eturas case, many Lithuanian travel agencies or spokes used a common platform or hub (“Eturas”) for limiting the possibility of applying discounts of more than 3%, which was likely to have an anticompetitive effect. On the other hand, Eturas acted as a facilitator by circulating the travel agencies’ discount cap message. The Court of Justice of the European Union asserted that the travel agencies’ actions constituted a concerted practice leading to collusion since the travel agencies can be presumed to know that other travel agencies might be using the same platform.
2. Messenger: In these cases, enterprises use algorithms to effect anticompetitive agreements that human beings have previously entered into. The Poster & Frames Case (UK) and the David Topkins Case (USA) are classic landmark decisions in determining antitrust regulators’ stand against algorithmic collusion. In both cases, online sellers agreed not to undersell other competitors, and in pursuance of this, they implemented an agreement through ‘automatic pricing software,’ which automatically adjusted online sellers’ prices to remain aligned to each other. Commissioner Margrethe Vestager considered ‘the risk that automated systems could lead to more effective cartels’ as a potential challenge for cartel enforcement.
3. Predictable Agent: In such cases, each firm/enterprise unilaterally creates an algorithm that continuously monitors and adjusts to another algorithm’s pricing strategies. It can be possible that no two firms are using the same set of algorithms, but the unique algorithms used by both firms are programmed to adjust according to pricing fluctuations of each other.
4. Self-Learning: In these cases, algorithms with excellent predictability skills, by continually monitoring and readapting to competitors’ actions, may be able to collude without human intervention. This type of collusion may happen with or without the knowledge of the programmer. Generally, this type of collusion happens due to the failure of the programmers in implementing necessary safeguards. According to an OECD report, it has been ascertained that algorithms will be able to achieve a cooperative equilibrium with continuous hit and trial method and at a faster speed than humans.
3. CRITICALLY ANALYSING THE CCI'S APPROACH
With regards to this issue, the CCI missed a critical opportunity and adopted a parochial view in the case of Samir Agrawal v. ANI Technologies Pvt. Ltd., which was also upheld by the NCLAT. The CCI in this dismissed a case of a possible hub and spoke cartel against cab aggregators, i.e., Ola and Uber. Further, in an appeal to the Hon’ble Supreme Court of India, the court upheld the orders passed by CCI and NCLAT, stating, that Ola/Uber do not aid cartelization between drivers and also held that drivers are independent individuals who function independently of one another.
Both Ola and Uber in their ‘terms and conditions’ state that they are merely a ‘technological platform’ and ask the users to recognize that they do not provide any type of transportation and logistics services and such services are provided by an independent ‘contractor’ or ‘third party.’ This means that the drivers are not employees/workers of Ola/Uber and are merely independent contractors. Albeit they are independent contractors, they entirely lack bargaining power as they are not at liberty to determine the price of each ride, which are determined by Ola/Uber’s pricing algorithms. Here, the concept of hub and spoke cartels gains significance as cartel members do not communicate directly in order to keep their behaviour aligned but use an intermediary to do so.
The Samir Agrawal case is in some sense, similar to the Eturas case discussed earlier. There, many Lithuanian travel agencies also used Eturas as a common hub to communicate regarding the fares’ discount cap. The primary contention of CCI while dismissing the case in Samir Agrawal was that for a cartel to exist, there has to be some conspiracy leading to collusion, and mere assent of drivers to use a common intermediary cannot be said to amounting to collusion. Here, CCI went on with a very narrow understanding of ‘collusion,’ thereby disregarding ‘tacit collusion.’ Similarly, in the Eturas case, there was no direct communication between the travel agencies, just like there is none between the cab drivers, nonetheless, speculation of the participation of travel agencies existed.
Moreover, the drivers entered into an agreement with Ola/Uber actively knowing that other drivers might also be doing the same. Following the reasoning adopted in the Eturas case, drivers’ actions acting as spokes, and Ola/Uber acting as hubs, culminated into a hub and spoke cartel. Thus, the author believes that CCI should have taken into consideration the reasoning adopted in the Eturas case and taken a more holistic approach while defining the collusion between Ola/Uber as ‘tacit.’ Further, the CCI should now use this reasoning if any case of the hub and spoke cartel comes up in the future with similar facts and circumstances.
4. POSITION OF INTERNATIONAL JURISDICTIONS ON TPC
4.1 United States of America
In the US, the antitrust framework is governed by the Sherman Act ("Act of 1890"), the Federal Trade Commission Act, and the Clayton Act. Section 1 of the Act of 1890 prohibits agreements between two or more individuals restraining free and fair trade. Although Section 1 of the Act of 1890 includes the term ‘agreement’, the question that if the term ‘agreement’ includes ‘tacit agreement’ has been widely debated in several landmark cases. In Bell Atlantic Corp. v. Twombly, the court held that the ambiguous behaviour of ‘tacit agreements’ is consistent with a conspiracy that aligns with the broad market and competitive strategies. Scholars like Richard Posner had argued that ‘tacit collusion’ should be included within the ambit of ‘agreements’ under Section 1 of the Act of 1890, but the courts have repudiated this claim. The courts do recognize tacit collusion as harmful to consumer welfare, but they treat it as unavoidable for rational oligopolists.
This view of the court was held in In Re: Chocolate Factory Antitrust Litigation, where the court held that conscious parallelism or tacit collusion, even though it has anti-competitive outcomes, is not prohibited under the Act of 1890 for two reasons. Firstly, conscious parallelism is not an agreement, and it is “fact of life” for oligopolies. Secondly, it was held that tacit collusion is lawful because the courts do not have a remedy for such conduct. Therefore, the courts treat tacit collusion as unavoidable for rational oligopolists. The author believes that the CCI should not adopt the reasoning provided by the court as it is contingent upon the furnishing of the ‘plus factors.’ The requirement of plus factors will make identification of oligopolists, indulged in conscious parallelism, difficult.
Further, this view of the courts was emphasized in the Martha Vineyards case, which involved some gas station owners who were accused of price coordination between rival competitors. The First Circuit Court ruled in favour of the gas station owners, proclaiming the coordination to be particularly favourable for ‘conscious price parallelism’. Whereas explicit collusion under Section 1 of the Act of 1890 is illegal, ‘tacit collusion’ is legal per se. While the Act of 1890 has a narrow ambit, the Federal Trade Commission Act ("FTC Act") has a broader ambit. Under the FTC Act, the plaintiff has to show that a practice is unfair because; (i) it has or likely to have injury to the consumers; (ii) consumers cannot be reasonably expected to avoid it; (iii) its harms should not be overshadowed by its benefits to the consumers. Therefore, the CCI should preferably adopt the reasoning provided under Section 5 of the FTC Act.
4.2 European Union
Article 101 of the Treaty on the Functioning of the European Union (‘TFEU’) deals with agreements or concerted practices having anticompetitive effects on free and fair competition in the market. For any agreement to be made punishable under Article 101 of TFEU, actual collusion must be proved. Thus, tacit collusion or tacit agreements are not covered under Article 101. In the Wood Pulp case, the European Commission found that the wood pulp producers had engaged in collusive practices, though in the absence of any explicit agreement. However, the European Court of Justice overruled the Commission’s findings stating that mere parallelism does not result in concerted practices leading to collusion unless there is some exchange of information between the market players. Thus, the absence of hard evidence and language of Article 101 of TFEU made the jurisdiction of the European Union unsuitable for regulating and tackling ‘tacit price collusion.’
5. THE WAY FORWARD
From the above discussion, it is evident that the use of AI poses many antitrust risks that need to be dealt with. The major problem is detecting the algorithmic collusion, which becomes more difficult in the absence of an explicit agreement. Thus, the first measure to mitigate the risk of algorithmic collusion is to mandate the self-assessment and auditing of algorithms used by enterprises. The CCI should consider constituting an internal committee of economists and data scientists for laying down guidelines for self-assessment and aligning of algorithms with the extant competition law provisions. It would allow firms/enterprises to check whether the algorithms used by them signalled collusion signs, shifting the burden of proof to the companies to undertake compliance programs as suggested in the recently released e-commerce report. Through this measure, firms/enterprises can ensure fair and pro-competitive practices.
Secondly, the author believes that the CCI should adopt some structural remedies like divestitures. In the oligopolistic market model, symmetry in market dynamics is assumed. The primary motive of the CCI should be to alleviate the negative effects of tacit collusion which can be done by creating asymmetries between market participants. Divestitures aim to create asymmetry in market dynamics by either creating a large number of competitors or refashioning the competition dynamics, thereby making tacit collusion more difficult to achieve. In an oligopoly, pricing decision of a firm is not only influenced by the pricing strategies of other firms, but also by various non-price factors such as distribution channels and coordination between enterprises. The creation of asymmetries might therefore prevent algorithms to achieve collusive results, which can be easily achieved in a symmetrical market model. For instance, in the case of airlines where common pricing algorithm is the norm, the slots at the airport could be sold to low-cost carriers not operating on the concerned route, thereby creating asymmetric conditions in the market. In asymmetric market conditions, AI also finds it difficult to self-learn and achieve collusive outcomes.
Thirdly, due to the increasing use of algorithms, firms/enterprises have developed a sense of collectivism (intentionally or unintentionally), attributing to AI’s role in matching/fixing prices across firms/enterprises. Individual firms in the relevant market, due to collusion, have become analogous to each other, yielding collective influence over the market. Under Section 4 of the Act, the CCI recognizes only unilateral dominance enjoyed by a firm or a group. The author believes that the CCI should equate such conduct as collective dominance. Although the CCI has expressed its inability to recognize such dominance as ‘collective,’ as the same cannot be penalized unless there is some legislative amendment to that effect.
Lastly, the CCI should adopt an approach identical to the approach used in Section 5 of the FTC Act in the USA. As discussed above, Section 5 of the FTC Act requires FTC to prove the apprehension of algorithmic collusion arising out of an arrangement. If used in India, it will allow the CCI to detect algorithmic collusion or ‘tacit price collusion’ between firms/enterprises at a nascent stage and prevent the firms from taking undue advantage emanating from such coordinated conduct.
The preceding paragraphs deal with the regulatory challenges posed by the use of algorithms. Unfortunately, the CCI missed the trick and failed to comprehend the nuances of tacit collusion in the Samir Agrawal case. The existence of ‘tacit collusion’ can be indeed expected to upscale the non-identification of cartels in the digital sphere. Further, the Draft Competition (Amendment) Bill 2020 suggests some changes that can be a step in the right direction, one of them being the inclusion of ‘hub and spoke cartels’ within the ambit of ‘cartels.’ Such inclusion will penalize hub and spoke cartels and shall expectantly help address tacit price collusion.