• No results found

Protecting competition in the digital age- algorithmic tacit collusion at the intersection of EU competition law and regulation

N/A
N/A
Protected

Academic year: 2023

Share "Protecting competition in the digital age- algorithmic tacit collusion at the intersection of EU competition law and regulation"

Copied!
51
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

Protecting competition in the digital age- algorithmic tacit collusion at the intersection of EU competition law and

regulation

Christopher Hristov 13984519

Thesis supervisor: Dr. Giacomo Tagiuri

LLM European Competition Law – Faculty of Law 2021-2022 Word count: 15966

(2)

2

Table of contents:

1. Introduction 5

2. Pricing algorithms and their benefits 7

2.1. What are pricing algorithms and how do they operate 7

2.2. Benefits of the use of algorithms 10

3. Anticompetitive effects of the use of algorithms- four scenarios 11

3.1. Algorithms as facilitators of anticompetitive conduct 12

3.2. Algorithmic hub and spoke scenario 14

3.3. Algorithms as means of achieving and sustaining tacit collusion 14

4. Plausability of algorithmic tacit collusion- an empirical question 17

4.1. The pervasiveness of pricing algorithms in digital markets 18

4.2. Evidence in favour of the emergence of algorithmic tacit collusion 18

4.3. Counterevidence 20

4.4. The role of this research in the wider context of academic research on EU competition law 21

5. In seach of solutions- the proposals of scholars for addressing algorithmic tacit collusion 23

5.1. A per se prohibition of specific types of algorithms 23

5.2. Liability of producers of pricing algorithms 25

5.3. Algorithmic tacit collusion as a type of concerted practice 26

5.4. Limitation in the algorithms’ speed as a potential remedy 28

5.5. Market inquiries as facilitatory tools 28

6. Regulation as another instrument in the toolbox 29

6.1. The concept of regulation 31

(3)

3

6.2. Competition law and regulation compared 33

6.3. The interface between competition law and regulation 35

6.3.1. Competition law and regulation as substitutes 36

6.3.2. Competition law and regulation as complimentary regimes 37

7. A solution in sight? Tackling the challenges presented by algorithmic tacit collusion in a holistic manner 39

7.1. The necessity of combining regulatory and competition law remedies 39

7.2. Finding a holistic solution to the problem of algorithmic tacit collusion 42

8. Conclusion 46

9. Bibliography 48

(4)

4

Abstract

Algorithmic pricing has become an ever more widespread phenomenon in both digital and brick- and-mortar settings. With the development of algorithmic technologies this brings up the

prospect of such algorithms to be used by undertakings for the purposes of reaching supra competitive prices in ways that reminisce the practice of tacit collusion. If it materializes, such a development could circumvent the scope of EU competition law due to the fact that tacit

collusion is currently not prohibited and would result in harm to both consumers and competition as a whole. This paper discusses precisely this problem. It outlines how algorithmic tacit

collusion could be achieved and subsequently discusses the ways in which it can be addressed by policymakers within the EU. In doing so, this paper explains the negative effects that algorithmic tacit collusion could bring to both markets and consumers alike. It then proceeds to analyse potential remedies that can be deployed as countermeasures. Deviating form the currently existing academic literature on the topic, this paper proposes a new approach towards the construction of strategies for addressing algorithmic tacit collusion. More specifically, it maintains that the solution of the problem can be found not only through the use of EU competition law but also through ex ante regulation. Having this in mind, the paper then proposes a combination of remedies that serves as a strategy for tackling algorithmic tacit collusion.

(5)

5

1. Introduction

It is rarely disputed that digital technologies are an integral part of our modern societies and economies. From mundane tasks such as ordering a product online to the execution of highly complex business strategies by multinational companies, technologies play important roles not only in our personal lives but also in the way businesses conduct their operations. However, amongst all new developments in science, there is one field which plays an especially important role in our daily lives and that is the field of algorithmic technologies. According to some scientists, algorithms are a pervasive force in our societeis and they are capable of tracking, predicting and even influencing the actions that we take in our everyday lives.1 Becuase of these capabilities, the use of algorithms has resulted in significant benefits when it comes to the operation of markets. They have led to improved automation, efficiency and quality for both businesses and consumers.2 Algorithms have been able to take on tasks that previously required human decision-making and with the constant development of algorithmic technologies this number of tasks is only set to increase in the future. However, at the same time, even though it is capable of bringing with itself significant efficiencies, the raise of algorithms can also have negative effects on markets. In this regard, one of the most relevant technological developments is the emergence of algorithms which are capable of automatically calculating and setting the prices of individual products, also known as pricing algorithms.

The purpose of this thesis is to analyze the role of pricing algorithms in the opearation of markets and the ways in which they can influnence competition. More specifically, it explores the emerging issue of the possible use of such algorithms as tools for carrying out a specific type of anticompetitive strategy that has been referred to in the economic literature as ‘tacit collusion’.

To this end this thesis seeks to outline the harm that algorithmically induced tacit collusion can bring to the compeitive process and discuss how policymakers can prevent it. In doing so, it situates itself within the academic literature on algorithmically induced tacit collusion. It

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

<www.oecd.org/competition/algorithms-collusion-competition-policy-in-the-digital-age.htm> accessed 24 May 2022, page 5.

2 Ibid.

(6)

6 identifies the overarching approach that scholars have adopted thus far of treating EU

competition law as the main source of potential remedies for tackling the issue and critically evaluates it as being insufficient. Consequently, this thesis proposes a new approach for constructing a strategy for addressing the problem of algorithmically induced tacit collusion.

This approach entails the inclusion of ex ante regulation as source of additional remedies that would compliment those provided by EU competition law and contribute to the creation of more effective solutions to the problem. In doing so, the thesis also interacts with the academic

literature that discusses the interface between competition law and regulation. In this sense, it treats the debate on algirithmically induced tacit collusion as an example that shows how policymakers ought to treat the relationship between ex ante regulation and competition law.

Lastly, after doing that, the thesis makes its final contribution towards the academic literature by arguing how a future strategy for tackling the issue of algorithmically induced tacit collusion has to be formulated.

The central research question is:

“How does the emergence of pricing algorithms affect the ability of undertakings to participate in tacit collusion and to what extent should the framework of European competition law change

in order to regulate their use?”

In answering it, the thesis is divided into 8 chapters. Chapter 2 starts by explaining what pricing algorithms are, how they operate, and the efficiencies that they can create for businesses. Then, chapter 3 outlines the main ways in which the use of pricing algorithms may result in businesses carrying out anticompetitive practices. After that, chapter 4 acquaints the reader with the academic debate on the empirical plausibility of algorithmic tacit collusion occurring in the near future.

Following on, chapter 5 provides the foundation for the discussion of the research question by outlining the main solutions to the problem that have been proposed in the academic literature so far. Then, chapter 6 provides a criticism of the approach adopted by the academic literature of focusing on finding a competition law solution to the problem and explores the relationship between competition law and regulation as two distinct regimes that can both serve as sources of potential remedies. After that, building on top of the chapters preceding it, chapter 7 presents a

(7)

7 new approach for how a strategy for overcoming algorithmic tacit collusion can be formulated and proposes a combination of remedies to this end. Finally, chapter 8 concludes.

Furthermore, throughout its text, this thesis makes use of a variety of different source. It explores the relevant academic legal and economic literature on algorithmic tacit collusion and connects it to the legal literature on market intervention through both competition law and ex ante

regulation. Furthermore, it also makes use of the relevant provisions from the founding treaties of the EU that establish the rules of EU competition law, as well as the Court of Justice of the EU. The objective of this thesis is to contribute towards the debate of what measures the EU and its member-states should adopt in order to respond to the future challenges that could be created by the emergence of algorithmic tacit collusion.

2. Pricing algorithms and their benefits

2.1. What are pricing algorithms and how do they operate?

To start with, it must be pointed out that there is no one universally accepted definition of what constitutes an algorithm.3 As a result, different authors have used different definitions for the purposes of their writings. However, for the purposes of this master’s thesis, a simple and more understandable definition of what constitutes an algorithm would be the following: “An algorithm is 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”.4

Building on that, pricing algorithms are, as the name aptly suggests, algorithms whose function is to calculate and set the prices of specific products. To illustrate how a simple pricing algorithm could work an example can be given. Imagine, that on a given marketplace there are two

3 Ibid 8.

4 Thomas Cormen, Charles Leiserson, Ronald Rivest, Clifford Stein, Introduction to algorithms (4th edn, The MIT Press 2022), page 5; Cento Veljanovski, ‘Pricing Algorithms as Collusive Devices’ (2022) 53(4) IIC - International Review of Intellectual Property and Competition

Law <https://doi.org/10.1007/s40319-022-01177-8 > accessed 24 May 2022, page 606.

(8)

8 companies that sell similar or identical products and one of the companies has decided to use a pricing algorithm to calculate and set the prices of its product. In this situation, in order to compete, this company decides that it want to always price its products at 95% of the price level of the products of its rival. As a result, in order to achieve this goal, the algorithm would perform a set of simple operations (procedure). First, it will see what is the price that has been set by its competitor. Then it will calculate what 95% of that price would be and after it is done with that calculation, it will perform the act of changing/setting the price of the respective product.5 Furthermore, algorithms can be categorized by the method in which they learn and the possibility of their functioning to be interpreted by humans. In terms of their method of learning, algorithms can be categorized in many ways. For example, according to their method of learning, algorithms can be “supervised learning algorithms”, “unsupervised learning algorithms”, and “reinforced learning algorithms”.6

Supervised learning algorithms are algorithms that aim to learn how to analyse an input and label it correctly. For example, if the algorithm is presented with a picture of a cat, its task would be to analyse it without being told what it is and determine that it is indeed a picture of a cat. The reason why they are named this way is because, to learn, they use a set of labelled data that is being given to them by an individual (e.g., a data scientist). To use the abovementioned example, the algorithm can be given a picture of a cat and also the knowledge that this picture is of a cat (the label). In essence this means that the algorithm attempts to learn the general rules/patterns between the input data and the labels. Then, after, it has been ‘trained’, the algorithm is presented with never-before-seen data (input) and its function is to analyse it and come up with an

appropriate label for it.7 In contrast to this, unsupervised learning algorithms are also presented

5 Salil Mehra, ‘Antitrust and the Robo-Seller: Competition in the Time of Algorithms’ (2016) 100 Minnesota Law Review, page 3.

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

<www.oecd.org/competition/algorithms-collusion-competition-policy-in-the-digital-age.htm> accessed 24 May 2022, page 9.

7 Ibid 7.

(9)

9 with a set of data that is, however, not labelled. Rather the algorithm itself is designed to analyse such unlabelled data and detect patterns, similarities and hidden structures on its own.8

Last but not least, reinforced learning algorithms are algorithms which are put in an environment to perform a specific task. This could be for example, driving a vehicle or playing a game. Then, while they are performing their task, the algorithms will learn on how to do it as effectively as possible by learning through trial and error.9

Furthermore, in terms of their interpretability algorithms are categorized as “black box”

algorithms whose decision-making process cannot be interpreted by humans and interpretable algorithms whose decision-making process can be examined and assessed by humans.10 However, for the purposes of this master’s thesis, the focus falls on the legal implications stemming from the introduction of pricing algorithms into the business practices of companies, rather than the scientific classification and definition of algorithms. This is due to the fact that the description and scientific classification of algorithms is a complex topic and for better or worse, it is more appropriate that it is left for the more competent experts in other fields, such as mathematics or software engineering, to participate in that debate. However, what is important is what the abilities of pricing algorithms are and what could be the consequences of their use for the competitive process.

In order to carry out their task of setting prices, pricing algorithms take into account a multitude of different factors (also referred to as input data). For more simple algorithms the input can be limited to only the prices of competitors.11 However, more sophisticated algorithms can take into account a higher number of factors such as calculations about the current levels of supply and demand on a given market, as well as predictions of future supply and demand, currently

8 Ibid.

9 Ibid.

10 Veljanovski (n 4) 607.

11 Marshall Fisher, Santiago Gallino, Jun Li, ‘Competition-Based Dynamic Pricing in Online Retailing: A Methodology Validated with Field Experiments’ (2017) 64(6) Management Science

<https://doi.org/10.1287/mnsc.2017.2753 >accessed 24 May 2022, page 2496.

(10)

10 available stock and personal behaviour and characteristics of the consumer like age, gender, educational background, etc…12

2.2 Benefits of the use of algorithms

As has been noted by the OECD, algorithms can outperform humans in many tasks since they can process larger amounts of data in a faster manner and with lower costs when compared to humans.13 Moreover, also in contrast to human decision-makers, algorithms are capable of taking consistently identical decisions when working with the same data set. This is because unlike humans, algorithms are not influenced by factors such as the time of the day, hunger and personal mood and will always reach the same conclusion when working with the same set of data.14 As a result, companies can substitute or assist human employees by adopting the right type of algorithm that would best serve their particular business needs. All of this allows them to optimize their business processes and brings with itself important supply-side efficiencies that benefit both them and their consumers.

One of the main ways in which, algorithms can make markets more efficient is by providing sophisticated price discovery. This is because algorithms can quickly and automatically respond to changes in supply and demand by taking into account a multitude of factors such as currently available stock, predictions of future supply and demand, personal characteristics of the

consumers, etc…15 As a result, by using algorithms, businesses will be better able to find the optimal price levels for a given product even in rapidly changing market conditions.16 This on its

12 Peter Seele, Claus Dierksmeier, Reto Hofstetter and Mario Schultz, ‘Mapping the Ethicality

of Algorithmic Pricing: A Review of Dynamic and Personalized Pricing’ (2021) 170 Journal of Business Ethics <https://doi-org.proxy.uba.uva.nl/10.1007/s10551-019-04371-w> accessed 24 May 2022, page 702; OECD (n 6) 16.

13 OECD (n 6) 11.

14 Sloan, Robert H. and Richard Warner, ‘When Is an Algorithm Transparent?: Predictive Analytics, Privacy, and Public Policy’ (2017) IEEE: Security & Privacy <http://dx.doi.org/10.2139/ssrn.3051588>

accessed 29 July 2022, page 4.

15 OECD (n 6) 14.

16 Ambroise Descamps, Timo Klein, and Gareth Shier, ‘Algorithms and competition: the latest theory and evidence’ (2021) 20(1) Competition Law Journal < https://doi.org/10.4337/clj.2021.01.04> accessed 24 May 2022, page 35.

(11)

11 own means that markets where suppliers use pricing algorithms may be better able to achieve equilibrium.

Another benefit stemming from the use of algorithms is that when algorithms are able to react more quickly to changes in supply and demand, perishable goods such as groceries, hotel rooms and airline tickets are more likely to be sold and not go to waste. In addition to this, the fact that perishable goods can be more efficiently sold also means that the management of the inventory of such goods will be better facilitated and improved 17

Furthermore, the ability of algorithms to automatically determine and optimize price levels may be especially useful for companies selling a large number of different products as this would be a streamlining of a task that is crucial for their operation and is usually more costly. This on its own could lead to cost reductions which could then be passed on to consumers.18

Lastly, it should be noted that this set of examples is not an all-encompassing list of the supply- side efficiencies that algorithms are capable of producing. With the advancement and

development of algorithm technologies, it is more likely than not that new ways of achieving even more supply-side benefits will emerge. Rather, the examples given in this sub-section serve to show that pricing algorithms can have important positive consequences for both producers and consumers alike and that their current and future impact on markets should not be

underestimated by any means.

3. Anticompetitive effects of the use of algorithms- four scenarios

Even though pricing algorithms are capable of bringing with themselves important

efficiencies for businesses and consumers alike, they are also not without their downsides. Due to their versatility and effectiveness, algorithms can be used in many different scenarios that may give rise to competition concerns. Furthermore, algorithms may also create problems that are

17 Competition and Markets Authority, ‘Pricing algorithms’ (2018)

<https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/74635 3/Algorithms_econ_report.pdf> accessed 24 May 2022, page 20.

18 Descamps, Klein, and Shier (n 16) 35.

(12)

12 outside the domain of competition law. For example, algorithms have been said to allow for the raise of practices such as dynamic pricing (a practice of dynamically adjusting prices in order to achieve revenue gains, while responding to a given market situation with uncertain demand) and personalized pricing (a practice where firms charge different prices to different consumers based on their willingness to pay and other personal characteristics). These practices bring with

themselves a set of concerns such as ethical issues related to fairness and potential discrimination on grounds such as race, ethnicity, gender, etc… 19However, while they are related to the use of pricing algorithms, these concerns form part of a debate that is different than the one to which this thesis seeks to contribute. As a result, the practices of dynamic pricing and personalized pricing will only be discussed in the context of the role that they play in influencing the process of free market competition.

3.1. Algorithms as facilitators of anticompetitive conduct

The first type of situations is one where pricing algorithms are used to implement an anti- competitive agreement that has been concluded by the relevant parties. Such scenarios amount to a competition law violation with modern age means. In the literature on algorithms and

anticompetitive collusion, it has also been referred to by some scholars as the ‘messenger scenario’. Notably, this name has been coined by professor Ariel Ezrachi and professor Maurice Stucke.20 In this type of situations, humans agree on the anticompetitive actions that the

companies they represent will follow (for example, fix prices, reduce output, allocate markets, etc…) and after that they turn to use algorithms as means of carrying out their agreement in

19 Laura Drechsler, ‘The Price Is (Not) Right: Data Protection and Discrimination in the Age of Pricing Algorithms’ (2018) 9(3) European Journal of Law and Technology <https://ssrn.com/abstract=3401907>

accessed 24 May 2022, page 5; Seele, Dierksmeier, Hofstetter and Schultz (n 12) 704-706.

20 Ariel Ezrachi and Maurice Stucke, ‘Artificial intelligence and Collusion: when computers inhibit competititon’ (2017) 2017 University of Illinois Law Review <https://illinoislawreview.org/wp-

content/uploads/2017/10/Ezrachi-Stucke.pdf> accessed 24 May 2022, page 1784; Lea Bernhardt and Ralf Dewenter, ‘Collusion by code or algorithmic collusion? When pricing algorithms take over’(2020) 16(2- 3) European Competition Journal <https://doi.org/10.1080/17441056.2020.1733344> accessed 24 May 2022, page 326; Ulrich Schwalbe, Algorithms, ‘Machine Learning, and Collusion’ (2018) ) Institute of Economics, University of Hohenheim <http://dx.doi.org/10.2139/ssrn.3232631> accessed 24 May 2022, page 5.

(13)

13 practice.21 Algorithms are useful in implementing such anticompetitive agreements because they can be used for example to transmit information between cartelists about planned price increases, supply reductions or special offers.22 Furthermore, they could also be used by companies to monitor the prices of the other cartel participants and automatically set their own prices in compliance with the agreement or change their prices and execute a ‘price war’ as punishment in case of a deviation from the agreement by the other cartelists. In this type of situations,

algorithms do not eliminate the need for communication between the undertakings during the establishment of the anticompetitive agreement, rather, they have the effect of stabilizing it and extending and intensifying the harm it causes.23

It should be noted that thus far there have been real life cases where algorithms have indeed been used to implement anticompetitive agreement. In the context of EU competition law, a notable example is the case of Eturas and others where an online travel booking platform had been used to facilitate collusion by travel agencies in terms of the amount of discount that they would provide through the use of said booking platform.24 Another example worth mentioning, that has however occurred in the context of US antitrust law, is the case of US v Topkins where two sellers on the Amazon Marketplace had used algorithms to coordinate their sales at collusive non-competitive prices.25

21 Ezrachi and Stucke, ‘Artificial intelligence and Collusion’ (n 20) 1784.

22 Bernhardt and Dewenter (n 20) 326 and 327.

23 Bernhardt and Dewenter (n 20) 327; Francisco Beneke, ‘Mark-Oliver Mackenrodt, Remedies for algorithmic tacit collusion’ (2021) 9(1) Journal of Antitrust Enforcement

<https://doi.org/10.1093/jaenfo/jnaa040 > accessed 9 July 2022, page 165.

24 Case C-74/14 "Eturas" UAB and Others v Lietuvos Respublikos konkurencijos taryba EU:C:2016:42,

paragraph 50.

25 Salil K. Mehra, ‘US v. Topkins: can price fixing be based on algorithms?’ Journal of European Competition Law & Practice (2016) 7 (7), <https://doi.org/10.1093/jeclap/lpw053> accessed 24 May 2022, page 472.

(14)

14

3.2. Algorithmic hub and spoke scenario

The second type of situations is one where two or more undertakings coordinate their actions through the use of pricing algorithms and a vertical relationship with a third party. This has also been referred to as a ‘hub and spoke’ scenario. This scenario includes two types of cases.

The first type are cases where undertakings purchase their respective pricing algorithms from the same third-party. As a result, two or more undertakings could have similar or identical

algorithms which could then lead to identical market conduct which would effectively amount to conscious or unconscious horizontal coordination.26 The second type are cases where

undertakings would electronically coordinate with a third party (for example by sending their costs data) which would then suggest or outright set the profit-maximizing price.27

However, it should be noted that when it comes to the ‘hub and spoke’ scenario the competing undertakings have not come into contact with each other. The collusion in such a situation does not come exclusively from the algorithms themselves. The algorithms of the two undertakings do not communicate between each other, nor are they programmed to “know” each other’s future actions or “learn” to read and predict each other’s actions. Rather, the collusive outcome is a function of the vertical relationship of the undertakings with a third party (the hub).

3.3 Algorithms as means of achieving and sustaining tacit collusion

The third type of situations is one where undertakings use algorithms in order to reach and facilitate a form of behaviour that is referred to in competition law literature as ‘tacit

collusion’ or ‘conscious parallelism’. This type of situations is the main object of attention of the analysis contained in this thesis. The terms tacit collusion’ and ‘conscious parallelism’

themselves refer to a scenario where undertakings are capable of coordinating on a given market without concluding an agreement or participating in concerted practices pursuant to article 101 TFEU. However, while they do not communicate with each other, undertakings participating in

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

<www.oecd.org/competition/algorithms-collusion-competition-policy-in-the-digital-age.htm>accessed 24 May 2022, page 27.

27 Ezrachi and Stucke, ‘Artificial intelligence and Collusion’ (n 20) 1787; Niccolò Colombo, ‘Virtual Competition: Human Liability Vis-a-Vis Artificial Intelligence's Anticompetitive Behaviours’ (2018) 2(1) European Competition and Regulatory Law Review <https://doi.org/10.21552/core/2018/1/5> accessed 24 May 2022, page 13.

(15)

15 tacit collusion are still capable of achieving anticompetitive effects similar or identical to those resulting from explicit collusion. 28 In order to occur, tacit collusion requires at least three conditions. First, there needs to be a sufficient degree of transparency on a specific market. This is necessary in order for the different competitors on the market to be able to quickly and easily monitor how their counterparts are behaving on the market so that, if necessary, they could retaliate if another undertaking deviates from the common conduct. Second, the actions that the undertakings have taken in a coordinated manner have to be sustainable. In essence, this means that if one of the competitors decides to ‘cheat’ by deviating from the common conduct (for example by lowering prices), the other competitors have to be able to nullify the benefits of its cheating (by for example also lowering prices and thus negating any significant profits that could have flowed from the lower price such as attracting new clients). Lastly, the third condition that needs to be present is that there needs to be an absence of effective competitive constraints. In essence, this means that the reactions of competitors, customers and consumers should not be such as to jeopardize the results expected from the coordination.29

Coming back to the main topic at hand, algorithms could be used in order to more easily achieve tacit collusion. They could do so by changing the dynamics of the market and enhancing the abovementioned conditions of transparency and sustainability. For example, algorithms could constantly scan the marketplace for any changes in price levels (thus contributing towards the transparency of the market) and if such changes occur, quickly react to punish these deviations by automatically setting new lower prices (thus contributing towards the sustainability of the common conduct). 30 Furthermore, another way in which algorithms can increase the likelihood of tacit collusion is by enhancing the implementation of a practice called ‘price signaling’. Such practice entails that companies increase their prices as signals to competitors (a form of ‘offer’ to raise prices) which if followed upon would lead to a collusive-like outcome without a need for direct communication.31 The way in which pricing algorithms can increase the likelihood of signalling is by reducing the potential costs that signalling may have. For example, if one

28 Richard Whish and David Bailey, Competition Law (10th edn, OUP 2021), page 591.

29 Ibid 592.

30 Ezrachi and Stucke, ‘Artificial intelligence and Collusion’ (n 20) 1789.

31 OECD (n 26) 28.

(16)

16 undertaking is to signal a price increase by increasing its prices for a set of products, if its

competitors do not also raise their prices, said undertaking is likely to lose money. Algorithms can significantly reduce these costs if both the undertaking in question and its competitors are using algorithms to set their prices. This is due to the fact that, as has already been mentioned, algorithms are able to scan the market and adjust price levels in a fast manner. Because of that, if after signalling a price increase, an algorithm does not quickly see a subsequent increase by the algorithms of competitors, it will know that its signal has not been accepted and it will be able to quickly change the price back to its original level. Thus, there will be a smaller loss of profit because prices will be higher for a smaller period of time.32

Finally, it should be noted that in this type of situations where pricing algorithms are used to reach and facilitate tacit collusion, they do not agree to coordinate with each other in the conventional sense that could amount to an agreement or concerted practices under article 101 TFEU.

The fourth type of situation is one where pricing algorithms are capable of learning to achieve tacit collusion without assistance by humans who work at the business that uses them. For example, in such a scenario, the algorithm is given an objective of maximizing profit. If it is sufficiently complex, it could learn by itself and experiment with its pricing strategy in order to achieve an optimal outcome. Consequently, as a way of achieving its task of maximizing profit, the algorithm could find that one way is to tacitly collude with competitors by for example enhancing the transparency of the market. 33 This this type of situations, algorithms are capable of operating autonomously and in doing so reach a collusive outcome without coordinating in the conventional sense with the algorithms of other competitors. Furthermore, it should be pointed out that in such a scenario it is not necessary that the developers/creators of the algorithm or the businesses that use it intend for it to participate in conscious parallelism, nor is it necessary that they are aware if it even does so and if yes, when does it do so and for what period of time. In essence, here the relationship between the algorithm and the business that uses it is characterized by more limited interactions between them and a higher degree of delegation of decision-making autonomy by the business to the algorithm. This limited relationship, however, has big

32 Ibid 29.

33 Competition and Markets Authority (n 17) 28.

(17)

17 implications in terms of the legal framework of EU competition law. Currently, tacit collusion is not prohibited under EU competition law.34 At the same time, the risks associated with

algorithms participating in tacit collusion can be more significant than those of regular tacit collusion reached by humans. Since, as has been stated above, algorithms could be better suited to optimizing price levels as compared to humans, there is a concern that a widespread adoption of algorithms may lead to an increase in the number of markets where tacit collusion can occur.

This is because the larger the number of undertakings which use algorithms is, the more any imperfections in human decision-making are going to be substituted for the efficiencies of

machines. For example, it has been stated that tacit collusion is a problem mainly associated with oligopolistic markets which are characterized by the fact that only a few firms operate on them.35 However, if sophisticated pricing algorithms are employed, this could open the possibility for tacit collusion to be achievable on what are currently robustly competitive markets with a higher number of competitors operating on them.

4. Plausability of algorithmic tacit collusion- an empirical question

As can be seen from section 3.3, the theoretical model of algorithmic tacit collusion has already been formulated by the academic scholarship. However, connected to it is an important but different debate that concerns its practical side. More specifically, this debate is about whether algorithms are currently capable of participating in algorithmic tacit collusion or whether they will obtain the capabilities to do so in the near or medium term. Like any other debate, it has supporters who agree with either side and the purpose of this chapter is to provide an overview of some of the main evidence presented by each side. With this, it aims to inform the reader about what is currently known about the practical feasibility of algorithmic tacit collusion and outline the purpose and value of the theoretical research contained in this thesis.

34 Richard Whish and David Bailey, Competition Law (10th edn, OUP 2021), page 596.

35 Ibid 588-591.

(18)

18

4.1. The pervasiveness of pricing algorithms in digital markets

During the last decade, pricing algorithms have become an important and widespread tool adopted by businesses. For example, in its final report on the E-commerce Sector Inquiry in 2017, the European Commission concluded that two thirds of online retailers used ‘automatic software programmes that adjust their own prices based on the observed prices of competitors’.36 Furthermore, the report also concluded that the availability of real-time pricing information on the market may also trigger automatised price coordination and that a wide-scale use of such software may, depending on market conditions raise competition concerns.37 Since the report was published in 2017, naturally one can assume that during the time that has passes since then, the number of pricing algorithms used by businesses has only grown. Furthermore, pricing algorithms have also been adopted in brick-and-mortar settings. For example, pricing algorithms have been adopted by retailers operating in German and Dutch gasoline markets.38 This

relatively widespread adoption of pricing algorithms is the reason why their potential capabilities to contribute towards undertakings achieving tacit collusion are being hotly debated.

4.2. Evidence in favour of the emergence of algorithmic tacit collusion

So far, there has been a substantial number of scholars who have expressed views in support of the idea that algorithmic tacit collusion is a credible threat or at the very least that it can be a credible threat. One such scholar is Michal Gal. According to her, algorithms are capable of facilitating tacit collusion in markets that possess high barriers to entry.39 This is due to the fact that they make it easier for undertakings to reach a common understanding on the terms of an ‘agreement’ of what the supra-competitive equilibrium will be, while also increasing

36 Commission, ‘Final report on the E-commerce Sector Inquiry COM(2017) 229 final, 10 May 2017, paragraph 13.

37 Ibid.

38 Stephanie Assad, et al. ‘Algorithmic pricing and competition: Empirical evidence from the German retail gasoline market’ (2020) CESifo Working Paper No. 8521; Ariel Ezrachi and Maurice Stucke,

‘Sustainable and Unchallenged algorithmic tacit collusion’ (2020) 17(2) Northwestern Journal of Technology and Intellectual Property

< http://dx.doi.org/10.2139/ssrn.3282235> accessed 29 July 2022, pages 247-249.

39 Michal Gal, ‘Algorithms as Illegal Agreements’ (2019) 34(1) Berkeley Technology Law Journal <

https://doi.org/10.15779/Z38VM42X86> accessed 4 July 2022, pages 73 and 81-90.

(19)

19 their ability to detect deviations from it and their ability to easily create a strong and credible threat of retaliation to deviations.40 This line of reasoning overlaps to a certain extent with the opinion of Ezrachi and Stucke on the matter. According to them, algorithmic tacit collusion is likely to occur in concentrated and transparent markets involving homogenous products where deviation from a supra-competitive equilibrium can be punished effectively and where the

reactions of outsiders (e.g. competitors and customers) should not jeopardize the expected results from the coordination.41 Furthermore, as they also point out there are already software vendors who promote their pricing algorithms as a way to avoid price wars and increase prices and margins.42 Another supported of the idea that algorithmic tacit collusion will become reality is Salil Mehra. According to him, this is due to the fact that the use of algorithms (which he refers to as ‘robosellers’) leads to an increase the accuracy in detection of price changes, greater speed in response to changes and a reduced irrationality in the setting of discount rates.43 Because of that, markets where algorithms are employed will produce supra competitive prices and the lower the number of competitors is, the higher the price level would be. Accordingly, algorithms would reach the highest possible supra competitive price in markets with only two competitors.44 However, at the same time, markets with more than two competitors would still yield supra competitive prices. Furthermore, in addition to this, there have also been experiments have been carried out by Calvano et al. with a type of reinforcement learning algorithms known as ‘Q- learning algorithms’. 45 According to them, the results show that ‘relatively simple pricing algorithms systematically learn to play collusive strategies’.46 These collusive practices were reached autonomously between the algorithms through trial and error, without communication between them, and without any instructions given to them to collude.47

40 Ibid 73 and 81-90.

41 Ezrachi and Stucke, ‘Sustainable and unchallenged algorithmic tacit collusion’ (n 38) 228 and 230.

42 Ibid 229.

43 Salil Mehra, ‘Antitrust and the Robo-Seller: Competition in the Time of Algorithms’ (2016) 204 Minnesota Law Review < https://scholarship.law.umn.edu/mlr/204 >accessed 9 July 2022, page 1340.

44 Ibid 1345.

45 Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolò and Sergio Pastorello,’Artificial Intelligence, Algorithmic Pricing, and Collusion’ (2020) 110 American Economic Review

<https://www.ftc.gov/system/files/documents/public_events/1494697/calzolaricalvanodenicolopastorello.

pdf> accessed 10 July 2022, page 3268.

46 Ibid.

47 Ibid.

(20)

20

4.3. Counterevidence

There have been, however, scholars who do not share the view that algorithmic tacit

collusion is a proven inevitability. For example, according to Veljanovski that there is currently there is not enough evidence that would definitively prove that algorithmic tacit collusion is possible.48 He states that currently, there is a lack of case-law on the matter, as well as a lack of experimental evidence. The reason for the latter is that the experiments conducted so far use models which present simpler conditions than those that would occur in reality.49 Furthermore, this idea that the prospects of algorithmic tacit collusion have not been empirically proven is also share by Dorner, according to whom, the evidence for the ability of algorithms to tacitly collude in real markets is scant.50

However, even though some scholars might strongly accept of reject the plausibility of the emergence of tacit collusion, the overall debate is in fact a nuanced one with many agreeing that as of now, there is an underlying uncertainty of what the answer is. For example, according to Schwalbe, algorithmic tacit collusion ‘is a possible outcome, but it is not as quick and easy or even unavoidable as it is often assumed in the legal discussion of algorithmic collusion’.51 Others, such as Abada and Lambin, have claimed that ‘while the evidence that algorithms can learn to earn supra-competitive profits is now overwhelming, the extent to which such seeming collusion will actually arise in practice remains an open question…’.52 Furthermore, the Furman Report commissioned by the UK concludes that it is hard to predict whether algorithmic

collusion will become a reality, however, it is an are with the potential to move fast where it is

48 Cento Veljanovski, ‘Pricing Algorithms as Collusive Devices’(2022) 53(4) IIC - International Review of Intellectual Property and Competition Law < https://doi.org/10.1007/s40319-022-01177-8 > accessed 24 May 2022, page 608.

49 Ibid 610-615.

50 Florian Dorner, ‘Algorithmic collusion: A critical review’ (2021) Institute of Science, Technology and Policy, ETH Zurich <https://arxiv.org/ftp/arxiv/papers/2110/2110.04740.pdf> accessed 29 July 2022, page 30.

51 Ulrich Schwalbe, Algorithms, ‘Machine Learning, and Collusion’ (2018) Institute of Economics, University of Hohenheim <http://dx.doi.org/10.2139/ssrn.3232631> accessed 24 May 2022, page 17

52 Ibrahim Abada and Xavier Lambin, ‘Artificial Intelligence: Can Seemingly Collusive Outcomes Be Avoided?’ (2020) <http://dx.doi.org/10.2139/ssrn.3559308> accessed 10 July 2022, page 6.

(21)

21 important to stay alert to potential harms.53 Moreover, it also issues a ‘strategic recommendation’

that the evolution of machine learning algorithms and artificial intelligence has to be monitored in order to ensure that it does not lead to anti-competitive activity or consumer detriment.,54 Lastly and quite importantly, in a joint report, the French and German NCAs have also

concluded that whether or not algorithmic tacit collusion has already occurred or will occur in practice is an open question.55

4.4. The role of this research in the wider context of academic research on EU competition law

Having in mind all the positions expressed by the different sides in the debate, the conclusion that can be drawn from them is that when discussing the plausibility of algorithmic tacit

collusion, the academic community is faced with uncertainty. The analysis of some scholars and the experiments they have conducted present compelling evidence as to why algorithmic tacit collusion is a plausible issue that can have substantially negative consequences for consumer welfare. However, at the same time not everyone is completely convinced of its inevitability.

Furthermore, another factor that adds to the uncertainty is that technologies such as pricing algorithms are likely to be developed by businesses rather than researchers due the incentives that they have in terms of obtaining profits from the sale of such technologies. However, since companies are not likely to easily disclose their inventions, the models on which the academic debate currently relies on have been developed by scholars as part of their academic research.

Because of this, it is theoretically possible that there is even a whole other ‘world’ of pricing algorithm R&D initiatives which are close to or have already managed to create pricing

53 Digital Competition Expert Panel, ‘Unlocking digital competition—report of the UK Digital Competition Expert Panel’ (2019)

<https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/78554 7/unlocking_digital_competition_furman_review_web.pdf> accessed 10 July 2022, page 15.

54 Ibid.

55ADLC and BkartA, ‘Algorithms and competition. Joint report of the Autorité de la Concurrence and Bundeskartellamt’ (2019) <https://www.autoritedelaconcurrence.fr/sites/default/files/algorithms-and- competition.pdf> accessed 10 July 2022, page 52.

(22)

22 algorithms capable of participating in tacit collusion which the relevant competition law

authorities have yet to detect.

The existence of such uncertainty in the topic of algorithmic tacit collusion presents the academic community with a choice between two approaches. First, it can adopt a passive position and refuse to discuss the potential negative consequences of algorithmic tacit collusion for the EU competition law framework and by proxy also for consumer welfare. This means that it will refuse to acknowledge the risks that the future might hold and ultimately be surprised if or when such risks become reality. Second, knowing that there is indeed a prospect for algorithmic tacit collusion to occur and having in mind the fact that technological progress in the past

decades has advanced in ways that one could have rarely imagined before, the academic community can take a proactive approach. Such an approach involves a forward-looking assessment of the potential negative effects that algorithmic tacit collusion would bring and the proposal of possible solutions. This would allow for the legal framework of EU competition law community to be prepared as early as possible for future challenges. As Pierre Schlag has stated

‘Courts have dockets. Legal academics have time.’ and because of this academic research is well suited to provide intricate analyses that contribute towards the creation of a well thought out solutions to new problems.56 In the current case, these contributions, by themselves, will also play a role in ensuring the vitality of the EU competition law framework due to the fact that they would aid the EU courts and legislator to take appropriate decisions that will promote its

effective development.57 Ultimately, precisely because of this, this thesis chooses to pursue the second approach due to the beneficial effects that it has for the development of EU competition law as a mechanism for protecting consumer welfare.

56 Pierre Schlag, 'Spam Jurisprudence, Air Law, and the Rank Anxiety of Nothing Happening (A Report on the State of the Art)' (2009) 97 Geo. L.J. < https://scholar.law.colorado.edu/articles/265> accessed 29 July 2022, page 822.

57 Rob Gestel and Hans- Wolfgang Micklitzage, ‘Why Methods Matter in European Legal Scholarship’, (2014) 20(3) European Law Journal <https://doi.org/10.1111/eulj.12049 > accessed 30 June 2022, page 294.

(23)

23

5. In search of solutions- the proposals of scholars for addressing algorithmic tacit collusion

The potential use of algorithms for the purposes of achieving and sustaining tacit collusion presents a disruption to the current legal framework of EU competition law. This is because in its current state, EU competition law focuses not on prohibiting, for example, the raising of prices itself but the way in which such raises occur. Namely, it prohibits communication between competitors that would result in anticompetitive effects. However, as has already been shown in this paper, the practice of tacit collusion does not require that such communication in the

conventional sense takes place. Because of this, the practice of tacit collusion is currently not prohibited under the EU competition law regime. If taken at face value and left unchanged, the logic behind this current state of the law is likely to leave algorithmic tacit collusion

unsanctioned. Consequently, the result of that would be that algorithmic tacit collusion would become a clever way for undertakings to participate in more widespread collusion

without actually falling under the scope of EU competition law prohibitions and bearing the consequences thereof.

Because of this possibility of an escape from liability, it is incumbent on researchers to develop and propose ideas for solutions that would serve to prevent and/or remedy the potential negative effects stemming from algorithmic tacit collusion. To this end, the purpose of this chapter is to outline solutions that have been presented by scholars as potentially suitable ways for tackling the challenges created by algorithmic tacit collusion.

5.1. A per se prohibition of specific types of algorithms

One of the main methods for tackling algorithmic tacit collusion that have been proposed is a per se prohibition on certain pricing algorithms and/or on pricing algorithms that possess certain properties that support supercompetitive prices.58 According to Harrington, such a prohibition could be illustrated with a simple example.59 If one imagines that the algorithms of

58 Joseph E Harrington, ‘Developing Competition Law For Collusion By Autonomous Artificial Agents’

(2018) 14(3) <https://doi.org/10.1093/joclec/nhy016> accessed 9 July 2022, page 350.

59 Ibid.

(24)

24 two undertakings have two choices of stable price levels. One price level is low and equal to the competitive price and the other is high and consequently supra competitive. Furthermore, the algorithms calculate the current price based on the previous prices set by their competitor’s algorithm. When the competitors set a low price, the algorithms would then also set the same low price or even a lower one, thus punishing their rivals for their conduct. As a result, this would ensure that prices are kept to high levels due to the fact that any lowering of the price levels will result in a punishment and a loss of profits. According to Harrington, a prohibition on algorithms which base their calculations on the past prices set by their rival would tackle this problem and reduce the risks of tacit collusion. This is because in such a case, undertakings would not be able to use algorithms that implement a reward-punishment scheme which would result in a

practically collusive outcome.60 However, it should be noted that Harrington admits that this example is simplistic in nature and ignores many of the challenges that would come with the implementation of such a ban on algorithms. Nevertheless, he still claims that it also provides an appropriate illustration of how a per se prohibition on certain pricing algorithms could work.61 Furthermore, Harrington also states that a per se prohibition on algorithms with specific

properties is possible because of a fundamental difference between cases of ordinary tacit collusion and algorithmic tacit collusion. Notably, while the collusive strategy for an ordinary tacit collusion is inside the heads of the human decision-makers, the strategy for algorithmic tacit collusion is contained in the code and is in principle observable.62 This also means that

evidentiary methods can be created for determining when algorithmic tacit collusion has indeed been reached by undertakings. To this end, Harrington proposes two types of tests- static tests and dynamic tests. Static tests entail an examination of the algorithm’s code without turning it on. In this way, the logic of the algorithm can be analysed in order to discern whether it possess properties that are per se prohibited. However, this approach is not applicable in the case of

‘black box’ algorithms which cannot be interpreted by humans. For such cases, dynamic testing can be used where the algorithm is put into a simulated environment in order to determine whether it is prone to participating in algorithmic tacit collusion.63

60 Ibid.

61 Ibid.

62 Ibid 349.

63 Ibid 355.

(25)

25

5.2. Liability of producers of pricing algorithms

Another solution that has been proposed in the literature is the establishment of liability not only for the undertakings participating in collusive pricing but also the undertakings that

programme and create the algorithms themselves. The proposal of such a solution is borne out of the nature of pricing algorithms and the structure of the industry surrounding them.64 Developing complex pricing algorithms that are based on ground-breaking technology requires a large amount of expertise. Because of this, such technologies are developed by specialized companies which then sell them to the undertakings that would directly participate in collusive pricing.

Therefore, if only the direct participants are fined, this would mean that a whole other category of relevant actors is ignored.65 Establishing liability for producers would in theory address the source of the problem as they are the creators of the algorithms and in some cases also work together with their customers in order to create a pricing algorithm that would be fit for their business operations. Furthermore, it will also lead to a dual-liability model which would reflect the responsibilities that both the producers and buyers have in designing their pricing algorithms.

Consequently, the result of this would be to incentivise companies that purchase pricing algorithms to demand that they do not lead to collusive outcomes and for the creators of such algorithms to take up these demands seriously.66 In this sense, thorough cooperation between producers and purchasers will not only be the result of good client-service but will also be stimulated by the law. What is interesting about this approach, is that there has already been a similar case. Namely, in AC-Treuhand v Commission, the ECJ has held that facilitators of horizontal agreements can be held liable.67 However, this does not mean that implementing it would be an easy task. In some cases, such as those where reinforcement learning is used, algorithms can learn price optimization rules by itself. Because of that it is harder to assess the fault on the side of the algorithm’s creators and users.68 To tackle this issue, Ezrachi and Stucke have proposed a strict liability rule where companies would be held liable regardless of their

64 Francisco Beneke, ‘Mark-Oliver Mackenrodt, Remedies for algorithmic tacit collusion’ (2021) 9(1) Journal of Antitrust Enforcement < https://doi.org/10.1093/jaenfo/jnaa040 > accessed 9 July 2022, page 166.

65 Ibid.

66 Ibid.

67 Case C-194/14 AC-Treuhand (2015) EU:C:2015:717, paragraphs 26- 47.

68 Beneke and Mackenrodt, (n 64) 167.

(26)

26 intentions or negligence.69 However, even though it could make enforcement easier, it will likely lead to overdeterrence that would disincentivise businesses to use pricing algorithms and will erase the efficiencies brought by them.70

5.3. Algorithmic tacit collusion as a type of concerted practice

One common trait of all proposals that have been outlined so far in this chapter is that they require the creation of completely new concepts in EU competition law in order to bring

algorithmic tacit collusion under its scope. However, not all proposals subscribe to this approach.

According to Luca Calzolari, one of the ways in which EU competition law can respond to the emergence of algorithmic tacit collusion is by extending the scope of the already existing concept of concerted practices.71 This could be done in two steps.

First, it has to be established that undertakings are to be held liable for the decisions that their autonomous pricing algorithms take. To this end, Calzolari submits that intent and imputability play a very limited role in the context of EU competition law because it establishes a quasi-strict liability regime when dealing with antitrust offences. According to him the case-law of the CJEU contains ‘virtually no reference’ to the role of intent for the application of articles 101, 102 and 106(1) TFEU. 72 Furthermore, he also points out that there is a regime for strict liability for corporate groups that is rooted in the concept of ‘undertaking’.73 This is because the rules of EU competition law prescribe that when a subsidiary is wholly owned by its parent company, there is a rebuttable presumption that the latter is liable for any anticompetitive conduct carried out by the former. Moreover, according to the case-law of the CJEU, parent companies have a

69 Ariel Ezrachi, and Maurice Stucke, ‘Two Artificial Neural Networks Meet in an Online Hub and Change the Future (Of Competition, Market Dynamics and Society)’ (2017) Oxford Legal Studies Research Paper No. 24/2017 < http://dx.doi.org/10.2139/ssrn.2949434> accessed 9 July 2022, pages 39- 41.

70 Beneke and Mackenrodt (n 64) 167.

71 Luca Calzolari, ‘The Misleading Consequences of Comparing Algorithmic and Tacit Collusion:

Tackling Algorithmic Concerted Practices Under Art. 101 TFEU’ (2021) 6 European Papers <

https://www.europeanpapers.eu/en/e-journal/misleading-consequences-of-comparing-algorithmic-and- tacit-collusion> accessed 29 July, page 1211.

72 Ibid 1212 and 1213.

73 Ibid 1215.

(27)

27 responsibility to ensure their subsidiaries are complying with the competition rules.74

Consequently, according to Calzolari, much like in the case of parent companies and their subsidiaries, strict liability may and should also apply in situations where autonomous pricing algorithms take decisions to collude even if they were not instructed to do so.75

Second, the concept of concerted practices is extended to include a scenario where two undertakings use algorithms in order to reach and sustain a supra competitive collusive price.

In order to explain how this can be the case, Calzolari starts by outlining that a concerted practice occurs when undertakings knowingly substitute the risks of competition for practical

cooperation, which culminates in a situation which did not correspond to the normal conditions of the market. Then he states that by instructing algorithms to monitor market conditions (e.g.

prices), they are reducing the uncertainty about the current and future behaviour of their

competitors. Furthermore, since many if not all undertakings on a given market are likely going to be using similar pricing algorithms, it can be concluded that by doing so, they are replacing uncertainty (i.e. risks of competition) with knowledge.76 Because of this Calzolari submits that article 101 TFEU does capture algorithmic tacit collusion since it falls under the concept of concerted practices ‘or at the very least, because algorithms’ monitoring activities represent a mechanism of (public or private) exchange of information among competitors’.77 Furthermore, he also states that when two pricing algorithms reach a collusive price level, there ought to be a rebuttable presumption of an existence of a concerted practice. This means that the Commission would not need to prove the undertakings’ intentions to conspire, and it would be for the latter to prove that in that specific case a concertation did not occur, or that their conduct fulfils the conditions of article 101(3) TFEU.78

74 Ibid 1218.

75 Ibid.

76 Ibid.

77 Ibid

78 Ibid 1212.

(28)

28

5.4. Limitation in the algorithms’ speed as a potential remedy

Another solution that has been proposed strikes directly at one of the main features that make pricing algorithms so appealing for businesses. Namely their ability to change prices at an

incredibly high speed. As has been established in section 3.3, the speed at which prices can be used for the practice of price-signalling through which a supra competitive price levels would be achieved. Furthermore, in cases where for one reason or another, a collusive price level has been achieved, pricing algorithms can eliminate the possibility of a deviation from it by reacting and lowering prices seemingly instantaneously, thus annulling the benefits of such deviation.

Because of this, it has been proposed that this can be solved through an introduction of a limit on the speed at which pricing algorithms so that they are able to change the price of a given product at certain intervals.79 While such an approach would present a simple, yet somewhat crude solution to the problem, it can also be criticized due to the fact that it would create significant inefficiencies for companies that could ultimately impact their profitability.80 However,

according to Bertrand and Dewenter, there is way to solve the concerns of these inefficiencies.81 It entails the adoption of what they refer to as ‘lags in trading activities’. To this end, they

provide an example of how in 2016 a national stock exchange for US equities introduced a delay in trading activities of 350 microseconds. While such a delay is not observable by humans, according to the authors, it can hamper price coordination and as a result, small delays or

restrictions in the rate of price changes can affect coordinated behaviour without deterring future technological advances.82

5.5. Market inquiries as facilitatory tools

As can be seen in this chapter, the solutions proposed by scholars mostly focus on directly addressing the anticompetitive effect of algorithmic tacit collusion. Generally speaking, however, EU competition law provides for more types of policy responses than only remedies.

79 Lea Bernhardt & Ralf Dewenter, ‘Collusion by code or algorithmic collusion? When pricing algorithms take over’ (2020) 16(2-3) European Competition Journal

<https://doi.org/10.1080/17441056.2020.1733344 > accessed 9 July 2022, pages 338 and 339.

80 Ibid 338 and 339.

81 Ibid 339 and 340.

82 Ibid.

(29)

29 According to article 17 of Regulation 1/2003, can carry out a market inquiry into sectors of the economy and types of agreements. In doing so it can require the undertakings or associations of undertakings concerned to provide information about all of their agreements, decisions and concerted practices. Additionally, pursuant to articles 18, 23 and 24 of the same regulation, it can also take statements, conduct inspections and even impose fines for the provision of incorrect information.83 In light of this, according to Beneke and Mackenrodt, the European Commission can conduct such an inquiry as part of its strategy to deal with algorithmic tacit collusion. 84 This is because doing so will provide valuable information on what types of algorithms are used, to what degree are they used in markets and what are their effects on competition. Subsequently, this information can serve an important facilitating role in the context of a general strategy for tackling algorithmic tacit collusion by informing future enforcement action.85

6. Regulation as another instrument in the toolbox

When one acquaints themselves with the literature on how of algorithmic tacit collusion should be approached, there are two common trends that can be identified amongst the solutions that have been proposed thus far.

First, a substantial amount of the academic writings (such as articles, book chapters, etc…) focus on presenting one or two remedies. While such an approach allows the authors to explain their solution in detail, it fails to present a holistic picture of how the problem can be tackled as a whole. For example, in some instances it is possible that the weaknesses of one remedy may be compensated by the parallel introduction of another remedy. Furthermore, it is also possible that the effects of two remedies might overlap to a significant extend, thus nullifying the need for their parallel introduction, or that their effects negatively affect each other, thus leading to ineffective enforcement. Because of this, in order for an academic piece of writing to be able to inform future policymaking, it is best if it presents a solution that takes into consideration and discusses a combination of a variety of remedies.

83 Council Regulation (EC) No 1/2003 of 16 December 2002 on the implementation of the rules on competition laid down in Articles 81 and 82 of the Treaty [2003] OJ L 1/1, articles 17,18,23 and 24.

84 Beneke and Mackenrodt (n 64) 173 and 174.

85 Ibid.

Referenties

GERELATEERDE DOCUMENTEN

Wie dat advies erbij pakt, leest daarin dat ‘een Tweede Kamer die is voortgekomen uit latere verkiezingen dan die welke mede waren ingegeven door de ontbinding ex artikel 137’

In die gewone omgang is die term ‘beurtkrag’, bekend as ‘load shedding’ in Engels, wat deur Eskom ingevoer is, summier verwerp.¹⁷⁹ Daar is verduidelik dat dit

When Enlightenment premises were declared dead in the Cold War era, many social scientists such as Marshall and Parsons, in search for a place for social science during the Cold War

De verschillen tussen de planten opgekweekt in &#34;droge&#34;, normaal vochtige en natte potten waren vrij groot. De planten opgekweekt in &#34;droge&#34; omstandig­ heden

Wel worden, naast sulfiet ook natriumhydroxide en water aan het ontzwavelingsadsorbens toegevoegd; het natriumhydroxide verlegt enerzijds het chemisch evenwicht van de reactie

Gebleken is dat de teelt- en oogstkosten voor voederbieten circa 400 gul- den per ha hoger zijn dan voor mais.. Dit wordt met name veroorzaakt door de hoge kosten van het zaaizaad en

De ammoniakemissie in de afdeling voor vleesvarkens met een bolle dichte vloer voor in de hokken en alle roosters van metalen driekant, was gemiddeld over vijf ronden 1,9 kg

De conclusie is dat klimaatadaptatie niet gebaat is bij een te strakke integratie van dit beleid in andere beleidssectoren (zoals water en ruimtelijke ordening) op