Hub-and-Spoke Collusion under EU Competition Law: Liability of Participants in Algorithmic Collusion

(1)

Master’s Thesis

LL.M. European Competition Law and Regulation

Hub-and-Spoke Collusion under EU Competition Law:

Liability of Participants in Algorithmic Collusion

July 24

th

, 2020

Léa Verdoodt

Under the supervision of

12262528

Dr. Jan Broulík

(2)

ABSTRACT

The purpose of this thesis was to assess how the current framework of EU competition law could address the question of liability of the competitors and IT developers involved in a hub-and-spoke scheme having resulted in collusion on prices.

The research was based on papers of scholars having discussed hub-and-spoke specifically and other related topics. Further, papers of the OECD and contributing countries have enabled to further understand the challenges related to liability in hub-and-spokes scenarios. Moreover, in light of the prohibitions of Article 101(1) TFEU and relevant case-law, this thesis could bring forward interesting answers to the question.

Therefore, this thesis has first clarified the notion of pricing algorithms and establish the context of law as an introduction to the core of the discussion. After having defined algorithmic hub-and-spoke collusion, the thesis delves into the analysis of the current EU competition case-law to address the question. From this analysis, it seems the EU will be able to interpret the case-law within the context of an intentional use of a common algorithm. Applying the rulings of these cases to the scenarios where pricing algorithms were unintentionally used by several competitors was not adapted. Hence, this thesis brings forward potential ex ante policies which, if enacted, would greatly reduce the risks of tacit collusion to occur.

(3)

ABBREVIATIONS

AI Artificial Intelligence

CJEU Court of Justice of the European Union

ECJ European Court of Justice

EU European Union

(5)

INTRODUCTION

Algorithms have been around for a long time. One of the first algorithms was developed in ancient Greece at the emergence of mathematics.1 In the digital era, algorithms are used in many

more fields, going from finances to health or the automotive sector.2 Lately, companies active on

online markets and high-tech industries have been increasingly relying on computer algorithms.3

This coincides with the evolution of Big Data enabling companies to access, process and analyze enormous volumes of data at strong speed. 4 Used to optimize many aspects of a business’

process, this thesis will only discuss pricing algorithms, i.e. algorithms to fix, monitor and recommend prices. With the 2016 e-commerce sector inquiry, the Commission observed that many retailers track their competitor’s online prices – mostly with the use of automatic software programs, i.e. algorithms – and eventually adjust their prices depending on the results obtained.5

Lately, pricing algorithms have been the center of discussion in the European Union.6 A few

competition authorities and organizations have assessed the impact of pricing algorithms on competition.7 Essentially, the concerns refer to the use of pricing algorithms as a tool to facilitate

collusion.8 Already in 2015, Ezrachi and Stucke identified four scenarios in which the use of

pricing algorithms may foster collusion.9 This thesis will discuss the scenario known as

hub-and-spoke collusion. In ‘Virtual Competition’, Ezrachi and Stucke describe the classic hub-and-hub-and-spoke collusion as an array of similar vertical agreements resulting in horizontal collusion between

1 Marcus du Sautoy, The Creativity Code: Art and Innovation in the Age of AI (1st edn, HUP 2019), 40 2 Lorenz Marx, Christian Ritz and Jonas Weller, ‘Liability for Outsourced Algorithmic Collusion – A Practical

Approximation’ (2019) 2 Concurrences, para. 7 <https://www.concurrences.com/fr/revue/issues/no-2-2019/pratiques/liability-for-outsourced-algorithmic-collusion-a-practical-approximation>

3 OECD, ‘Algorithms and Collusion: Competition Policy in the Digital Age’ (2017), 7 and 11

4 Ariel Ezrachi and Maurice E. Stucke, ‘Artificial Intelligence & Collusion: When Computers Inhibit Competition’

(2015) Oxford Legal Studies Research Paper No. 18/2015

<https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2591874>; Simone Gambuto, ‘Algorithms, Big Data and Tacit Collusion New Challenges for Competition Law’ in Enrico Adriano Raffaeli (ed), Antitrust between EU Law and national law/Antitrust fra diritto nazionalee diritto dell'unione europea: XII conference/XIII convegno (Bruylant 2019), 253-255

5 Commission, ‘Commission Staff Working Document Accompanying the document Final Report on the

E-commerce Sector Inquiry’ SWD (2017) 154 final, para. 149

6 Hogan Lovell, ‘Digital competition policy on the move : price algorithms in the German Monopolies

Commission’s spotlight – EU Commission launches public consultation process’ (Summer 2018) Antitrust, Competition and Economic Regulation Quarterly

<https://www.hoganlovells.com/~/media/hogan-lovells/pdf/2018/2018_09_20_acer_newsletter_june_2018_v11.pdf>

7 E.g. the French Autorité de la Concurrence and the German Federal Cartel Office have jointly published a report

on algorithms and collusion in November 2019 and the OECD has published its research and called for contribution on the same topic.

8 OECD, ‘Algorithms and Collusion’ (n 3). ; Autorité de la Concurrence and Bundeskartellamt, ‘Algorithms and

Competition’ (2019) <https://www.autoritedelaconcurrence.fr/sites/default/files/algorithms-and-competition.pdf>

(6)

competitors.10 The vertical agreements are concluded between the hub as the common supplier

or retailer, and the spokes – individually of its competitors active on the same market.11 In the

context of algorithms, the function of the “hub” is executed by the automated pricing algorithm developed by a third-party. This algorithm is itself implemented by several competitors who use it as a tool to fix its prices based on the decisions and strategies it developed. This leads to a triangular relationship between the hub and the spokes where competitors align their prices following a decision brought forward by the algorithm. Alignment on prices results from the decision-making process of the algorithm; which processed the data of all these different competitors to define the profit-maximizing prices they each should set.12

Many legal challenges have been associated with hub-and-spoke collusion, but this thesis will focus on the question of liability. Although the prices have been established by an algorithm, the question arises in what extent should humans be liable for the decision of an algorithm which resulted in collusion on prices? In this regard, Commissioner for Competition Margret Vestager declared in a speech in 2017: “It's true that the idea of automated systems getting together and

reaching a meeting of minds is still science fiction.

But illegal collusion isn't always put together in back rooms. There are many ways that collusion can happen, and some of them are well within the capacity of automated systems.

So as competition enforcers, I think we need to make it very clear that companies can’t escape responsibility for collusion by hiding behind a computer program.”13

In the context of liability for hub-and-spoke collusion, a distinction is made between competitors knowingly using the same algorithm and competitors merely engaging in the parallel use of an algorithm. In the former, liability should not be difficult to concede. Indeed, competitors coordinated their use of the same algorithm either by entering into an agreement or through indirect contact.14 The third-party’s algorithm is then used as a tool to exchange sensitive

10 Ariel Ezrachi and Maurice E. Stucke, Virtual Competition: The Promise and Perils of the Algorithm-Driven

Economy (1st edn, HUP 2016), 46

11 OECD, ‘Hub-and-Spoke Arrangements – Background Note by the Secretariat’ (DAF/COMP(2019)14), 2 12 Ezrachi and Stucke, Virtual Competition (n 10) 47.

13 Margrethe Vestager, Speech at the Bundeskartellamt 18th Conference on Competition in Berlin (16 March

2017), available at <https://ec.europa.eu/competition/speeches/index_theme_1.html>

(7)

information and enable a more effective collusion.15 In the latter situation, collusion derives from

pure parallel behavior where competitors unwittingly used a common algorithm. Competitors did not intend to coordinate their behavior as they were not aware that their competitors relied on the same algorithm.16 In both cases, the use of the same algorithm by competitors results in

collusive outcomes17, but the extent of liability of the parties involved may be difficult to

establish.

This thesis attempts to determine whether the current European competition law framework is sufficient to address the challenging assessment of liability of the competitors and IT developers involved in an algorithmic hub-and-poke collusion scheme. Although pricing algorithms are used by many actors on the economic market, the focus of this thesis is narrowed down to the so-called outsourced pricing algorithms. This corresponds with the aim of the thesis which is to define liability of the competitors having used a common algorithm from a single IT developer. The scenarios discussed in this thesis differ from situations where a pricing algorithm is used by a single platform and imposes prices to several of its suppliers active on the platform.18

Establishing liability in the latter situation diverges from liability in the case of pricing algorithms sold to companies which are usually independent from each other. This thesis will thus analyze whether the current competition law framework can address liability of the hub and the spokes in the cases where the competitors involved used a pricing algorithm outsourced by an IT developer – i.e. a pricing algorithm sold on the market to any company that desires to use this product.

Therefore, the thesis has been divided in several chapters. The first two chapters are set to help the reader understand the basic notions of algorithms and provide the context of EU competition law that is discussed in this thesis. The first chapter aims to clarify the meaning of pricing algorithms and related terms that will be mentioned several times in the thesis. The second chapter attempts to shed some light on the concept of collusion and clarifies the distinction between explicit and tacit collusion. Additionally, it describes the concept of algorithmic

15 OECD, ‘Algorithms and Collusion – Note from the European Union’ (DAF/COMP/WD(2017)12), paras. 21 and

29

16 Autorité de la Concurrence and Bundeskartellamt, ‘Algorithms and Competition’ (2019), 32-42

<https://www.autoritedelaconcurrence.fr/sites/default/files/algorithms-and-competition.pdf

17 OECD, ‘Algorithms and Collusion’ (n 3) 28.

18 An example would be the Uber ride sharing platform. It’s the app’s pricing algorithm that determines the prices

of a ride and imposes that price on all Uber drivers and all customers, leaving no space of decision for the drivers themselves. Retrieved from: Julian Nowag, “Algorithmic Price-Setting by Platforms”

(8)

collusion. The third chapter introduces the concept of hub-and-spoke collusion. It further explains how alignment occurs under the hub-and-spoke scenario and identifies the challenges associated with this collusion scenario. The fourth chapter of the thesis delves into the discussion on liability of the hub and the spokes. The analysis will involve the concepts and notions explained in the previous chapters. This chapter attempts to determine how liability can be established in the case of intentional common use of a single algorithm considering relevant case-law. It further brings forward potential answers to the question on liability in the case of inadvertent common use of an algorithm. Finally, the conclusion will summarize the findings of the analysis of this thesis and answer to the question whether the current framework of EU competition law is sufficient to assess liability of the participants to a hub-and-spoke collusion scheme.

(9)

I. PRICING ALGORITHMS

This chapter provides an explanation of the term “algorithm” and the concepts related to it. Specifically, the first section presents a general definition of algorithms in the digital era and their association with Big Data. The following section specifies the concept of pricing algorithms. The last section introduces the difference between internally developed or outsourced algorithms.

1. DEFINITION OF ALGORITHMS

Algorithms are described in several ways and a single universally accepted definition of algorithms is still missing.19 The Cambridge Dictionary defines algorithms as “a set of

mathematical instructions or rules that, especially if given to a computer, will help to calculate an answer to a problem.”20 Another definition of algorithms proposed in the literature21 and

adopted by the OECD22 describes an algorithm as “an ambiguous, precise list of simple

operations applied mechanically and systematically to a set of tokens or objects. The initial state of the token is the input; the final state is the output.”23 This definition can be illustrated with the

example of a cake recipe. The recipe is the list of operations and the ingredients are the set of tokens. The cake that comes out of the oven is the final output.24

a. Algorithms in the Digital World

In the digital era, algorithms have been developed to enable computers to perform complex calculations and data processing which are too intricate for humans.

Accordingly, algorithms limited to computer science can be described as follows: “any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values as output.”25

20 Algorithm (n.c.) in Cambridge Dictionary. Retrieved from

<https://dictionary.cambridge.org/dictionary/english/algorithm?q=algorithms>

21 Robert A. Wilson (Professor of Philosophy) and Frank C. Keil (Professor of Psychology) presented their

definition of an algorithm in their book on the history and philosophy of cognitive science “The MIT Encyclopedia of Cognitive Sciences” (1999)

23 Robert A. Wilson and Frank C. Keil, The MIT Encyclopedia of the Cognitive Sciences (MITECS) (The MIT

Press 1999), 11

24 ibid. 11; OECD, ‘Algorithms and Collusion’ (n 3) 8.

(10)

For the purpose of this thesis, we will only discuss these digital algorithms. Precisely, pricing algorithms used by businesses.

b. Big Data and Algorithms

The input value operated by digital algorithms is data. Any relevant piece of information can be used as data and processed by the algorithm.26 The notion of Big Data is attributed to the

explosion of information of recent years.

Three components can be attributed to Big Data.27 First, the growing volume – or amount – of

available data. The velocity – or speed – at which data can be collected, processed and analyzed. Finally, the variety of the available data that has considerably increased in recent years. 28

Volume, velocity and variety have been recognized to be the three elements characterizing Big Data.29 Knowing the improved availability of data30, it is argued that it is the evolution of Big

Data that is at the origin of the development of algorithms in recent years.31 In other words,

pricing algorithms would not be relied on by more and more businesses if Big Data had not evolved so much.

2. PRICING ALGORITHMS

Pricing algorithms are described as algorithms that use a computational procedure to determine a price as an output based on an array of inputs relevant to price information.32 This thesis will

only discuss such algorithms which are characterized as algorithms implementing continuous

26 Competition & Markets Authority, ‘Pricing Algorithms: Economic working paper on the use of algorithms to

facilitate collusion and personalised pricing’ (2018), paras. 2.23 and 2.24

<https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/746353/Algorit hms_econ_report.pdf>

27 Doug Laney, ‘3D Data Management: Controlling Data Volume, Velocity, and Variety’ (2001) META Group

<https://blogs.gartner.com/doug-laney/files/2012/01/ad949-3D-Data-Management-Controlling-Data-Volume-Velocity-and-Variety.pdf>

28 Competition & Markets Authority (n 26) para. 2.21.

29 The characterization of Big Data by the « 3 V » has been used by many scholars such as M. Stucke or A. Gal,

along with the OECD in their papers and the national competition authorities.

30 Avigdor Gal, ‘It’s a Feature, not a Bug: On Learning Algorithms and what they can teach us’ (2017) OECD

Roundtable on Algorithms and Collusion <https://one.oecd.org/document/DAF/COMP/WD(2017)50/en/pdf>

31 ibid. para. 1; OECD, ‘Big Data: Bringing Competition Policy to the Digital Era – Background Note by the

Secretariat’ (DAF/COMP(2016)4), para. 14

(11)

price changes.33 The definition of a pricing algorithm includes several purposes: price

monitoring, price recommendation and price-setting.34 When the term “pricing algorithm” will

be mentioned in this thesis, it will be referring to the broad definition of the concept including the three purposes.

Price monitoring implies that the algorithm will collect information and data on the competitor’s prices. Companies will thus systematically and almost instantly adjust their prices whenever a change is detected.35 Price recommendation guides on the adjustment of prices according to

demand and supply.36 Lastly, price-fixing algorithms calculate at what price a product should be

sold to maximize the companies’ profits.37 The price is calculated based on the businesses own

data, the competitors’ data and the consumers’ demands.38

The input of a pricing algorithm includes information on the company’s own prices, previous pricing, profit, costs of production and storage as well as information on customers. Moreover, the company’s competitors’ information on similar matters will be analyzed as well. Ultimately, information unrelated to the economic market – such as the weather – will also be considered. 39

a. Examples of Pricing Algorithms

Pricing algorithms are mostly used in e-commerce. Indeed, they are inherently more efficient in the context of e-commerce where data can be collected and processed easily and prices can be adjusted automatically almost instantly. Whereas on the offline market, data is not as accessible and prices cannot be adjusted as rapidly as with a computerized system.40 The e-commerce sector

inquiry of the European Commission supports this statement by providing figures on the use of pricing algorithms by online retailers. The Commission found out that more than half of the responding retailers track their competitors’ prices. Moreover, almost 70% of these responding retailers use automatic software programs – i.e. pricing algorithms – of which around 80%

33 OECD, ‘Algorithms and Collusion’ (n 3) 16. 34 Competition & Markets Authority (n 26) para. 2.5. 35 OECD, ‘Algorithms and Collusion’ (n 3) 26.

36 OECD, ‘Personalized Pricing in the Digital Era – Background Note by the Secretariat’ (DAF/COMP(2018)13),

para. 19; OECD, ‘Algorithms and Collusion’ (n 3) 27.

37 Competition & Markets Authority (n 26) paras. 2.8 and 2.10. 38 OECD, ‘Algorithms and Collusion’ (n 3) 16.

40 Le Chen, Alan Mislove and Christo Wilson, ‘An Empirical Analysis of Algorithmic Pricing on Amazon

Marketplace’ (2016) Proceedings of the 25th International Conference on World Wide Web (WWW ’16), 1339

(12)

subsequently adapt their prices.41. Nowadays, pricing algorithms are also used by many other

industries such as airlines and hotel-booking websites.42

A great example on the functioning of pricing algorithms is the astonishing price at which the book “Making of a Fly” was offered on Amazon in April 2011. Two new versions of the book were offered by two distinct sellers on the Amazon platform. One version was offered at the price of $1 730 045,91 and the other at $2 198 177,95. It appeared both sellers were using a pricing algorithm. This resulted in a price spiral, originally starting at a commercial price, to rise to a million dollars’ price in only ten days. Eventually, the books were sold at a normal commercial price of approximately $100.43

3. OUTSOURCED OR VERTICALLY INTEGRATED PRICING ALGORITHMS

In practice, two types of pricing algorithms are used by businesses. On the one hand, a company may be using a “vertically integrated” algorithm.44 This is an algorithm developed by the business

itself to set the prices for its own products. Generally, algorithms are internally developed by large businesses with the resources and expertise to do so. 45 On the other hand, a company may

buy an algorithm developed by a specialized IT companies.46 These are identified as

“outsourced” algorithms and are sold “off the shelf” by the algorithms developers.47

In this thesis, the discussion will focus on the outsourced algorithms. Indeed, the problematic discussed in this thesis relates to these outsourced algorithms bought by competitors from a single IT developer which then results in the forming of a hub-and-spoke scheme.

41 Commission, ‘Staff Working Document on the E-commerce Sector Inquiry’ (n 5) para. 149. 42 Joe Weinman, ‘Cloud Pricing and Markets’ (2015) IEEE Cloud Computing, 1

<https://ieeexplore.ieee.org/document/7091800>

43 Michael Eisen, 'Amazon’s $23,698,65593 book about flies' (It is NOT junk, 22 April

2011) <http://www.michaeleisen.org/blog/?p=358>; Lea Bernhardt and Ralf Dewenter, ‘Collusion by code or algorithmic collusion? When pricing algorithms take over’ (2020) European Competition Journal, 1

<https://www.tandfonline.com/doi/full/10.1080/17441056.2020.1733344>; Ezrachi and Stucke, ‘Artificial Intelligence & Collusion’ (n 4) 3.

44 Marx, Ritz and Weller (n 2).

46 Monopolkommission, ‘Algorithms and Collusion’ (2018) XXII. Biennal Report of the Monopolies Commission,

para. 252

<https://www.monopolkommission.de/images/HG22/Main_Report_XXII_Algorithms_and_Collusion.pdf>

(13)

II. ALGORITHMIC COLLUSION

Since this thesis discusses a new form of collusion – namely hub-and-spoke collusion – it is important to first recall how EU competition law defines collusion. Accordingly, the first section of this chapter lays down the definitions of collusion and clarifies the distinction between explicit or tacit collusion within the framework of Article 101(1) TFEU. The second section explains how algorithms facilitate or help sustain collusion. It further demonstrates how algorithms alter characteristics of the economic market.

1. INTRODUCTION TO COLLUSION

Collusion refers to any form of combination, conspiracy or agreement between competing firms in which competitors attain higher profits than on a non-cooperative market.48 Collusion is a joint

profit maximization strategy for firms to raise or fix prices and to reduce output to increase profits.49 To attain a sustainable collusive scheme, businesses must put in place a structure and

agree on a common policy which is then monitored and of which any deviations can be punished.50

Collusion occurs in two forms. It is either explicit or tacit collusion. For explicit collusion, competitors maintain their anticompetitive coordinating conduct with explicit agreements – whether written or oral. Adversely, for tacit collusion competitors achieve coordination without any need for an explicit agreement. Indeed, competitors are independent from each other and each participant unilaterally decides of its own profit-maximizing strategy. These decisions are, however, based on the strategies implemented by their competitors.51 Such conduct usually

appears in transparent and concentrated markets with homogeneous products, making it easy for competitors to enjoy the benefits of such market structure.52

The distinction between explicit or tacit collusion is significant for the discussion on liability of the participants of algorithmic hub-and-spoke collusion, as will be demonstrated in Chapter IV of this thesis.

48 R.S. Khemanu and D.M. Shapiro commissioned by the OECD, Glossary of Industrial Organisation Economics

and Competition, (OECD 1993), para. 25

49 ibid. para. 25; OECD, ‘Algorithms and Collusion’ (n 3) 19. 50 OECD, ‘Algorithms and Collusion’ (n 3) 19.

51 ibid. 19.

(14)

2. COLLUSION UNDER EUROPEAN COMPETITION LAW

Under EU competition law, the focus lies on the means used by competitors to achieve a collusive outcome. Indeed, all agreements, decisions by undertakings or concerted practices which have as their object or effect the prevention, restriction or distortion of competition are restricted under Article 101(1) TFEU.53

The definition of explicit collusion includes the notion of an agreement. Hence, where there is an agreement between competitors resulting in collusion, the infringement of Article 101(1) can be established.54 The term “agreement” has been broadly defined by the CJEU as “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 agreements is expressed.”55 In this regard, explicit collusion arising from an agreement falls

under Article 101(1) TFEU as long as there is proof of a “meeting of the minds” and irrespective of the form of the agreement.56

Moreover, explicit collusion can also result from a concerted practice between competitors.57

The CJEU defines a concerted practice as “a form of coordination between undertakings which, without having reached the stage where an agreement properly so-called has been concluded, knowingly substitutes practical cooperation between them for the risks of competition.”58 The

CJEU further clarified that concerted practice amounts to any direct or indirect contact between competitors with the object or effect to either influence the conduct of an actual or potential competitor or to disclose their company’s strategy.59 However, the CJEU recognized the right of

competitors to independently adapt their pricing strategies to any current or potential conduct of its competitors.60 Hence, when direct or indirect contact between competitors has been

53 OECD, ‘Algorithms and Collusion’ (n 3) 19. 54 ibid. 19.

55 Case T-41/96, Bayer AG v Commission of the European Communities [2000] ECLI:EU:T:2000:242, para 69 56 OECD, ‘Algorithms and Collusion’ (n 3) 19.

57 Autorité de la Concurrence and Bundeskartellamt (n 8) 26.

58 Case 48/69, Imperial Chemical Industries Ltd. v Commission of the European Communities (Dyestuff) [1972]

ECLI:EU:C:1972:70, para. 64; Joined cases C-89/85, C-104/85, C-114/85, C-116/85, C-117/85 and C-125/85 to C-129/85, A. Ahlström Osakeyhtiö and others v Commission (Woodpulp) [1993] ECLI:EU:C:1993:120, para. 63

59 Case 40/73, Suiker Unie and Others v Commission [1975] ECLI:EU:C:1975:174, para. 174 60 Dyestuff (n 58) para. 118.; Suiker Unie and Others (n 59) para. 174.

(15)

established in the form of a concerted action, which then leads to market conduct implementing the concerted action then the causal link between the first two elements is presumed to exist thus confirming the existence of concerted practice within the meaning of Article 101(1) TFEU.61

For competition enforcers, the real challenge is the concept of tacit collusion. The elements characterizing an “agreement” or a “concerted practice” are missing since firms implement pricing strategies independently from each other. As mentioned above, tacit collusion occurs within a particular market structure. From an economic perception, on a transparent market with few market players and a homogeneous product, a supra-competitive pricing strategy may be seen as a normal and rational behavior.62 In the absence of and understanding or communication

between companies, tacit collusion is described as conscious parallelism without any coordination whatsoever. It appears when certain market characteristics are in place which allows companies to undertake similar strategies. However, the margin between conscious parallelism and concerted practices is slight.63 With tacit collusion there is no contact whatsoever between

competitors adapting their prices. Parallel conduct does not automatically encompass concertation or coordination, whilst a concerted practice that affects companies’ independence consistently falls under Article 101 TFEU.64 A grey area of business conduct thus emerges where

competitors are able to coordinate their behaviors – by virtue of the market’s characteristics – without it being identified as concerted practice simply because there was no proof of communication between the competitors involved.65 The increasing use of pricing algorithms

constitutes such grey area and is further explained in the following section.

3. ALGORITHMS FACILITATING COLLUSION

Algorithms are said to pose new risks to the proper functioning of the competitive economic market in the European Union. Indeed, the evolution of new technologies and Big Data in

61 OECD, ‘Hub-and-Spoke Arrangements – Note by Germany’ (DAF/COMP/WD(2019)104), para. 14.; Gian Luca

Zampa and Paolo Buccirossi, ‘Hub and Spoke Practices: Law and Economics of the New Antitrust Frontier’ (2013) 9/1 Competition Law International, 100

<https://heinonline.org/HOL/Page?collection=journals&handle=hein.journals/cmpetion9&id=2&men_tab=srchres ults>

62 This is also known as the « oligopoly problem » on markets where only a few undertakings jointly dominate.

(Retrieved from R. Whish and D. Bailey, Competition Law (9th edn., OUP 2018) 572) 63 OECD, ‘Algorithms and Collusion’ (n 3) 19-20.

64 Pierre Honoré and Guillaume Fabré, ‘European Union – Algorithmic Pricing under Article 101 TFEU’ (2019)

Global Competition Review <https://www.lexology.com/library/detail.aspx?g=6f23d01a-150d-400a-b594-6d5542054025>

(16)

conjunction with algorithms have resulted in new and unique anticompetitive activities. These scenarios have either not yet been identified or are outside the scope of application of EU competition law. Accordingly, it seems the EU competition law framework is not yet equipped to address the threats formed by algorithms and this involves implications on the competitive process and competition policy.66

a. New anticompetitive outcomes

The use of pricing algorithms creates new dynamics widening the circumstances in which anticompetitive outcomes may take place.67 It allows companies to hide behind an algorithm

helping them initiate, implement or consolidate a cartel.68 The algorithm is then used as a tool to

monitor and enforce a pre-existing agreement to coordinate on prices. Alternatively, companies may be using an identical pricing algorithm in the context of a concerted practice between competitors. This does not preclude the inadvertent use of a single pricing algorithms by several companies thus unknowingly engaging in parallel conduct.69 As mentioned in the previous

section, it is the inadvertent parallel behavior leading to tacit collusion which presents challenges beyond the reach of European competition law. Hence, the danger does not reside in the use of algorithms limited to enhance anticompetitive agreements or concerted practice. This would amount to explicit algorithmic collusion resulting from a pre-existing agreement or concerted practice and would be captured by Article 101(1) TFEU.70 Rather, algorithms enabling tacit

collusion to occur in a subtler fashion and escaping the scope of competition law troubles scholars and competition authorities.71 The following sub-section presents the influence of algorithms on

the markets. More precisely, it explains how algorithms enable tacit collusion outcomes to appear on any market.

66 Marx, Ritz and Weller (n 2) para. 4.; Ariel Ezrachi, ‘EU Competition Law Goals and the Digital Economy’

(2018) Oxford Legal Studies Research Paper No. 17/2018, 5 <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3191766>

67 Ezrachi and Stucke, Virtual Competition (n 10) 36.

68 Dan Vlad Roman, ‘Digital markets and pricing algorithms – a dynamic approach towards horizontal

competition’ (2018) Chinese Center for Competition Law <http://www.competitionlaw.cn/info/1007/26679.htm>

70 OECD, ‘Hub-and-Spoke Arrangements – Note by Germany’ (n 61) para. 45.

71 Ezrachi and Stucke, Virtual Competition (n 10) 36.; This assumption is based on the fact that many scholars,

competition authorities have published papers and researches on the topic of algorithmic tacit collusion. Accordingly, this thesis refers to many of these papers.

(17)

b. Algorithms Altering Market Characteristics

As noted in the second section of this chapter, tacit collusion is more likely to occur in markets with particular characteristics. Evidently, these characteristics are still present on markets where the use of algorithms is frequent. First, algorithmic tacit collusion is likely to arise in concentrated markets involving a similar product and where companies have shifted to online pricing. Secondly, on the online market, deviation by a competitor is detected faster with the use of pricing algorithms. A pricing algorithm may then quickly calculate the retaliation measures needed to punish deviation. Companies can also adjust prices as many time as they desire every day and almost immediately when deviations are detected. Finally, to enable sustainable tacit collusion, a market should not be endangered by buyer power or the entry of competitors on the market.72 However, it appears that pricing algorithms have altered the market characteristics

typically associated with tacit collusion. Indeed, the new technology provided by algorithms and the increased transparency and the frequency of interaction allows companies to collude on price on markets where the three characteristics mentioned above are per se not present.73 An increased

market transparency is linked to the incentive of companies in gaining competitive advantage through the use of algorithms and therefore collecting real-time data of its competitors. Likewise, an online environment with great transparency on prices results in algorithms being even more reliable and efficient tools capable of reporting or adjusting prices instantly. Finally, pricing algorithms nowadays are way more efficient than humans at calculating the profit maximizing prices.74 This further triggers companies to rely on these tools which reinforces the altering of

the competitive market and enables collusion to exist on any market even in the absence of typical characteristics previously associated with tacit collusion.

c. Conclusion

Regardless of the scenario at hand, whether it is explicit or tacit algorithmic collusion, both types are a threat to the economic market. Collusion leads to supra-competitive prices between competitors further resulting in a lower output of goods, deadweight losses and ultimately a

72 Ariel Ezrachi and Maurice E. Stucke, ‘Two Artificial Neural Networks Meet in an Online Hub and Change the

Future (of Competition, Market Dynamics and Society)’ (2017) University of Tennessee Legal Studies Research Paper, 3-5 <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2949434>

73 Ariel Ezrachi and Maurice E. Stucke, ‘Sustainable and Unchallenged Algorithmic Tacit Collusion’ (2018)

University of Tennessee Legal Studies Research Paper, 2-5

<https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3282235>; Monopolkomission, ‘Shaping competition policy’ Shaping competition policy in the era of digitalisation’ (2018) Submission for the European Commission – DG Competition Call for contributions, para. 13-15.

(18)

decrease of consumer welfare.75 Moreover, as demonstrated in the above paragraph, since the

traditional market characteristics leading to tacit collusion are transformed, there is indeed a real threat linked to algorithms.

75 Peter Georg Picht and Gaspare Tazio Loderer, ‘Framing Algorithms: Competition Law and (Other) Regulatory

Tools’ Max Planck Institute for Innovation & Competition Research Paper No. 18-24, 16. <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3275198>

(19)

III. HUB-AND-SPOKE COLLUSION

This chapter introduces the concept of algorithmic hub-and-spoke collusion. Although not defined under European competition law, the OECD and some scholars have already brought forward potential descriptions of this new type of collusion. 76 Hub-and-spoke collusion was first

developed in the context of offline collusion.77 In this regard, the chapter will first clarify the

“classic” scenario of spoke collusion. The second section will then introduce hub-and-spoke collusion arising from the use of pricing algorithms. Moreover, the section will clarify how the use of a similar pricing algorithms leads to companies colluding on prices. Additionally, it will present the legal challenges associated with algorithm-driven hub-and-spoke collusion.

1. DEFINITION OF “CLASSIC”HUB-AND-SPOKE COLLUSION

There is no definition of hub-and-spoke collusion under European competition law. Nonetheless, hub-and-spoke collusion has been characterized as a cartel coordinated through indirect exchanges of information via a third-party.78 In other words, the hub-and-spoke scenario occurs

where a cluster of similar agreements between a hub and each of the spokes results in horizontal collusion on the downstream market – i.e. on the market of the spokes. The hub is identified as the common third-party active on a different level of the supply chain or on another market. The spokes are the industry’s competitors each individually concluding an identical agreement with the hub. Hub-and-spoke collusion is thus a triangular dynamic involving two distinct features: it first requires an array of vertical agreements to then result in horizontal collusion. 79

The following section explains hub-and-spoke collusion emerging when competitors rely on a similar pricing algorithm.

76 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) 2.; Nicolas Sahuguet and Alexis

Walckiers, ‘A Theory of Hub-and-Spoke Collusion’ (2017) 53 International Journal of Industrial, Organization, 354-355 <https://www.sciencedirect.com/science/article/pii/S0167718716300285?via%3Dihub>; Ezrachi and Stucke, Virtual Competition (n 10) 46-50.; Patrick J. G. Van Cayseele, ‘Hub-and-Spoke Collusion: Some Nagging Questions by Economists’ (2014) 5/3 Journal of European Competition Law & Practice, 164

<https://academic.oup.com/jeclap/article/5/3/164/1825839>

77 OECD, ‘Algorithms and Collusion – Note from the EU’ (n 15) 6-9.; This assumption is further supported by

papers issued by competition authorities used in footnotes 8 and 46.

78 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) 2.

79 ibid. 2.; Ezrachi and Stucke, Virtual Competition (n 10) 46-50.; Patrick A. Perinetto, ‘Hub-and-spoke

arrangements: future challenges within Article 101 TFEU assessment’ (2019) 15/2-3 European Competition Journal, 283 <https://www.tandfonline.com/doi/full/10.1080/17441056.2019.1662209>; Zampa and Buccirosi (n 61) 98-99.

(20)

2. ALGORITHMIC HUB-AND-SPOKE COLLUSION

As underlined in the introduction, with greater access to Big Data comes the growing use of algorithms as a supporting tool for the decision-making process of companies.80 Companies have

also been increasingly relying on outsourced algorithms to maintain their position on the markets where pricing has become even more dynamic.81 The Commission and National Competition

Authorities are thus facing new types of collusion.82 Although hub-and-spoke collusion is not

uniquely related to algorithms it seems these pricing algorithms are medium towards even more facilitated and sustainable collusion.83

As indicated in the introduction, pricing algorithms may be used in various instances but our focus lies on the use of outsourced pricing algorithms – i.e. pricing algorithms sold by IT developers to companies active on the economic market. Previously, the use of a pricing algorithms was only available for large companies having the knowledge and resources to develop a pricing algorithm. Nowadays, pricing algorithms are available for smaller companies as well as a result of the development and following marketing of such algorithms by IT developing companies.84

This section will clarify the meaning of algorithm-driven hub-and-spoke collusion and further explains how collusion develops in that context. Finally, it will demonstrate the legal challenges associated to hub-and-spoke collusion and thereby introduce the following chapter.

a. Definition of Algorithm-Driven Hub-and-Spoke Collusion

As mentioned in the first chapter, algorithms can either be developed internally by a firm or are outsourced by an IT firm. Algorithmic hub-and-spoke collusion occurs when several competitors use the same outsourced pricing algorithm. If multiple competitors on a same market rely on the same or a similar pricing algorithm developed and sold by an external IT firm it is likely that these competitors will coordinate their pricing. This common third-party algorithm will execute

80 Ariel Ezrachi and Maurice E. Stucke, ‘Algorithmic Collusion: Problems and Counter-Measures – Note by A.

Ezrachi & M.E. Stucke’ (DAF/COMP/WD(2017)25), 10

82 Ezrachi and Stucke, Virtual Competition (n 10) 36.; OECD, ‘Hub-and-Spoke Arrangements – Note by BIAC’

(DAF/COMP/WD(2019)112), para. 34

83 Ezrachi and Stucke, ‘Sustainable and Unchallenged Algorithmic Tacit Collusion’ (n 73) 6 and 8.; OECD,

‘Algorithms and Collusion – Note from Italy’ (DAF/COMP/WD(2017)18), para. 1 and 10

(21)

the function of the hub and facilitate alignment of the competitor’s prices by virtue of its computerized decision-making process. 85 It is the increasing use of a common pricing algorithm

that stabilizes prices and reduces competition on a market.86 Although there are presently no

known EU cases of algorithmic hub-and-spoke, it seems this scenario presents the most immediate risk as it only requires competitors to rely on the same pricing algorithm.87

This thesis further makes a distinction between two scenarios of algorithmic hub-and-spoke collusion. The first scenario occurs when companies knowingly use the same pricing algorithm. In such scenario, the aim of the competitors is to reach horizontal collusion and to use pricing algorithm as smokescreen to hide their anticompetitive behavior.88 In this context, the pricing

algorithm takes the role of the hub in what first seems to be a normal and rational economic behavior when it is actually implementing or facilitating a pre-existing agreement or concerted practice. The second scenario concerns competitors inadvertently using a single pricing algorithm. In this case, the competitors’ use of a single algorithm resulting in collusion on prices is merely the consequence of their parallel behavior, but not their original purpose.89 In this

context, the use of a common algorithm is genuinely a result of normal and rational economic behavior. The competitors act independently of each other but make their decisions based on their competitors’ strategies or on the trends of the economic market in general. Given that algorithms are increasingly used in all markets combined, the use of a pricing algorithm can only be qualified as normal.

Regardless of the scenario that ensues, for a pricing algorithm to decide on prices, particularly sensitive information of a company must be shared with that algorithm.90 Information on prices

concern current and previous prices, discounts, production costs, volume, turnovers, etc.91 In a

hub-and-spoke scenario, the hub will use the pricing information and parameters of several competitors to calculate the profit maximizing price for its clients. This indirect information exchange using a similar algorithm enables alignment in two different ways.92

85 OECD, ‘Algorithms and Collusion’ (n 3) 28.; Ezrachi and Stucke, Virtual Competition (n 10) 47-48. 86 Ezrachi and Stucke, Virtual Competition (n 10) 94.

87 Competition & Markets Authority (n 26) para. 5.35. 88 Perinetto (n 79) 281 and 284.

90 European Commission, Guidelines on the applicability of Article 101 of the Treaty on the Functioning of the

European Union to horizontal co-operation agreements (2011/C 11/01) (“Horizontal Guidelines”), para. 73

91 Autorité de la Concurrence and Bundeskartellamt (n 8) 38. 92 ibid. 33.

(22)

b. Alignment at Code Level or Data Level

In a joint paper discussing the competitive risks associated with the use of algorithms, the French Autorité de la Concurrence and the German Bundeskartellamt have identified two distinct circumstances in which alignment of algorithmic decision-making could arise.93 According to

these two national competition authorities, alignment could occur at “code level” or at “data level” and result in collusion on prices.94

Alignment at “code level” means alignment appears at the level of the algorithms. Whilst algorithms are usually individualized per customer – i.e. companies or spokes – the purpose of a pricing algorithm does not vary so much. Furthermore, the methodology implemented to calculate a price is limited to certain features. Usually, companies on a same market will calculate prices based on the same economic features. Hence, because of the common purpose and eventually a similar methodology of calculation, alignment may occur even when algorithms are not identical.95

Alignment at “data level” indicates alignment at the input level used for the calculation of a price. There are three distinct scenarios relating to data level alignment. First, an IT developer might be using a shared pool of data to calculate the price for maximizing joint profit of several competitors by implementing only one common algorithm to that one pool. This is the most far-reaching scenario of alignment at data level identified. The second scenario involves separate calculations for each competitor. However, it does not prevent an IT developer from still using a shared pool of data. Hence, the separate calculations will still imply the use of confidential and sensitive information of other competitors. So even if the data used to calculate a price is not directly shared with the competitors, the use of shared data pool will lead to an alignment of competitor’s prices. Lastly, the competition authorities referred to an alignment of parts of the input data when an IT developer relies on data from a public data source or a specific commercial data supplier.96

93 Ezrachi and Stucke identified this as well (on p. 14 of ‘Two Artificial Neural Networks’ (n 72)). 94 OECD, ‘Hub-and-Spoke Arrangements – Note by Germany’ (n 61) 12

95 Autorité de la Concurrence and Bundeskartellamt (n 8) 33. 96 ibid. 33-34.

(23)

In this regard, a shared data pool, the use of a common or similar algorithm and the methodology of calculation can be characterized as the factors leading to collusion between competitors in a hub-and-spoke scheme.

c. Enforcement Challenges

In its note to the OECD on algorithms and collusion, the EU acknowledged that the use of algorithms and their potential harmful consequences for consumers and competition in general raises questions as to what would be the appropriate response to the risks brought by this new tool.97 On this topic, Ezrachi and Stucke identified three enforcement challenges posed by

algorithmic collusion.98

First, algorithmic tacit collusion may be difficult to detect. As enforcers lack the tools to investigate artificially enhanced prices on a market, detecting anticompetitive practices will be complicated. Especially on a market dominated by algorithms, it may be burdensome for the competition authorities to distinguish between market prices set by computerized systems. The process of algorithms may be difficult to trace back, hence making it difficult for enforcers to identify infringements.99

Second, the authors indicated that antitrust law is not fixed and will have to adapt its policy according to new issues brought by Artificial Intelligence. Indeed, the competition law framework was designed to resolve more predictable and manageable issues resulting from controllable technology.100 AI is the computer science that studies how to design intelligent

machines able to carry out tasks which are too difficult for humans. Algorithms are the practical embodiment of these machines. 101 The AI technology is accompanied by new problems to which

we will have to adapt the surrounding policy as to establish who and to what extent a human can be responsible for the actions of algorithms.102

97 OECD, ‘Algorithms and Collusion – Note from the EU’ (n 15) para. 29. 98 Ezrachi and Stucke, ‘Algorithmic Collusion – Note to the OECD’ (n 80), 17. 99 ibid. 22-25.

100 Commission, ‘Commission Staff Working Document on the free flow of data and emerging issues of the

European data economy’ SWD(2017) 2 final, 43

101 Commission, ‘White Paper on Artificial Intelligence – A European Approach to excellence and trust’

(COM(2020)65 final), 2

(24)

Lastly, algorithmic collusion poses questions with respect to liability. Precisely, tacit algorithmic collusion. In the same vein as the EU’s note, Ezrachi and Stucke indicate that tacit collusion by means of algorithms is a challenging issue as regards the interpretation of current EU competition law. Hence, the legal framework may need to be adapted accordingly. Either way, it seems the EU will not allow companies to hide behind an algorithm to escape liability for illegal pricing practices.103

This thesis focuses on the challenge to define liability in the case of algorithm-driven hub-and-spoke collusion and is discussed under the following chapter. Therefore, the two first challenges presented in this sub-section will not be further discussed.

103 OECD, ‘Algorithms and Collusion – Note from the EU’ (n 15) para. 38.; Ezrachi and Stucke, ‘Algorithmic

(25)

IV. LIABILITY OF THE HUB AND THE SPOKES FOR THE

COMMON USE OF AN OUTSOURCED PRICING

ALGORITHM

The EU distinguishes algorithmic explicit collusion from algorithmic tacit collusion arising from hub-and-spoke collusion schemes.104 Irrespective of the form of collusion, the issue lies with the

extent of knowledge of the competitors involved in the anticompetitive coordination.105 Although

European competition law only captures explicit collusion, it seems the EU does not wish to allow companies to escape liability irrespective of the circumstances.106 In this regard, this

chapter will try to define liability of the hub and the spokes when a common pricing algorithms is knowingly used by several competitors (section 1) or when it is unintentionally used by several competitors (section 2). Therefore, the current framework of EU competition law will be applied as well as relevant CJEU case-law. A landmark judgment on hub-and-spoke collusion resulting from the use of algorithms is still missing107, nevertheless, a few cases have shed some light on

the potential approach of the EU in algorithmic hub-and-spoke collusion cases.

1. INTENTIONAL COMMON USE OF AN OUTSOURCED PRICING ALGORITHM

Hub-and-spoke collusion can occur when firms agree to implement a pricing algorithm to set identical prices on a product. In such a scenario, competitors engage in explicit collusion where the algorithm of a third-party performs the functions of the hub for the competitors.108 Explicit

collusion is captured by the prohibition of Article 101(1) TFEU. The fact that the prices are set through an algorithm does not alter the illegality of explicit collusion.109 Since competition law

restricts the means used to achieve collusion, the enforcement authorities would have to prove the existence of a pre-existing agreement or a concerted practice where competitors – and in some extent the hub – agreed to set identical prices through an algorithm.110 Competition enforces

104 CD, ‘Algorithms and Collusion – Note from the EU’ (n 15) paras. 24-29.

105 OECD, ‘Hub-and-Spoke Arrangements – Note by the European Union’ (DAF/COMP/WD(2019)89, para. 30. 106 OECD, ‘Algorithms and Collusion – Note from the EU’ (n 15) para. 38.; Margrethe Vestager (n 13).

107 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) para. 73.; These difficulties will not be

extensively discussed since this thesis focuses on liability.

108 OECD, ‘Algorithms and Collusion – Note from the EU’ (n 15), para. 24. 109 Picht and Loderer (n 75) 17.

110 I.e. a “meeting of the minds” amounts to an “agreement” under EU competition law (from Bayer AG (n 55),

(26)

proving “concurrence of will” or market behavior explained by a joint intention is sufficient to hold parties liable.111

The first sub-section will analyze the liability of the hub (sub-section a) and is followed by an analysis of the liability of the spokes (sub-section b). The last sub-section will discuss our findings.

a. Liability of the Hub

There are no EU cases specifically discussing liability of the hub. Nevertheless, we can deduct from case-law that the hub in the case of explicit collusion will be identified as the third-party having a role of facilitating and sustaining collusion in a hub-and-spoke scheme.112 Evidently, if

it can be established that the algorithm was designed by the IT developer specifically to facilitate and sustain collusion, then establishing the liability of the hub will be straightforward.113 In other

circumstances, the current case-law has discussed situations where the third-party – i.e. the hub – was held liable for facilitating collusion between competitors.

The first case discussing liability of a third-party is AC-Treuhand.114 In its ruling, ECJ held the

consultancy firm AC-Treuhand liable for facilitating a cartel between competitors. The Court established that a third-party could be held liable as facilitator of horizontal collusion despite not being active on the affected market. Indeed, if a third-party could escape liability simply because its economic activity does not relate to the market affected by the collusion then the effectiveness of the prohibition of Article 101(1) TFEU would be jeopardized.115 Hence, a hub can be held

liable if it (i) contributed or intended to contribute to the attainment of the common objectives of the spokes, and (ii) was aware of the competitor’s actual or planned coordination or that it could reasonably have foreseen it and thus contributed to the realization of collusion through its services. Moreover, the role of the hub must be essential to the attaining and sustaining of the collusive outcome.116 Contribution to the anticompetitive objectives can be active or passive, but

111 Monopolkommission, ‘Algorithms and Collusion’ (n 46) para. 207.

112 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) Box 12. 113 Ezrachi and Stucke, Virtual Competition (n 10) 53.

114 Case C-194/14 P, AC Treuhand AG v European Commission [2015] ECLI:EU:C:2015:717 115 ibid. para. 36.

116 In this case, AC Treuhand as consultancy firm had quite an essential role as it organized and attended meetings,

collected and supplied data to its clients, etc. (from OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) Box 12).

(27)

as long as the hub did not publicly distance itself from the competitor’s behavior or report it then it is considered to have agreed to facilitate the continuation of the anticompetitive conduct.117

This ruling was confirmed in the Icap case.118 The Court again established that Icap had quite a

significant role as it circulated spreadsheets of quotes including sensitive information concerning the banks included in the infringement.119 Moreover, the Court specifies a very intriguing point

that Icap should have expected its conduct was illegal, certainly if it had sought appropriate legal advice, knowing the extensive case-law on the prohibitions set out in Article 101(1) TFEU.120

This statement is indeed interesting as it demonstrates the significant importance of EU case-law on Article 101 TFEU and believes that companies should at least assume their behavior is illegal, or seek legal advice whenever a potential question anticompetitive conduct arises.

b. Liability of the Spokes

It seems that establishing liability of the spokes in explicit hub-and-spoke collusion cases is straightforward. If horizontal collusion is proven, the hub-and-spoke collusion scheme will be assessed as an “ordinary” collusion case. Hence, the spokes of an explicit hub-and-spoke scheme will be held liable for an infringement of Article 101(1) TFEU as long as the enforcement authority can establish the anticompetitive conduct of the spokes and accordingly their intent.121

The anticompetitive conduct in a hub-and-spoke scheme is the exchange of sensitive information between the competitors. Sensitive information is indirectly exchanged through a third-party.122

Additionally, if sensitive information is exchanged following an agreement or in the context of concerted practice, then the spokes’ conduct will fall under the prohibition of Article 101(1) TFEU.

When the competition enforcer finds proof of a pre-existing agreement that was implemented through a pricing algorithm, liability of the spokes is quite easily established. However, proving horizontal collusion resulting from a concerted practice will generally be quite difficult since in

117 AC Treuhand (n 114) para. 30-31.

118 Case T-180/ 15, Icap and Others v European Commission [2017] ECLI:EU:T:2017:795

119 ibid. para. 198.; OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) Box 12 120 Icap (n 118) para. 197.

121 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) para. 80. 122 Horizontal Guidelines (n 90) para. 55.

(28)

hub-and-spoke schemes there is none or hardly any direct contact between the spokes.123

Nevertheless, the ruling of two cases permit to hold competitors liable for their conduct if certain conditions are fulfilled.124 Neither of these cases involve algorithms, they are nevertheless

relevant to assess the potential extent of liability the ECJ may establish in cases of explicit collusion in hub-and-spoke schemes.

The first case is the Eturas case where a pricing software was used to implement a pre-existing agreement.125 This case does not mention hub-and-spoke collusion, yet the facts are similar to

what could occur with hub-and-spoke schemes.126 In the case, the online booking platform Eturas

offered to thirty travel agencies the option to give discounts on online bookings. In a message addressed to the travel agencies, Eturas proposed to implement a software rule capping the discounts to a maximum of 3%. In the event a travel agency wanted to grant a greater discount, the travel agencies could do so individually and if taking additional steps. Regarding these facts, the ECJ ruled that the travel agencies were presumed to be aware of the message concerning the discounts from the day it was dispatched as they behaved in accordance with it. The Court thus ruled that the agencies having received the message on the 3% cap and subsequently complying with the discount cap could be presumed to having participated in a concerted practice within the meaning of Article 101(1) TFEU. In this sense, all parties were liable for the infringement of Article 101(1) TFEU if (i) the agencies were aware of the message or objective and consistent indicators that they tacitly consented to the anticompetitive conduct; (ii) subsequent behavior of the agencies could be observed and (iii) the cause and effect between the message and the agencies’ behavior is linked.127 Finally, a travel agency could only escape liability if it had

publicly distanced itself from the concertation.128

123 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) para. 80.; OECD, ‘Hub-and-Spoke

Arrangements – Note by the European Union’ (n 105) para. 34.

124 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) para. 80.

125 Case C-74/14, Eturas and Others v Lietuvos Respublikos konkurencijos taryba [2015] ECLI:EU:C:2015:493 126 Case C-74/14, Eturas and Others v Lietuvos Respublikos konkurencijos taryba [2015] ECLI:EU:C:2015:493,

Opinion of AG Szpunar, para. 65; The ECJ precised that the Eturas case does not resemble to hub-and-spoke collusion. Nevertheless, it has been used by the EU in a note to the OECD on algorithmic collusion (‘Algorithms and Collusion – Note from the EU’ (n 15). Further, the OECD used the interpretation of the CJEU in the case when discussing liability of the “hub” in its report of hub-and-spoke arrangements (‘Roundtable Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11)).

127 Eturas (n 125) paras. 42-45.

128 OECD, ‘Hub-and-Spoke Arrangements – Note by the Secretariat’ (n 11) Box 12.; OECD, ‘Algorithms and

Collusion – Note from the EU’ (n 15) para. 25.; OECD, ‘Hub-and-Spoke Arrangements – Note by Germany’ (n 61) para. 47-48

(29)

A few years later, the ECJ rules in the case VM Remonts on hub-and-spoke collusion.129 The case

concerns a “pure” hub-and-spoke scheme where a third-party service provider shared sensitive information to two clients both competing for the same tender.130 Particular to this case is that

one of the spokes was not aware of the hub’s behavior. The ECJ ruled that an undertaking may nevertheless be held liable for the acts of the independent service provider which lead to a concerted practice if the undertaking “could reasonably have foreseen that the service provider retained by it would share its commercial information with its competitors and if it was prepared to accept the risk which that entailed.” 131 In this regard, the Court established that a spoke may

be liable for the actions of an independent third-party if only one of the following conditions is met: (i) the service provider was in fact acting under the direction or control of the undertaking concerned, or (ii) the undertaking was aware of the anticompetitive objectives pursued by its competitors and the independent service provider and intended to contribute to them by its own conduct, or (iii) the undertaking could reasonably have foreseen the anticompetitive acts of its competitors and the service provider and was prepared to accept the risk which they entail.132

With its last condition, the ECJ settled quite a low standard to prove the intent of a spoke and enables competition enforcers to prove horizontal collusion more easily.

c. Conclusion on Liability for Intentional Hub-and-Spoke Collusion

Although the cases discussed did note involve the use of pricing algorithms, their facts and ruling are quite relevant to the issue of liability of hub and spokes in the context of explicit collusion. Since there is no reason to establish liability differently simply because collusion occurred through a pricing algorithm, there is no doubt that hub and spokes will be held liable for the infringement of Article 101(1) TFEU in light of the cases discussed.

The potential interpretation on the liability of the hub poses a few questions. Holding a hub liable for facilitating collusion between competitors may be excessive. Firstly, because a hub is generally not active on the same market as the one impacted by the horizontal collusion. Secondly, the hub is vertically related to the spokes and is thus not part of the horizontal collusion. Finally, some argue the hub might not necessarily be aware of its role in the hub-and-spoke scheme if the hub-and-spokes have set this scheme up.133 These arguments can be even more true

129 Case C-542/14, SIA ‘VM Remonts’ and Others v Konkurences padome [2016] ECLI:EU:C:2016:578 130 Perinetto (n 79) 291.

131 VM Remonts and Others (n 129) para. 31. 132 VM Remonts and Others (n 129) para. 33.

(30)

in the context of an algorithmic hub-and-spoke scheme. Indeed, an IT developer is generally not active on the same market as the undertakings using its pricing algorithm and the developer is thus not part of the horizontal collusion. Moreover, the IT developer may not even be aware that competitors on a same market use its pricing algorithm as a tool to collude on prices. However, as argued by the ECJ in the AC-Treuhand case, enabling an IT developer to escape liability would jeopardize the effectiveness of Article 101(1) TFEU. Furthermore, if a competition enforcers find proof of the IT developer’s knowledge concerning the competitors’ plan and still took the risk to participate, or that the IT developer knew or acknowledged that its algorithm could be implemented by competitors to obtain collusive outcomes, it would seem odd to enable the developer to escape liability. An IT developer may as well totally benefit from the fact that competitors are aware of the common use of that one pricing algorithm and thereby attract even more clients. It is different if it can be proved that the algorithm was designed to enhance collusion, then liability of the IT developer cannot be neglected. It could be that liability of the hub will have to be decided on a case-by-case basis. However, this is not a very efficient solution as it is an additional burden for competition enforcers. Yet, in the absence of tools able to detect the roles and extent of knowledge of participants to hub-and-spoke collusion scheme this is currently the only available alternative.

Liability of the spokes could easily be established in light of the conditions set out in Eturas and

VM Remonts. Although it may not be easy to detect anticompetitive agreements or concerted

practice when these involve a pricing algorithm, assessing liability under the conditions should not be burdensome. Indeed, merely requiring awareness and behavior conforming with it makes it easier for competition enforcers to assess liability.134 Moreover, VM Remonts has extended the

circumstances in which a spoke may be held liable since a spoke may be held liable for a hub’s anticompetitive behavior if the spoke could only foresee the risk it took by exchanging sensitive information to an independent service provider.135 Although some may argue the Court went too

far in its ruling, it follows the assumption that companies on a market are aware of the risk of information being shared to their competitors since algorithms are increasingly being used.

134 Eturas (n 125) paras. 42-45.

Hub-and-Spoke Collusion under EU Competition Law: Liability of Participants in Algorithmic Collusion

Master’s Thesis

LL.M. European Competition Law and Regulation