Algorithmic Price Coordination and the Limits of Competition Law

Introduction : Two sellers on the same online marketplace can find themselves pricing each other’s products within a few rupees for months, without either ever sending an email to the other. Both will have attached their price listings to a repricing tool, which tracks the competitor’s price and makes automatic adjustments of its own product’s price, often within seconds. The result can be equivalent to a cartel of prices: discounts do not appear on both products simultaneously, and neither seller appears willing to lower its price enough to draw away customers. Competition laws are silent on such situations, being written at a time when coordination of this sort had to be achieved through phone calls and meetings. Section 3 of India’s Competition Act, 2002, Article 101 of the Treaty on the Functioning of the European Union (TFEU), Section 1 of the US Sherman Act, and Section 1 of the UK Competition Act 1998, all prohibit any agreement or, in the EU, a concerted practice. But none of them contemplate a situation where the coordinating mechanism is an algorithm, not a person, which neither seller has created themselves but purchased as a commodity.

This blog post explores the ways in which Indian, US, UK and EU competition laws approach the challenge presented by algorithmic coordination, analyzing the tests for anti-competitive agreements used in each territory, the newly proposed hub-and-spoke theory of harm in India, the difficulties presented by algorithms to the collection of evidence, and finally identifying where the line between anti-competitive facilitation and independent, pro-competitive parallelism might be drawn in each case.

Legal Provisions

A. India: The Competition Act, 2002

Section 2(b) of the Competition Act, 2002 defines an “agreement” generally and includes “any agreement” between parties whether express or otherwise, written or unwritten. Section 3(1) makes it unlawful “for any agreement which causes, or which, directly or indirectly, has the effect of causing or which, directly or indirectly, is likely to have the effect of causing an appreciable adverse effect on competition in India.” Until 2023, that deemed having an appreciable adverse effect on competition was limited to those persons “carrying on the same or substantially similar trade” leaving a lacuna for technology vendors or intermediaries that may facilitate cartelization without competing directly with anyone. The Competition (Amendment) Act, 2023 fills this lacuna by adding another proviso to Section 3(3): a person not dealing in the same or substantially similar trade as the cartel is “deemed to be a party to the cartel agreement” if it “participates or has participated or knowingly facilitates or has facilitated the commission of any act for the furtherance of such agreement.” This is the legal basis for the hub-and-spoke liability in India, which applies to algorithmic pricing because the modern-day “hub” often takes the form of a software intermediary.

B. United States: The Sherman Act

Section 1 of the Sherman Act prohibits “every contract, combination…., or conspiracy, in restraint of trade.” Unlike the Competition Act, the language of the Sherman Act has consistently been interpreted by American courts to require an actual agreement, not just interdependent behavior, for an anti-competitive outcome. With that said, firms that simply respond to another firm’s behavior in a way that causes prices to converge on one value do not violate the Sherman Act, even if the prices are nearly identical. The problem raised by algorithmic pricing cases is therefore not that such collusion through coding is illegal, as it very often is, when a price-fixing agreement can be established. It is rather whether using the same algorithm, which allows two or more companies to exchange information, can be classified as an agreement itself.

C. European Union: Article 101 TFEU

Article 101 (1) prohibits both agreements and ‘concerted practices’ between undertakings. The idea of a concerted practice was created precisely to capture coordination that falls short of a contract: a form of cooperation which deliberately replaces the risks of ordinary competition among firms by practical cooperation between the firms. This is a much lower threshold than the Sherman Act’s contract-or-conspiracy requirement, and it is the provision under which the Court of Justice of the EU decided Eturas, discussed below.

D. United Kingdom: The Competition Act 1998

Chapter 1 prohibition in UK Competition Act 1998, which covers anti-competitive agreements and concerted practices by organisations in UK, is similar to Article 101. Guidance published by the Competition and Markets Authority in 2021 regarding algorithms illustrates that competition rules apply to whatever methodology the party adopting an agreement adopt: a decision not to undercut a competitor is illegitimate, whether it is captured in an email exchange or encoded in repricing software.

Legal Analysis

A. Independent parallel pricing is not unlawful

The starting premise for any of the jurisdictions discussed above will be the same: firms that act independently but are reacting to the same factors and happen to reach similar prices cannot be considered to have conspired in setting them. If two gas stations on the same street raise their prices when crude oil prices increase, there is no collusion just as there is none between two ride-sharing companies that adjust their fares after the same rainstorms. In each of these examples, the firms simply react to traffic, demand, and the weather rather than each other’s actions, and this nuance, which is sometimes called conscious parallelism, is the reason why a shared pricing algorithm is not enough to prove conspiracy.

B. A common vendor as the hub

The more difficult case is where several competitors feed their live prices, costs, or occupancy data into one third-party algorithm and that algorithm calculates a recommended price for each client partly on the basis of what its other clients’ charge. Here the vendor becomes a channel for one competitor’s confidential pricing intentions to reach another, even though no employee of either competitor has spoken directly to the other. A vendor was not directly amenable to the antitrust laws of India before 2023, since it did not compete in the same trade as the sellers it served. The amended proviso to Section 3(3) overcomes this difficulty, since it interprets the vendor as participating in the cartel wherever it will facilitate or will intend to facilitate the execution of the agreement. In litigation in the United States concerning RealPage, however, the Department of Justice Statement of Interest asserts that the combination of the sharing of non-public rent and occupancy data amongst competitors via the use of a common algorithm in conjunction with a reasonable expectation that landlords would follow the suggested rental levels created the agreement required by Section 1, even though no landlord communicated directly with another.

C. Algorithmic awareness without a shared vendor

A more interesting case concerns two competing companies with totally different algorithms, where neither looks at the other's publicly posted prices and reacts to them but without any information exchange, common vendors, or human contact. Sometimes this is called algorithmic awareness, and none of the four regimes have a definitive answer. While the EU's concerted-practice doctrine provides a wider theoretical framework than does the Sherman Act's contract requirement, it has always required some kind of contact beyond one of the companies had a bot that looked at the other's webpage. Purely reactive scraping of public prices, under the current doctrine, is more likely to be seen as a peculiarly quick means of price-matching than collusion, no matter how uncomfortable it would make a regulator.

D. Evidentiary challenges

Even if it is possible to identify a particular shared vendor or hub, it can still be difficult to distinguish between the outcome of an illegality and an innocuous coincidence. According to Eturas, the mere fact of a message being relayed through a common channel would not be enough to demonstrate that it was the result of a collusive design, because implicit acceptance, as shown by a failure to object to a restrictive message, to report the matter to the competition authority and to circumvent the purported conspiracy, could be sufficient for an inference of conspiratorial reliance, subject, of course, to a rebuttal and to the presumption of innocence. In an algorithmic cartel case, investigators need to demonstrate three things: that the algorithm that produced the collusive outcome was using private information, not publicly available prices; that the collusive outcome was an output that the algorithm was designed to produce rather than a coincidence; and that the users of the algorithm actually followed it. Samir Agrawal highlighted the absence of the first two pieces of evidence, since each firm was using its own algorithm for its own platform and there was no evidence that either algorithm had access to the other’s fares.

Relevant Case Laws

Samir Agrawal v. Competition commission of India & Ors., Civil Appeal No. 3100 of 2020 (Supreme Court of India, 15 December 2020): Having upheld the order of the CCI dated 6 November 2018 and the order of NCLAT dated 29 May 2020, the Supreme Court dismissed the petition against the alleged collusion between Ola and Uber to fix prices of their drivers through surge pricing mechanism. To this end, the Court held that neither did these aggregators share any information on the implementation of their business practices, nor did they operate via a common third party ‘hub’ that controlled prices at the same time. Algorithms of the two operated on separate data, taking into account various elements (the weather, demand, traffic, etc.) in their unique way, responding to similar supply-demand patterns but not exercising control over one another. The Court concluded that algorithms that work separately with no common reference to share or analyse data in real time to set prices were not cartel.

In re RealPage, Inc., Rental Software Antitrust Litigation, MDL No. 3071 (M.D. Tenn.): antitrust case where the DOJ, and a few other States' attorneys general, have accused RealPage's algorithm of colluding, by sharing confidential data about rent and occupancy of competing landlords and issuing pricing recommendations to them that are expected to be followed. The DOJ, in its November 2023 Statement of Interest said that setting prices by a shared, data-pooling algorithm could constitute a per se Section 1 violation, even if there is no communication between the landlords.

Eturas UAB and Others v. Lietuvos Respublikos konkurencijos taryba, Case C-74/14 (CJEU, 21 January 2016): Travel agencies which offered their services via an online booking platform, which the system administrator advised them of the capping of online discounts which were subsequently technically enforced, was ruled by the CJEU as not establishing a concerted practice merely from the fact of receipt of the message by the travel agencies, but may find tacit acquiescence (by the absence of complaint to the competition authority) may be a sufficient basis for finding participation in a cartel, subject to rebuttal.

Competition and Markets Authority decision against Trod Ltd and GB eye Ltd (12th August 2016) is the first ever UK algorithm cartel decision. The case involved two Amazon marketplace sellers of licensed posters and frames who configured commercial repricing software to ensure that neither seller lowered their prices below the others between 2011 and 2015. Trod was fined £163,371 while GB eye was awarded leniency for reporting the cartel. The decision highlights that Chapter I liability extends to the agreement not to compete, which was manifested through the software configuration.

United States v. David Topkins, No. 3:15-cr-00201 (N.D. Cal. 2015): The first ever Sherman Act criminal prosecution hinged on pricing algorithms and occurred in the case of Topkins and his competitors of posters on Amazon Marketplace, who were using the software to automatically adjust the prices in accordance with each other’s moves, thus forming a cartel; he was found guilty and was forced to pay a USD 20,000 penalty. The cartel’s criminal conduct related to the human coordination of the pricing algorithms, which were not intrinsically illegal.

Practical Implications

A compliance review of any pricing system should watch out for these characteristics, in descending order of risk:

A vendor or platform that aggregates non-public price, cost, inventory or occupancy data from multiple competing clients and feeds it back into a shared model, rather than pricing each client from its own data and public information.

Contractual terms, ranking rules, or informal practice that deter a client from deviating from the algorithm's recommended price, since an expectation of conformity is what turns a mere recommendation into something close to an agreed rate.

Repricing software whose matching logic, discount caps or floors were set following discussion with a competitor, as in the Trod/GB eye arrangement, rather than configured unilaterally by each firm.

A player in a narrow market that is used by nearly every other player has a higher likelihood to be a nexus and not an uninterested supplier to the technology market.

Businesses can avoid this by acquiring or building pricing engines that are trained on only their data and publicly available market information, keeping records of the algorithm's design rationale and data inputs to showcase an independent, rational response to market conditions if questioned about it, and reviewing vendor contracts for parity or most-favoured-nation clauses Section 3(4) of the Competition Act, as amended in 2023, now reaches wider than before it did.

Conclusion

Existing doctrine such as Section 3 of the Competition Act as amended in 2023, the Sherman Act, the Chapter I prohibition and the Article 101 TFEU can all tackle the issue of algorithmic cartel formation, so long as there is an element of human coordination, or a genuine data-pooling mechanism, which can be demonstrated to have occurred. The coordinated action of pricing algorithms that base themselves on non-proprietary, publicly available data does not constitute anti-competitive behavior, as the Supreme Court ruled in the Samir Agrawal case. The only remaining grey area concerns price-fixing protocols based on mutual, but non-collusive awareness of one another’s algorithms; none of the four jurisdictions discussed have unequivocally condemned such price-fixing. However, with the facilitator liability introduced by the 2023 amendment and the CCI’s proposed 2025 market study on the subject of artificial intelligence, India is likely to have competition laws that update themselves before any new competition law legislation can be finalized.

Author: Paridhi Malik in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at  Khurana & Khurana, Advocates and IP Attorney

References (Endnotes)

1. Competition Act, 2002, s. 2(b) (India).

2. Competition Act, 2002, s. 3(1) (India).

3. Competition Act, 2002, s. 3(3) (India).

4. Competition (Amendment) Act, 2023, s. 4, inserting the second proviso to s. 3(3) of the Competition Act, 2002 (India).

5. Sherman Antitrust Act, 15 U.S.C. § 1 (1890) (United States).

6. Treaty on the Functioning of the European Union, art. 101(1).

7. Competition Act 1998, c. 41, Chapter I prohibition (United Kingdom).

8. Competition and Markets Authority, Algorithms: How They Can Reduce Competition and Harm Consumers (Jan. 2021) (United Kingdom).

9. Competition (Amendment) Act, 2023, s. 4 (India).

10. In re RealPage, Inc., Rental Software Antitrust Litigation, MDL No. 3071, Statement of Interest of the United States (M.D. Tenn., filed Nov. 15, 2023).

11. Eturas UAB and Others v. Lietuvos Respublikos konkurencijos taryba, Case C-74/14, ECLI:EU:C:2016:42 (CJEU, 21 Jan. 2016).

12. Samir Agrawal v. Competition Commission of India & Ors., Civil Appeal No. 3100 of 2020 (Supreme Court of India, 15 Dec. 2020).

13. United States v. Topkins, No. 3:15-cr-00201-WHO (N.D. Cal. 2015) (plea agreement).

14. Competition and Markets Authority, Online Sales of Discretionary Consumer Products (Trod Ltd and GB eye Ltd), Case 50223, Decision of 12 August 2016 (United Kingdom).

15. Eturas UAB and Others v. Lietuvos Respublikos konkurencijos taryba, Case C-74/14, ECLI:EU:C:2016:42 (CJEU, 21 Jan. 2016).

16. In re RealPage, Inc., Rental Software Antitrust Litigation, MDL No. 3071, Statement of Interest of the United States (M.D. Tenn., filed Nov. 15, 2023).

17. Samir Agrawal v. ANI Technologies Pvt. Ltd. & Ors., Case No. 37 of 2018, CCI Order dated 6 Nov. 2018; affirmed by NCLAT, Competition Appeal (AT) No. 11 of 2019, Order dated 29 May 2020; affirmed by the Supreme Court of India, Civil Appeal No. 3100 of 2020, Judgment dated 15 Dec. 2020.

18. Competition Commission of India, Market Study on Artificial Intelligence and Competition (Oct. 2025).

19. Ministry of Corporate Affairs, Draft Digital Competition Bill, 2024 (India).

20. Competition (Amendment) Act, 2023, s. 5, amending s. 3(4) of the Competition Act, 2002 (India).