A pricing algorithm is a computer program which monitors and/or sets prices and can be possessed and applied unilaterally by a business entity, can exist an outsourced software or as a simple publicly available price comparison website. Such programs allow to rapidly respond to market changes, adjust prices, optimise business rules and costs [1] and therefore became very popular with both online and block-and-mortar businesses. At the same time, pricing algorithms can distort competition by increasing market transparency, simplifying communication between market players and detection of deviations from collusive agreements and therefore have attracted significant attention of antitrust authorities around the world [2].
The antitrust authorities of EU countries and the European Commission have already investigated numerous cases of pricing collusion by means of algorithms and have developed certain established approaches which were reflected in the latest draft Guidelines on the applicability of Article 101 of the Treaty on the Functioning of the European Union to horizontal co-operation agreements (hereinafter – the “Draft Horizontal Guidelines”).
In this article, we will review the EU case law and the proposed rules regarding the application of algorithms and liability for anti-competitive effects resulting from the exchange of information through such software.
Algorithms facilitating explicit agreements
Antitrust agencies have traditionally been successful in prosecuting cases of overt collusion facilitated by pricing algorithms. It is generally recognised that overt collusion is expressly prohibited as ‘by object’ restriction and collusion facilitated by pricing algorithms is treated like any other concerted actions. Certain exemptions can exist for the monitoring of recommended and maximum resale prices unless such monitoring results in price fixing which shall be considered as prohibited by object.
In the classic Trod Ltd/GB Eye Ltd case in the UK the two merchants using Amazon repricing mechanism acknowledged that they have aligned their prices in the UK. Both merchants were using the repricing algorithm available on Amazon which is to be adjusted by rules determined by each merchant. Such rules include, for example, an x% price decrease for competing goods. The repricer allows to exclude prices of certain merchants from the algorithms by adding such merchants to the ignoring list. The two merchants had agreed between themselves not to compete on prices and put each other on the ignoring list so that they do not undercut each other prices which resulted in the price-fixing cartel.
In 2018 the European Commission fined Asus, Denon & Marantz, Philips and Pioneer for resale price maintenance which was facilitated by the use of price comparison websites (which in their turn are based on pricing algorithms) and a special pricing program which helped producers to trace prices of online retailers, detect deviation and maintain a certain level of retail prices [3].
Pricing algorithms as a means of tacit collusion
In addition to facilitating and maintaining collusive agreements between the entities, pricing algorithms can also lead to tacit price alignment in cases where entities unilaterally choose algorithms which monitor their rivals’ pricing behaviour and react to it in a certain way or in case companies unilaterally use sophisticated self-learning algorithms which being programmed for profit maximising can autonomously collude on higher prices.
It is considered that there are the following scenarios of tacit collusion [4]:
- Hub-and-spoke – where the use of the same pricing algorithm by multiple market players can lead to similar reactions to market developments and, as a result, similar pricing patterns. Moreover, if such market players are aware of algorithms used by their competitors, such use can lead to an indirect exchange of information and coordination of pricing policies.
- Predictable agent - use of simple pricing algorithms which react to market conditions in a certain predictable way, for example, ‘lowest price matching’ can lead to very transparent and predictable pricing. This can result in tacit collusion and parallel pricing [5]
- Autonomous algorithms - where algorithms are sophisticated enough to learn by themselves, they can tacitly coordinate prices.
Unlike cases of overt collusion facilitated by pricing algorithms, considering the absence of evidence of an agreement prosecuting cases of potential autonomous collusion by pricing algorithms will be challenging. EU competition law provides that business entities can be held liable for anti-competitive concerted actions only when there is evidence of an agreement to collude. In its turn, parallel pricing is not itself prohibited by antitrust laws and is considered illegal only if it cannot be explained by any reason other than collusion.
Hub-and-spoke scenarios
While European competition authorities have not challenged predictable agent and autonomous collusion scenarios yet, there are already several cases dealing with the hub-and-spoke situations. On the EU level this scenario was considered by the Court of Justice of the EU (hereinafter – the “CJEU”) in the Eturas case [6]. Eturas, an online travel booking system, sent through its system separate messages to each of its travel agents announcing the default cap of discounts at 3% and subsequently modified the system respectively [7]. The Lithuanian Competition Authority considered that due to such messages sent out by Eturas the travel agencies using the booking system were aware of each other’s discounts since they could reasonably assume that all the other users of that system would also limit their discounts to a maximum of 3% [8].
On the appeal of the decision, the Supreme Administrative Court of Lithuania asked the CJEU to clarify whether entities participating in the same platform can be presumed to be aware of the message sent and, if failing to publicly distance themselves from such message, be liable for anti-competitive concerted actions under Article 101(1) TFEU [9].
The CJEU confirmed that dispatch of the message may justify the presumption that the travel agencies in question were aware of the content of that message; however, the agencies should have an opportunity to rebut such presumption, while the competition authority failed to consider respective evidence [10]. Yet, the CJEU emphasised that the mere awareness of the content of the message is not enough to establish the fact of tacit collusion. Based on the previous case law, the court ruled that to prove a fact of tacit collusion it is also necessary to prove the fact of respective conduct and the cause-effect link between the alleged collusion and the market conduct [11]. Thus, a competition authority or the relevant court shall prove the three elements: (i) awareness of the message; (ii) conduct in correspondence to the agreement; and (iii) the causal link between the message and the market conduct of the competitors. Considering that the system was modified to apply the capped discount by default it is important that the agencies using the platform publicly distance themselves by means of clear and express objection or systematic application of a discount exceeding the cap in question to demonstrate the absence of the second and third conditions [12].
Following the above ruling, the Supreme Administrative Court of Lithuania considered the evidence of awareness and public distancing by the agencies and divided them into 3 groups: (1) the travel agencies that knew about the imposed restriction and did not oppose it; (2) the travel agencies that knew about the message and the system modification and opposed the imposed restriction; and (3) the travel agencies regarding which there was no evidence of their awareness of the restriction [13]. The court found that only the first group of travel agencies shall be held liable for the price fixing since only this group has fulfilled all of the above conditions.
More recently in 2020 the Danish Competition and Consumer Authority entered into a settlement agreement with Ageras A/S, a Danish service intermediation platform in the price standardisation and minimum price-fixing case [14]. Since 2018 Ageras had systematically informed professionals participating in the bids on the platform of a so-called “estimated market price” via a pop-up message when partners submitted bids deemed by Ageras as too low. A professional had a choice: to abstain from the bid, to adjust the price to a higher price than the estimated market price or to submit the original bid [15]. Furthermore, Ageras published several leads, which contained a “minimum quote” for the individual customer. All professionals with access to these leads were able to see the “minimum quote” regardless of their bids or participation in the bidding round. The Danish competition authority also considered that the professionals agreed to the price-fixing arrangement by failing to distance themselves from the practice. However, the decision was limited to Ageras only since it initiated the practices and implemented the algorithm and pop-up prompts [16]. It is not, however, clear how the competition authority would approach the liability of each particular professional using the platform.
Predictable agent and autonomous collusion scenarios
The competition authorities have not challenged such scenarios yet, however, it is generally recognised that such cases are possible, though in absence of any communication or other evidence of collusion it can be difficult to prove the alleged collusion.
Earlier, the German Monopolies Commission noted that in absence of evidence of anti-competitive collusive intent, to hold entities liable for algorithmic price alignment antitrust agencies will have to link commercial decisions of entities to algorithmic prices, which will be challenging in the absence of explicit evidence of collusion [17]. To do so, antitrust agencies will have to investigate longer periods preceding the alleged collusion to track the historical functioning of the algorithm to identify whether the price alignment is a result of self-learning or manual adjustments [18].
German and French antitrust authorities noted that they already have at their disposal means to efficiently investigate price alignment by self-learning autonomous algorithms, namely, information requests, dawn raids and interviews [19]. For example, they can request business entities to provide descriptions of implementing principles, explanation of inputs and outputs, usage patterns of the algorithm, frequency of learning, recalibration or manual adjustments, etc. [20]. Also, an antitrust authority can ask for internal documents such as specifications for the algorithms, user manuals or the code used in the development phase [21]. It is already known that self-learning algorithms must include a code that reveals an intent to collude [22]. The antitrust authorities might investigate the functioning of an algorithm to reveal such code. The authorities can also request a source code of the algorithm to approximate or recreate the algorithm in controlled conditions to understand whether the algorithm was programmed to collude [23]. However, considering that such measures would require significant technological resources and competences, it is expected that they will be applied as a last instance measure.
The approach prosed in the Draft Horizontal Guidelines
Following the number of case investigations as well as sector inquiries by the European Commission and the national competition authorities of the EU states, the European Commission provided some guidance in terms of information exchanges between competitors by means of algorithms in the new Draft Horizontal Guidelines [24]. The Draft Horizontal Guidelines deal mostly with hub-and-spoke scenarios as well as the use of algorithms as means to maintain otherwise reached collusion, while predictable agent and autonomous collusion scenarios remain uncovered. The Draft Horizontal Guidelines recognise that the use of algorithms can increase the risk of collusion outcome, however, algorithmic collusion can happen only under certain circumstances such as: (i) specific design of the algorithms; (ii) high frequency of interactions; (iii) low buyer power; (iv) presence of homogenous products/services [25]. The Draft Horizontal Guidelines do not however provide further detail regarding the “specific design” of an algorithm; however, it can be understood that it would entail the ability to swiftly match rivals’ price adjustments thereby eliminating their incentive to compete on prices as one the conditions for price collusion [26]. It is worth mentioning that scholars also mention other conditions for algorithmic collusion to happen. They include a high level of market concentration and collusion being more profitable than competition on prices which depending on market specifics can result in larger sales volumes rather than values [27].
In the case of hub-and-spoke scenarios, the Draft Horizontal Guidelines note that the liability of a participant on the market shall be defined on a case-by-case basis with a detailed analysis of the role of each participant. A participant in the information exchanges through a hub-and-spoke scheme can be held liable for collusion if such participant was aware of the anti-competitive objective pursued by such exchanges and intended to contribute to such objective by its own conduct [28]. The condition shall be considered as met if a market player expressly or tacitly agrees with the third party that information transferred to such third party will be shared with other market players. This also will apply if the undertaking receiving or sharing the information could reasonably have foreseen that the third party would share its commercial data with other market players and is ready to accept the risk entailed. The same shall apply if the algorithm collects publicly available information (such as prices from websites) which is by itself legal but various competitors have access to the data processed by the tool [29].
Separately, the Draft Horizontal Guidelines provide for conditions to hold third party software providers liable for collusion between the market players. Such entity can be held liable if it intended to contribute by its conduct planned or put into effect by the users of such software or could have reasonably foreseen the anti-competitive nature of the conduct and was prepared to take the risk [30]. Notably, in 2020 the Spanish Competition Authority opened the case proceedings against 7 companies for pricing collusion on the Spanish real estate intermediation market, among which there are two market players and five IT companies that provided the real estate brokerage software and algorithms used to maintain the collusion. The case is still under the investigation and the authority is considering whether the conduct has been facilitated by firms specialized in IT solutions through the design of real estate brokerage software and the algorithms embedded in them [31].
Separate attention is given to cases of unilateral disclosure of commercially sensitive information in particular through algorithms or other software. The Draft Horizontal Guidelines follow the established approach that such unilateral conduct can be caught by Article 101 TFEU even in absence of an explicit agreement to collude on pricing [32]. Yet, a mere announcement or a unilateral communication of prices or other sensitive information shall not be enough. The Draft Horizontal Guidelines refer to the CJEU ruling in the Eturas case, in which, as stated above, the CJEU noted that mere dispatching of the message does not mean that the other alleged participants of the concerted actions were aware of it and acted in conformity with such message [33]. While a competition authority may presume so, business entities shall have an opportunity to rebut such presumption. The Draft Horizontal Guidelines further emphasise that shall a business entity receive commercial sensitive information from a competitor it will be presumed that such entity took into account the received data and adapted its market conduct unless it does not publicly distance itself by means of a clear statement that it does not want to receive such information or by reporting such conduct to the competition authority [34]. The document does not, however, mention the possibility to distance from the received information by consistent conduct not in conformity with the received data, as was noted earlier in Eturas.
Such approach demonstrates the shift of the burden of proof from the competition authority to the defendant. Earlier, the German Monopolies Commission suggested reversing the burden of proof to entities using pricing algorithms, namely, to ensure that if an authority finds indications of anti-competitive algorithmic price alignment in the market, the burden of proof that the use of an algorithm has not contributed to the alignment shall be with the entities using such algorithms [35]. This reflects the general approach towards shifting the burden of proof in digital markets cases. The draft Digital Markets Act introduced by the European Commission in December 2020 provides for a proactive regulation of activities of core platforms thereby is reversing the burden of proof: the practices restricted by the act shall be considered harmful and illegal unless the entity designated as a gatekeeper demonstrates that compliance with an obligation would endanger the economic viability of the operation of the gatekeeper in the EU [36].
Conclusions
The above demonstrates that business entities can be held liable for any anti-competitive effect resulting from the application of algorithms. The entities using algorithms, either their own or supplied by third parties, must make sure that they understand not only the basic purposes and functions of the algorithms to be employed but also how your algorithms are working now and what they may do in the future, which information and how is shared with other market players. It is important to take a risk-based approach towards any information exchange through algorithms and make sure a business publicly distances from any commercially sensitive information received through a tool and which can contribute to a collusive outcome on the market.
*This document is for general guidance only and does not constitute legal advice.
____________________________________
1 Competition and Markets Authority, ‘Pricing Algorithms: Economic working paper on the use of algorithms to facilitate collusion and personalised pricing’ CMA94 8 October 2018 <https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/746353/Algorithms_econ_report.pdf> assessed 22 January 2019, para 2.7.
2 Oxera, ‘When Algorithms Set Prices: Winners and Losers’ (2017), Discussion Paper 19 June 2017, 18, <https://www.oxera.com/wp-content/uploads/2018/07/When-algorithms-set-prices-winners-and-losers.pdf.pdf.> assessed 10 February 2020.
3 See for example Asus (Case No 40465) [2018] <http://ec.europa.eu/competition/antitrust/cases/dec_docs/40465/40465_337_3.pdf> assessed 22 January 2019.
4 CMA (n 1) para 5.15; Ariel Ezrachi and Maurice Stucke, ‘Artificial Intelligence & Collusion: When Computers Inhibit Collusion’ (2015).
5 ibid para 5.22.
6 C-74/14 "Eturas" UAB and Others v Lietuvos Respublikos konkurencijos taryba [2016] ECLI:EU:C:2016:42 < http://curia.europa.eu/juris/liste.jsf?&num=C-74/14> assessed 22 January 2019.
7 ibid, para 10.
8 ibid, para 15.
9 Ibid, para 25.
10 Ibid, 40.
11 Ibid, 42.
12 Ibid, 50.
13 Lithuanian Courts, “SACL has rendered a decision in the travel agencies’ case” available at: <https://www.lvat.lt/en/news/sacl-has-rendered-a-decision-in-the-travel-agencies-case/390>, assessed 28 April 2022.
14 Danish Competition and Consumer Authority, ‘Danish Competition Council: Ageras has infringed competition law’, 30 June 2020, available at: <https://www.en.kfst.dk/nyheder/kfst/english/decisions/20200630-danish-competition-council-ageras-has-infringed-competition-law/>, assessed 28 March 2022.
15 Ingrid Vandenborre and Michael J Frese, ‘Pricing Algorithms under EU Competition Law’, 7 December 2021, Global Competition Review, available at: <https://globalcompetitionreview.com/guide/digital-markets-guide/first-edition/article/pricing-algorithms-under-eu-competition-law#footnote-002>, assessed 27 March 2022.
16 ibid no 15.
17 Monopolkommission, ‘Shaping Competition Policy in the Era of Digitisation’ (2018), para 29 < https://ec.europa.eu/competition/information/digitisation_2018/contributions/monopolkomission.pdf> assessed 10 February 2020.
18 Ibid, para 28.
19 Bundeskartellamt and Autorité de la Concurrence, ‘Algorithms and Competition’ (2019), 69, < https://www.autoritedelaconcurrence.fr/sites/default/files/algorithms-and-competition.pdf>, assessed 10 February 2020.
20 Ibid, 65-66.
21 Ibid, 66.
22 Charley Connor, ‘When robots collude’, 27 September 2019 <https://globalcompetitionreview.com/insight/gcr-q3-2019/1202826/when-robots-collude>, assessed 10 February 2020.
23 Ibid, see also Bundeskartellamt and Autorité de la Concurrence (no 8), 67.
24 Commission, ‘Guidelines on the applicability of Article 101 of the Treaty on the Functioning of the European Union to horizontal co-operation agreements’ (Draft), (2022), available at: <https://ec.europa.eu/competition-policy/public-consultations/2022-hbers_en#view-the-consultation-document>, assessed 15 March 2022.
25 Ibid, 418.
26 Ariel Ezrachi & Maurice E. Stucke, ‘Sustainable and Unchallenged Algorithmic Tacit Collusion’, (2019), Oxford Legal Studies Research Paper No. 16/2019, available at: < https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3282235>, assessed 3 March 2022; pages 9-10.
27 Ibid, 9,11.
28 Commission, no 24, para 437.
29 Ibid.
30 Ibid, para 438.
31 CNMC, ‘The CNMC opens antitrust proceedings against seven firms for suspected price coordination in the real estate intermediation market’, 19 February 2020, Press release, available at: <https://www.cnmc.es/sites/default/files/editor_contenidos/Notas%20de%20prensa/2020/2020219%20NP%20Intermediation%20Market%20EN.pdf >, assessed 5 March 2022.
32 Commission, no 24, para 432.
33 Ibid.
34 Ibid, 433
35 Monopolkommission, no 17, para 239.
36 Commission, Proposal for a Regulation of the European Parliament and of the Council on contestable and fair markets in the digital sector (Digital Markets Act), Brussels, 15.12.2020 COM(2020) 842 final.