I. Background
A widespread use of algorithms has recently raised concerns of possible anti-competitive behaviour as they can make it easier for firms to achieve and sustain collusion without any human interaction. Although there is no doubt that automated algorithms bring great benefits[1] to the society, the prospect of unaccountable computer algorithms fixing prices is attracting much attention from the regulators all over the world.[2] In November 2016, the OECD held a hearing discussion on Big Data to explore the implications on competition authorities' work and whether competition law is the appropriate tool for dealing with issues arising from the use Big Data.
Finding ways to prevent collusion between self-learning algorithms might be one of the biggest challenges that competition law enforcers have ever faced, and whose solution may involve artificially making market conditions more unstable and less prone to tacit collusion.[3]
II. The ACCC’s optimistic approach to colluding algorithms
Last month, Australia’s top competition official (ACCC chair Rod Sims) in a speech delivered to a Sydney conference stated:
“In Australia, we take the view that you cannot avoid liability by saying: ‘my robot did it’. (…) I am confident that our laws, particularly with the addition of the new concerted practices provision under section 45 of the Competition and Consumer Act 2010, but also the new misuse of market power provisions, can deal with either situation if they substantially lessen competition. (…) The new concerted practices prohibition should help shift the focus away from a requirement to establish a ‘meeting of the minds’ to consider whether these has been cooperation between competing businesses that substantially lessens competition. If robots are colluding, this provision will help us stop this conduct.”[4]
An independent competition policy review panel set up by the Australian government (the Harper Review) has recommended repealing these new provisions in favour of expanding Australia’s anti-competitive agreement provisions to cover ‘concerted practices’.[5]
The current law in Australia captures anticompetitive “contracts, arrangements and understandings (CAU)” that has the purpose, effect or likely effect of substantially lessening competition.[6] A CAU has been interpreted as requiring communication, a meeting of minds, and a commitment by at least one party.
The new proposal extends section 45 to ‘prohibit a corporation from engaging in a concerted practice that has the purpose, effect or likely effect of substantially lessening competition’.[7]
It avoids the need to prove CAU and commitment, but is it sufficient to prevent collusion between self-learning algorithms.
III. Agreements, concerted practices and tacit collusion (on steroids)
In the European Union, the concept of agreement is generally broadly defined. Article 101 of the Treaty on the Functioning of the European Union (TFEU) applies to all ‘agreements’ and ‘concerted practices’.[8] According to the courts, the proof of an agreement (…) “must be founded upon the direct or indirect finding of the existence of the subjective element that characterises the very concept of an agreement, that is to say a concurrence of wills between economic operators on the implementation of a policy, the pursuit of an objective, or the adoption of a given line of conduct on the market, irrespective of the manner in which the parties' intention to behave on the market in accordance with the terms of that agreement is expressed”.[9] In the absence of a formal agreement, the category of concerted practices can be applied. According to the Court of Justice of the European Union, the concept of a concerted practice refers to “a form of coordination between undertakings which, without having been taken to the stage where an agreement properly so-called has been concluded, knowingly substitutes for the risks of competition practical cooperation between them”.[10] Tacit collusion or tacit coordination on the other hand occurs when oligopolists coordinate their prices without any explicit communication or contact, which makes it possible for them to behave in a parallel manner and achieve supra-competitive profits. In the Wood pulp case, the Court provided with a very clear distinction between tacit collusion and the notion of concerted practice: If the conduct in question can be justified as independently chosen parallel behaviour as an intelligent adjustment to the anticipated conduct of the competitors, it will not be considered as an infringement of Article 101.[11]
Two types of algorithmic coordination outlined in ‘Virtual Competition’[12] by A. Ezrachi and M. Stucke seem to be particularly problematic: thePredictable Agent and the Digital Eye. In the first scenario, algorithms constantly monitor and adjust to each other’s prices and market data. They are programmed to follow price reductions and price increases when sustainable. The widespread use of such algorithms may result in algorithm-enhanced conscious parallelism.[13] In the second scenario, computers, in learning by experience, determine independently the means to optimize profit. Each firm’s algorithm determines whether it can profit by undertaking a competitive initiative. It is argued that eventually they will work out the oligopolistic pricing game.
IV. Is the optimism justified?
Similarly to an agreement, concerted practice requires at least two parties to a given practice to exist, as it needs to be concerted. In both of the scenarios firms unilaterally adopt their own pricing algorithms to set their own prices. Each firm has an independent economic self-interest in developing and relying on algorithms and it may even be contrary to the firm’s economic self-interest to rely on imperfect human pricing. The use of such algorithms, given their truly unilateral nature, does not seem to amount to a concerted practice.
Although an infringement could be established in most jurisdictions if the use of algorithms is found to have been used to facilitate collusion, in case of pure tacit collusion it would normally not result in any antitrust infringement.
V. Algorithms and oligopoly
In the algorithm-driven markets oligopolistic coordination may be facilitated even in non-oligopolistic markets.[14] Problems with algorithmic pricing are therefore quite similar to issues raised in the debate about whether classic oligopoly behaviour can be prosecuted as an unlawful agreement.
According to one view, an independent, profit-maximising firm can avoid taking steps to form an explicit price fixing agreement with its competitors, but it cannot simply ignore the actions and likely reactions of its rivals when setting its price[15]. Judge Breyer stated this succinctly in Clamp-All Corp. v. Cast Iron Soil Pipe Inst., 851 F.2d 478 (1988): “Courts […] have almost uniformly held […] that […] individual pricing decisions (even when each firm rests its own decision upon its belief that competitors will do the same) do not constitute an unlawful agreement […]. That is not because such pricing is desirable (it is not), but because it is close to impossible to devise a judicially enforceable remedy for "interdependent" pricing. How does one order a firm to set its prices without regard to the likely reactions of its competitors?”.
On the contrary, there are also arguments for re-interpreting or making greater use of standard antitrust tools to address this ‘gap’. Their proponents assume that from the consumer’s perspective, the oligopoly price is just as bad as the cartel price[16] and successful interdependent coordination leading to supra-competitive prices is in contrast with a more economically based approach to competition law. [17]
One of the exemptions to the rule that exempts tacit collusion from competition law liability is the existence of ‘plus factors’, i.e. economic actions and outcomes, above and beyond parallel conduct by oligopolistic firms, that are largely inconsistent with unilateral conduct but largely consistent with explicitly coordinated action.[18] It may be argued that an algorithm, or its design, constitutes a plus factor.[19]
But more general questions should be answered first. Is the current approach to agreements in fact too formalistic and therefore incapable of addressing harmful interdependence among firms? Should the legislators and courts rethink the current policy of exempting tacit collusion from the prohibition against anti-competitive agreements? And can a “meeting of algorithms” amount to an anti-competitive agreement or concerted practice?
The views expressed herein are those of the author and do not represent the views of the Office of Competition and Consumer Protection.
Please scroll down for the footnotes.
***
[1] For example using algorithms by consumers allows them to compare more offers in a more efficient and sophisticated manner; See M. Gal, N. Elkin-Koren, Algorithmic Consumers, Harvard Journal of Law and Technology, 2017.
[2] EC, Margrethe Vestager “Algorithms and Competition” (2017) at: https://ec.europa.eu/commission/commissioners/2014-2019/vestager/announcements/bundeskartellamt-18th-conference-competition-berlin-16-march-2017_en; CMA, Lord David Currie, “The Role of Competition in Stimulating Innovation”, 3 February 2017 at https://www.gov.uk/government/speeches/david-currie-on-the-role-of-competition-in-stimulating-innovation.
[3] Directorate for Financial and Enterprise Affairs, 'Big Data: Bringing Competition Policy to the Digital Era', DAF/Comp(2016)14, 27 October 2016 at https://one.oecd.org/document/DAF/COMP(2016)14/en/pdf.
[4] The whole transcript is available at https://www.accc.gov.au/speech/the-accc%E2%80%99s-approach-to-colluding-robots.
[5] On a side note, the ACCC has also established a new data analytics unit to help crack down on the use of machine learning and artificial intelligence in price-fixing, http://www.innovationaus.com/2017/11/ACCCs-crack-new-data-unit.
[6] Section 45 of the Competition and Consumer Act 2010 (CCA).
[7] https://www.australiancompetitionlaw.org/legislation/2017harper.html.
[8] Article 101 prohibits “all agreements […] and concerted practices […] which have as their object or effect the prevention, restriction or distortion of competition within the internal market.”
[9] Case T-41/96, Bayer AG v Commission, [2000] ECR II-3383, para. 173.
[10] Judgment of the Court of Justice of 16 December 1975 in Joined Cases 40-48/73 Coöperative Vereniging ’Suiker Unie’ UA and others v. Commission, para. 173.
[11] Judgment of the Court of 31 March 1993 in Joined Cases C-89/85 Wood pulp II, para. 71.
[12] A. Ezrachi, M. Stucke, Virtual Competition: The Promise and Perils of the Algorithm-Driven Economy, Harvard University Press 2016; see also A. Ezrachi, M. Stucke, Artificial Intelligence & Collusion: When Computers Inhibit Competition, Oxford Legal Studies Research Paper No. 18/2015, available at SSRN: https://ssrn.com/abstract=2591874.
[13] Or as the authors of Virtual Competition call it: “Tacit Collusion on Steroids”.
[14] M. Gal, Algorithmic-Facilitated Coordination: Market And Legal Solutions, Antitrust Chronicle of May 2017.
[15] “Summary of Discussion of the Hearing on Competition Enforcement in Oligopolistic Markets”, OECD Competition Committee, https://one.oecd.org/document/DAF/COMP/M(2015)1/ANN4/FINAL/en/pdf.
[16] G. Hay, Anti-Competitive Agreements: The Meaning of 'Agreement', Cornell Legal Studies Research Paper No. 13-09 (2013), Available at SSRN: https://ssrn.com/abstract=2220414, p. 15.
[17] L. Kaplow, On the Meaning of Horizontal Agreements in Competition Law, Harvard Law and Economics Discussion Paper No. 691. Available at SSRN: http://ssrn.com/abstract=1873430; L. Kaplow, An Economic Approach to Price Fixing, Harvard Law and Economics Discussion Paper No. 694. Available at SSRN: http://ssrn.com/abstract=1873412; L. Kaplow, ‘Direct Versus Communications-Based Prohibitions on Price Fixing, Harvard Law and Economics Discussion Paper No. 703. Available at SSRN: http://ssrn.com/abstract=1892095.
[18] See, e.g., William E. Kovacic et al., Plus Factors and Agreement in Antitrust, 110 MICH. L. REV. 393, 401 (2011).
[19]M. Gal, Algorithmic-Facilitated Coordination: Market And Legal Solutions, Antitrust Chronicle of May 2017; M. Gal, N. Petit, Algorithms as Plus Factors, forthcoming.