top of page
E. Delzanno; M. Scicolone

PRICING ALGORITHMS AND ANTITRUST ISSUES

In the modern economy, firms increasingly delegate important strategic choices to algorithms of various sorts. For instance, Uber’s former CEO stated: “We are not setting the price, the market is setting the price … We have algorithms to determine what that market is.”[1]

Already in 2015, more than a third of Amazon.com sellers used algorithms to determine their sales price[2]. During the pandemic online purchases increased dramatically and probably, for most of them, the price of the purchased product was neither decided by the seller himself nor by the manufacturer, but by an algorithm instead. This number has certainly grown, given the technological developments that have taken place and the increase in online demand.

However, Antitrust agencies are concerned that the autonomous pricing algorithms increasingly used by online vendors may learn to collude and, therefore, to avoid competitive pressure[3].

As if this wasn’t enough, the 2017 report[4] from the Organization for Economic Cooperation and Development (OECD) expressed concerns that algorithms might enable firms to achieve the same outcomes of traditional hard-core cartels through tacit collusion.


WHAT IS AN ALGORITHM

Any set of rules about turning digital inputs into digital outputs is an "algorithm". In other words, an algorithm is a decision-making tool.[5] In principle, any kind of software consists of one or more algorithms.

The latest generations of algorithms are more “autonomous” than they used to be in the past and adopt the same logic as artificial intelligence (AI). In particular, the algorithm is instructed by the programmer only about the aim of the game (“winning the game or generating the highest profit possible”). It does not know the rules of the game, instead, it is programmed to amend its own decision-making rules to account for experience. In other words, it runs automated field experiments to learn about the demand curve and each product’s profit-maximizing price[6].

On the one hand, these pricing algorithms enable businesses to set their prices more efficiently and effectively, reducing costs and increasing market efficiency. However, the observation of supra-competitive prices is not necessarily a proof of collusion, as high prices could be due to the algorithms’ failure to learn the competitive equilibrium[7]. Collusion between “profit maximiser algorithms” may occur even if they are not designed to conspire, as confirmed by the economic literature[8].

This has raised concerns that such reinforcement-learning algorithms may learn to collude without having been specifically instructed to do so and without communicating with one another. Therefore, many competition authorities have voiced their concerns that pricing algorithms may help firms to avoid competitive pressures and (knowingly or otherwise) coordinate with their competitors.


LEGAL FRAMEWORK

Article 101 TFUE prohibits anti-competitive agreements between two or more independent market operators, decisions of association or concerted practices. From a legal perspective what distinguishes an agreement from concerted practices is the existence of communication between competitors. Indeed, explicit agreements require at least a minimum of communication and a sense of mutual commitment, regardless of the form adopted[9]. On the contrary, concerted practices are “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].

By contrast, the US Supreme Court described tacit collusion as “the process, not in itself unlawful, by which firms in a concentrated market might in effect share monopoly power, setting their prices at a profit-maximizing supra-competitive level by recognizing their shared economic interests and their interdependence with regard to price and output decisions[11]. Indeed, both explicit and tacit collusion may result in a reduction of social welfare using either higher prices or lower output.

However, Article 101 "does not deprive economic operators of the right to adapt themselves intelligently to the existing and anticipated conduct of their competitors"[12]. Therefore, tacit collusion shall not fall within the scope of antitrust rules[13] because price-fixing results from unilateral and rational decisions taken by each of the undertakings active in a market. Indeed, when the market has some specific features (e.g., few players, barriers to entry, homogeneous products, high transparency), the reaction of firms usually consists in matching price undercutting by rivals in order to retain their customers. In this case, monopoly pricing is the only rational course of conduct, as firms lose the incentive to lower prices in the first place[14]. Price reductions would be detected and replicated by competitors before the undertaking that firstly applied rebates can benefit from this choice (e.g., by attracting new customers). Reductions would have the sole effect of reducing the overall earning of the sector, without benefiting the undertaking that first decides to discount[15]. Experience will teach pricing algorithms that the best strategy to perform their task (i.e., maximizing profits) is to avoid any alteration of the status quo. The capability of self-learning algorithms to learn from data and self-adapt with experience will lessen the struggle to compete even if such forms of artificial intelligence have not been instructed to collude.


PRICING ALGORITHM AND CARTELS

The EU defines a cartel as an agreement between a group of similar, independent companies to fix prices, to limit production or development, to share markets or customers between them or other similar types of restriction of competition. The reason cartels are illegal is that they lead to higher prices in combination with lower quality of the products.


i) the creation of cartels

Algorithms continuously and dynamically update prices, basing their decisions on the incessant examination of real-time data on market conditions. According to an inquiry carried out in 2017 by the EU Commission[16], “53% of the respondent retailers track the online prices of competitors, out of which 67% use automatic software programmes for that purpose. Larger companies have a tendency to track online prices of competitors more than smaller ones. The majority of those retailers that use software to track prices subsequently adjust their own prices to those of their competitors (78%)”.

This monitoring process is carried out through specific software called “spiders”, “scrapers” or “crawlers”.

In the simplest scenario, a group of undertakings may simply programme their pricing algorithms to coordinate prices among themselves. However, there can also be more complex scenarios. For example, pricing algorithms may be used by a group of colluding undertakings to monitor market conditions in order to spot the lowest price offered by non-colluding competitors at any given moment. Then, the undertakings may coordinate their behaviours to constantly fix their prices just below the lowest price charged by their non-colluding competitors. To achieve this, one of these companies (the 'sentinel') must program its pricing algorithm to monitor market conditions and to dynamically set its price just below the lowest price charged by non-colluding competitors. The algorithm of the other companies ('followers') must always match the price set by the sentinel[17].

From a theoretical perspective, it is not particularly relevant that algorithms facilitate the material execution of a cartel[18]. To this purpose, the relevant thing is the awareness of the anticompetitive practice, as confirmed by the European Court of Justice[19].


ii) the increase of cartels’ stability

In principle, cartels are unstable since there is a strong individual interest on the part of each of the colluding parties in deviating from the agreed terms by discounting prices to favour its own sales[20]. By doing so, the cheating undertakings have the opportunity to attract customers and to increase their market share at the expense of the other cartelists. However, it can be said that market transparency is indeed one of the most significant conditions which are likely to lead to the successful implementation of a cartel. Indeed, if there is a way to detect deviations from the agreement, the cheating party would not benefit from the price reduction, since others would mirror him. Therefore, the fact that undertakings use pricing algorithms to immediately detect deviations from a cartel should be considered as a rigorous way to implement a cartel.

Consequently, the Commission shall consider the use of similar pricing algorithms as a relevant factor to be assessed for the purpose of setting fines pursuant to art. 23 of the Regulation (EC) n. 1/2003.


REGULATION OF TACIT COLLUSION

These results would appear to suggest that current policy on tacit collusion may no longer be adequate in the age of Artificial Intelligence (AI). It is true that in most countries (including the U.S. and Europe), tacit collusion is not regarded as illegal. In fact, the treatment of tacit collusion has long been one of the most controversial issues in competition policy, and although the prevalent approach is tolerant, some jurisdictions take a more aggressive stance. The rationale is twofold: tacit collusion is very difficult to achieve and hard to detect. These assumptions imply that aggressive antitrust enforcement risks producing many false positives, while tolerant policy would result in relatively few false negatives. However, the advent of AI pricing could well alter the balance between the two types of errors.

It may prove difficult to ascribe to undertakings (let alone to hold them liable for) the autonomous decisions of their algorithms to cooperate among themselves if they were not programmed to collude[21]because the autonomous decision of the algorithm interrupts the causal link between conducts of undertakings and the anticompetitive effect[22]. On this point, it has been argued that “the implementation of pricing policies by one firm’s employees is unilateral conduct and is not actionable […] without evidence establishing an agreement with another firm over the purpose or effect of a pricing algorithm[23] Therefore, the application of art. 101 TFEU is precluded as algorithms have been programmed only to maximize profit do not collude.

However, as mentioned above, the CJUE often dismissed undertakings’ attempts to escape their antitrust liability based on the argument concerning the causal link (or the lack thereof) if they were aware of the anticompetitive harm. Moreover, according to the CJEU, a concerted practice includes every form of coordination among competitors which, regardless of the concurrence of wills between them, has the effect of altering the conditions of the market by replacing competition with cooperation.[24] Therefore, passive modes of participation in an infringement may also be considered as indicative of collusion and, as such, capable of violating art. 101 TFEU. In highly transparent markets, the use of pricing algorithms could arguably be qualified as a conduct falling within the notion of passive modes of participation in an infringement of art. 101 TFEU[25].

Moreover, the CJEU and the NCAs stated that undertaking may be found to be party to a concerted practice simply because it received information on the commercial activities of its competitors and, therefore, private exchanges of information between competitors may be qualified and fined as a cartel under art 101 TFEU [26]. However, also public exchange of information may be “considered as a restriction of competition by object” if it is carried out “with the objective of restricting competition on the market[27]. In this case, what matters is only that the information is made available to the general public, so that competitors may have access to them. The use of an algorithm to signal prices and modify them in view of competitors' responses alters the nature of the undertaking's commercial decision, which is to be regarded as concerted, rather than independent, with the consequence that such conducts "decided by means of algorithms, apparently autonomous, constitute in reality an explicit collusion, falling within the scope of Article 101, as a concerted practice"[28].


CONCLUSION

In conclusion, by virtue of the foregoing, it appears that tacit collusion by means of algorithms capable of monitoring the market and adapting prices accordingly may be brought within the scope of Article 101(1) TFEU as a concerted practice. Even if there is, in fact, a presumption of collusive conduct, there remains the possibility for market participants to overcome it by recourse to Article 101(3) TFEU. In other words, they have to prove that their conduct may generate objective economic benefits that outweigh the negative effects of the restriction of competition.

[1] M. Stoller, How Uber Creates an Algorithmic Monopoly to Extract Rents, Naked Capitalism (April 11, 2014), www.nakedcapitalism.com/2014/04/matt-stoller-how-uber-creates-an-algorithmic-monopoly.html. [2] Chen, L, A Mislove and C Wilson (2016), “An Empirical Analysis of Algorithmic Pricing on Amazon Marketplace”, in Proceedings of the 25th International Conference on World Wide Web, WWW'16, World Wide Web Conferences Steering Committee, pp. 1339-1349 [3] For instance, the seventh session of the FTC Hearings on competition and consumer protection, November 13-14, 2018, centered on the “impact of algorithms and Artificial Intelligence.”, the OECD sponsored a Roundtable on Algorithms and Collusion in June 2017, and in September 2017 the Canadian Competition Bureau released a discussion paper on the ability of algorithms to collude as a major question for antitrust enforcement (Big data and Innovation: Implications for Competition Policy in Canada). Moreover, the British CMA published a white paper on Pricing Algorithms on October 8, 2018. [4] OECD, ALGORITHMS AND COLLUSION: COMPETITION POLICY IN THE DIGITAL AGE 51 (Sept. 14, 2017), http://www.oecd.org/competition/ algorithms-collusion-competition-policy-in-the-digital-age.htm. [5] https://one.oecd.org/document/DAF/COMP/WD(2017)12/en/pdf [6] Karsten Hansen, Kanishka Misra and Mallesh Pai, ALGORITHMIC COLLUSION: SUPRA-COMPETITIVE PRICES VIA INDEPENDENT ALGORITHMS, 2020 [7] Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolò and Sergio Pastorello, ARTIFICIAL INTELLIGENCE, ALGORITHMIC PRICING AND COLLUSION [8] E Calvano, G Calzolari, V Denicolò and S Pastorello, ‘Artificial intelligence, Algorithmic Pricing and Collusion’ (2020) American Economic Review 3267, 3268 [9] Case T-41/96 Bayer, ECLI:EU:T:2000:242, para. 69 [10] Case 40/73 Suiker Unie, ECLI:EU:C:1975:174, paras. 26 and 1 [11] US Su- preme Court judgment of 21st June 1993 Brooke Group Ltd v Brown & Wiliamson Tobacco Corp. [13] It shall be noted that the point is very debated [14] P Siciliani, ‘Tackling Algorithmic-Facilitated Tacit Collusion in a Proportionate Way’ (2019) Journal of European Competition Law & Practice [15] Calzolari, Luca, The Misleading Consequences of Comparing Algorithmic and Tacit Collusion: Tackling Algorithmic Concerted Practices Under Art. 101 TFEU [16] Commission Staff Working Document accompanying the Final Report on the E-commerce Sector Inquiry of 10 May 2017, document SWD(2017) 154 final para. 149 [17] Calzolari, Luca, The Misleading Consequences of Comparing Algorithmic and Tacit Collusion: Tackling Algorithmic Concerted Practices Under Art. 101 TFEU [18] A Ezrachi and ME Stucke, ‘Algorithmic Collusion: Problems and Counter-Measures’ (2017) OECD Roundtable on Algorithms and Collusion [19] C‐74/14 Eturas and Others ECLI:EU:C:2016:42 [20] P Manzini, ‘Algoritmi collusivi e diritto antitrust europeo’ (2019) Mercato Concorrenza Regole [21] DI Ballard and AS Naik, ‘Algorithms, Artificial Intelligence, and Joint Conduct [22] A Capobianco, P Gon- zaga and A Nyeső, ‘Algorithms and Collusion: Competition Policy in the Digital Age’ (2017) OECD Paper for the Roundtable on Algorithms and Collusion [23] OECD, Algorithms and Collusion – Note by the United States (2017 [24] joined cases 40 to 48, 50, 54 to 56, 111, 113 and 114-73 Suiker Unie and Others v Commission ECLI:EU:C:1975:174 para. 191 [25] Calzolari, Luca, The Misleading Consequences of Comparing Algorithmic and Tacit Collusion: Tackling Algorithmic Concerted Practices Under Art. 101 TFEU [26] Cimenteries CBR v Commission ECLI:EU:T:2000:77 para. 1852; P Dole Food and Dole Fresh Fruit Europe v Commission ECLI:EU:C:2015:184; Agenzia Garante della Concor- renza e del Mercato (AGCM) decision of 30 September 2004 n. 13622, I575 Ras-Generali/Iama Consulting [27] Communication from the Commission of 14 January 2011 Guidelines on the applicability of article 101 of the Treaty on the Functioning of the European Union to horizontal cooperation agreements paras 73-74 [28] P Manzini, ‘Algoritmi collusivi e diritto antitrust europeo’ cit. 171-172

87 visualizzazioni0 commenti

Comments


bottom of page