top of page
  • Yash Arjariya

Algorithmic Collusion in an Oligopolistic Market: Difficulty in Gauging Conscious Price Parallelism

[Yash is a student at Hidayatullah National Law University, Raipur.]

An oligopolistic market structure is characterized by significantly fewer competing entities, homogenous products, and inelastic demand. Price parallelism and supra-competitive prices are, according to European and American jurisprudence, only natural outcomes of such market structures and thus not illegal per se. The jurisdictions take into account plus factors like communications and information exchanges to gauge whether such price parallelism is conscious price parallelism or not.

According to a European Union (EU) industry study performed in 2017, a significant majority of online merchants use pricing software for their enterprises. The use of an algorithm for automated pricing changes is a necessary activity for major merchants with a variety of several thousand distinct items. In the United States (US), David Topkins and Trod cases revealed examples of algorithmic collusion in which price-fixing algorithms were used to keep prices artificially high and collude with a rival company. This is critical when organisations employ the same pricing software even without knowing about it, leading to a scenario where both algorithms fix prices in the same direction. As a consequence, the price of two firms might – absolutely inadvertently – seem to be anticompetitive. Similar was the situation with the book The Making of a Fly, where the interplay of both sellers’ algorithms caused constant positive feedback, resulting in a price spiral.

Therefore, algorithmic collusion pertains to the concept that pricing algorithms might autonomously realise tacit collusion. By virtue of coordinating their price fixing through such algorithms, businesses may collectively maximize their earnings without an express agreement to do so. The article analyses the problem before the antitrust regulators in establishing guilt in such cases. The article then accounts for the jurisprudence around Competition Commission of India (CCI) in this regard. Finally, the article concludes by outlining the need for pro-active market screening and the adoption of a ‘harm-based’ approach by the regulators.

Subverting the Conventional Conception of Anti-Competitive Agreement

In algorithmic collusion, competitors may raise their prices as a signal for other competitors to follow, i.e., in algorithmically adjusted pricing, the competing product’s price will automatically be adjusted in response to the price hike. Algorithms re-adjust prices based on a multitude of factors, with the pricing of competing products receiving the most weight. As a result, price adjustments may result in artificially high pricing and price stabilization.

Across jurisdictions like the EU (Bayer AG v. Commission of the European Communities, paragraph 69), Australia, the USA, etc., establishing collusion necessitates the establishment of intention or concurrence of wills between parties. In conscious parallelism, competitors unilaterally increase prices above the competitive level in reaction to the actions of their rivals, but without “any unlawful arrangement or even any contact or communication among the competitors.” This makes it simpler to produce the same collusive consequences even without explicit agreements or coordinated activities, as the German Monopolkommission mentioned in its report.

The existing jurisprudence is not readily receptive to this idea and practise of collusion. In Eturas’s case, the Court of Justice of the EU undoubtedly meant to underline that the technology by itself is not capable of breaching competition. Participants must establish a certain level of communication and consent. In this context, TreuhandAG v. European Commission most liberally construed interaction between cartel participants as ‘mutual action,’ if not a firm’s response to an illegal cartel initiative. However, the ruling in Imperial Chemical Industries Ltd. v. Commission of the European Communities upheld the competitor’s right to change the pricing based upon taking into account the present or foreseeable conduct of his competitors. This caveat on the application of ‘mutual action’ becomes more glaring in the light of the increased use of self-learning or machine-learning algorithms, by virtue of which coordination can be done with the mere prediction of the other without any need to enter into any agreement to do so. In such cases, a firm can vigorously defend allegations of concerted action by contending that the algorithm used was simply reacting to a price increase on a competing product. As per Ezrachi and Stucke, this raises particular apprehension that competitors may elude detection for their price-fixing carried out through algorithmic collusion.

Indian Jurisprudence

The CCI’s decision in Samir Agrawal v. CCI (Samir Agrawal) may be considered a missed opportunity for dealing with algorithmic collusion. The regulator established the direct evidence test by stating that collusive behaviour requires a meeting of minds or agreement. This approach of CCI was along similar lines as in Re: Domestic Air Lines, wherein the Director General (DG) concluded parallel behaviour between airlines to increase prices after strikes by Air India employees. However, the CCI did not accept the findings of the DG on account of a lack of material evidence establishing agreement between the parties.

Re: Alleged Cartelization in the Airlines Industry (Airlines Industry) marks a departure from CCI’s existent jurisprudence. CCI developed a two-step test for dealing with algorithmic collusion. Firstly, CCI looked for human communication between entities. The role of algorithms was examined in the second step. The test met its logical end in Shikha Roy v. Jet Airways (Shikha Roy), which also marked the dawn of proactive invigilation against algorithmic collusion in India. Unlike the Airlines Industry case, where CCI looked into alleged algorithmic collusion after DG’s report, in Shikha Roy, CCI ordered DG to look at the role of algorithms during the investigation itself.

The two cases referred to above point towards standardization of the two-step test, which the CCI may find customary in its application while dealing with cases of algorithmic collusion. However, the CCI’s jurisprudence, though developing, is far from adequate on two grounds. First, when determining algorithmic collusion, the test is more specific in gauging whether the competitors used a common algorithm. The use of different platforms or software by competing firms may preclude an assessment of guilt by the regulator. Secondly, the test is still a hangover of the traditional approach because it necessitates gauging ‘intention.’ As laid out in the Airlines Industry case, the extent of ‘human intervention’ is one of three pivotal factors of Step 2 of the test while determining guilt. The author submits that cartels may arise in marketplaces where computers arrange and manage the interchange of information about pricing, despite the fact that the original purpose of the competitor was not to engage in forbidden anti-competitive agreements. Subjective ideas, such as intent, are exceedingly difficult to transfer and incorporate into the digital setting.

The Remedy

Identification of suspicious collusive behaviour is a significant issue because pricing algorithms have been in use for a long time; thus, it remains intriguing to assess whether we are attempting to compare anomalies with other anomalies that have become the ‘new’ normal as they have supplanted normal market conditions long ago. Scanning market developments is an important aspect of antitrust regulation in the age of pricing algorithms. Since the establishment of intention to collude remains an open question in pricing algorithms, the German Monopolies Commission specifically advises actively monitoring marketplaces for price anomalies. Against this background, it is of great utility to refer to the successful application of such a monitoring instrument in the London Interbank Offered Rate Case. The authorities used monitoring tools to monitor interest rates. They were alerted to signs of suspected collusion because of flagged interest rate changes that happened after 2007. Despite the tensions on the unsecured interbank money market, low interest rates were evidently recorded, prompting targeted screening for market manipulation and collusion in the US and Europe.

When pricing decisions are not directly attributable to market conditions like the cost of raw materials, transportation costs, marketing costs, etc., the price stability of competing products may attract suspicion through regular monitoring by regulators. However, due to the difficulty in applying traditional notions of ‘agreements’ to such a problem, the collusive activities may not be entirely traceable. In this respect, the US Policy Council of the Association for Computing Machinery supports the imposition of liability on every user of an algorithm, even if the pricing decisions are not entirely traceable. The author is of the suggestive stance that a transition from intent to consumer harm must be effected, and the quantum of consumer harm must become a valid and feasible threshold for differentiating between unlawful collusion and permissible oligopoly action when pricing algorithms are involved.


Through the CCI Market Study on E-commerce and Competition Law Review Committee Report, it can be safely concluded that CCI has passed the denial phase of the possibility of algorithmic collusion, albeit such an acknowledgement is merely theoretical. The recent jurisprudence of CCI in the Airlines Industry and the Shikha Roy case is an appreciable development from the denial phase after the Samir Aggrawal judgement. The author remains optimistic that the two-step test by CCI will only mature and manifest practical application in time to come. On a global scale, the studies by the OECD, the German Monopolies Commission, or the UK’s Competition and Markets Authority are a strong indication of emerging instances of behavioural coordination between competing entities on account of usage of their usage of pricing algorithms. Though parallel behaviour is not identified with concerted practice in an oligopolistic market structure, the author contends that it must be regarded as significant evidence pointing to the presence of collusive parallel conduct if it leads to circumstances of competition that do not correspond to the regular conditions of the market, taking into account the unique peculiarities of the market in question. As a corollary to the argument advanced in earlier section, the strict requirement of ‘agreement’ must be diluted while adapting the ‘harm-based’ approach supported by pro-active monitoring of pricing patterns with regards to market conditions.


Related Posts

See All


bottom of page