‘Market’ definition in ecosystems

Abstract

Despite the popularization of the concept of ‘ecosystems’ among academics and practitioners, ecosystems have so far failed to become a coherent unit of analysis in competition law. One of the reasons is that it is not yet apparent when ecosystem analysis should be preferred over standard market-based analysis, and we also lack empirical methodologies to define their boundaries. In this article we first describe contexts where formal definition of ecosystem boundaries (similar to market definition) is necessary or at least desirable to achieve quality analysis, and then describe methodologies on how to achieve that (hierarchical clustering, K-means clustering, network of complementarities, factor analysis, and snowball selection). Furthermore, we provide guidance on when each of the methodologies is pertinent. We describe the limitations of the substitutability, aftermarkets, and connected markets approaches in cases in which ecosystems are the main unit of analysis. Finally, we provide four examples of how, in such cases, these methodologies presented can be useful to practitioners: i) market inquiries, investigations, and pro-competition interventions; ii) designation of undertakings as having strategic market status; iii) conglomerate mergers and iv) calculation of fines and quantification of harm/effects.

1. INTRODUCTION

In recent years, there has been an increasing interest in ecosystems in the business, economics, and legal literature. The concept of ecosystems in the social sciences originated in the work of James Moore on business ecosystems, where the goal was to describe economic activity that was neither confined within the firm nor dispersed across the market, but rather managed by a firm and effectuated through a collection of surrounding firms.¹ Similar organizational forms were later explored through concepts such as keystone firms² and platforms,³ but it was not until the rise of digital markets that ecosystems emerged as a natural way to describe the economic activity that develops around firms that orchestrate the creation and provision of products and services to users (chiefly Big Tech). As such orchestrator firms grew more powerful and started attracting the attention of competition authorities, the concept of ecosystems infiltrated the vernacular of enforcers, courts, and regulators. Legal decisions,⁴ market inquiries,⁵ and legislative materials⁶ increasingly rely on the concept of ecosystems to refer to vaguely defined agglomerations of economic activities coordinated by a central actor. The culmination of ecosystems’ significance in competition law was their inclusion in the Commission’s new guidelines on market definition in 2024, where the Commission recognizes ecosystems as a distinct organizational form that may need to be defined as an aftermarket or bundle.⁷

Yet, despite the popularization of the term, ecosystems have failed to become a coherent unit of analysis in competition law. Fundamentally, if a given ecosystem—as a collection of interdependent products, services, economic activities, and actors—is to serve as the area within which competition law analysis is performed, one would have to be able to define the limits of that area. We posit that the hitherto failure to do so helps explain the lack of wider adoption of ecosystems in competition law analysis. It is impossible to conduct any kind of competition law investigation—whether Article 101/102 TFEU, merger control, or market inquiries—without first scoping out the limits of what products, services, actors, and business practices will be included in the investigation. While the boundary definition exercise does not always have to be exhaustive and precise (notably, anti-competitive agreements/restraints investigations have lower market definition requirements),⁸ competition authorities should be able, to the extent needed (and we identify below where this is the case), to demarcate the boundaries of their investigations, wherein they will look for actual or potential competitors, competitive effects, and remedies. We currently lack the tools to perform this exercise in ecosystems. The new Guidelines on Market Definition treat ecosystems as aftermarkets or bundles, which are organizational structures that only partially capture the breadth and complexity of ecosystems, and the case law attaches very specific conditions to the function and use of aftermarkets and bundles.⁹

As business ecosystems proliferate and grow in value, and as antitrust challenges are gathering pace, a more methodological approach will become necessary. Already, it is interesting to note that the references to and use of the construct of ecosystems in the early Commission cases (Google Android, Amazon Marketplace, Google/Fitbit)¹⁰ were far more tangential than in the Commission’s more recent Apple App Store Practices decision and the General Court’s also recent decision in Google Android,¹¹ which shows increasing reliance on the concept of ecosystems and its utility. There are situations where more precise, detailed, and rigorous mapping of ecosystems, such as that delivered by the methodologies presented herein, will be indispensable (see “The need for and suitability of formal definition of ecosystem boundaries” in Section 2). For instance, for the exercise of market inquiries/investigations where authorities need to map the market more comprehensively (as the Competition and Markets Authority (CMA) did for mobile operating ecosystems, which we discuss in Section 4),¹² a methodologically rigorous carving out of which ecosystems are present in the market, what they include, and how they are structured is essential for accurate analysis.

At present, it is the concept of market that serves as the delimiting tool in traditional competition law analysis. As the General Court explained in Mastercard, ‘[f]or the purposes of Article [102 TFEU] the proper definition of the relevant market is a necessary precondition for any judgment as to allegedly anti-competitive behaviour, since, before an abuse of a dominant position is ascertained, it is necessary to establish the existence of a dominant position on a given market, which presupposes that such a market has already been defined.’¹³ Similar scoping exercises are used in cases involving anti-competitive agreements, merger control, and market inquiries/sector investigations as well.¹⁴ Therefore, such boundary-setting exercises serve both as a necessary focusing tool and as a necessary measuring tool: a focusing tool in that it defines the area in which a competition law investigation is conducted, and a measuring tool in that it helps competition authorities assess and calculate the relative power of the actors falling within the set boundaries.¹⁵ One can reasonably expect that as ecosystems are increasingly regarded as a useful organizational form for competition analysis, the demand for a more precise methodology around determining their scope and structure will concomitantly become more central too.

We know of no study that describes how ecosystem boundaries should be set so that ecosystems can effectively delineate competition law investigations. While some literature exists on market definition in ecosystems, that line of research mostly discusses essential considerations for ecosystem definition rather than actual methodologies on how to perform it.¹⁶ The remaining existing literature on ecosystems and related structures is predominantly concerned with the concept of ecosystems and their necessity in competition law.¹⁷ It is motivated by the observation that the intricate, co-dependent, non-hierarchical relationships between products, services, and actors within ecosystems cannot be captured by the concept of the antitrust market, which predominantly relies on substitutability between products and/or services. The thinking is that in the economic order of ecosystems, products, services, and actors transcend markets, and they do not necessarily only compete with each other in a zero-sum game. But once one concedes to this argument and sees the validity of ecosystems as a unit of organization of economic activity (as is increasingly happening—see notes 4–6), there is no accepted methodology on how to reconstruct the ecosystem(s) within which competitive dynamics and potential competition law offences will be investigated.

We aim to fill that gap by presenting and synthesizing different methodologies on ecosystem definition, drawing partly from the scarce legal literature in the area,¹⁸ but mostly from the broader field of social sciences. It is impossible to present the methodologies in exhaustive detail. However, we take the first steps in that direction, and we provide initial guidance to demonstrate the relevance, applicability, and utility of the methodologies in ecosystem definition. By doing so, we contribute to operationalizing the use of ecosystems in competition law analysis, helping authorities, regulators, and courts identify tools to define the boundaries of ecosystems, which, similar to market definition, is most commonly the first step in any investigation. As an ancillary outcome, some of our discussed methodologies also reveal the structure of the studied ecosystem(s)—not only their boundaries. Once ecosystems are correctly drawn, investigators can more confidently determine competitive dynamics and detect competition distortions within them. As a result, our study helps minimize arbitrariness in ecosystem analysis and enables ecosystems to become the main unit of analysis when the economic activities investigated are structured as an ecosystem. We do not take a position on when investigators should rely on the concept of ecosystems (as opposed to markets)¹⁹; rather, we note that when the ecosystem is deemed to be the preferred unit of analysis, our study materially enhances the delineation of their boundaries and, by extension, the competition analysis to be performed thereupon.

The rest of the article is structured as follows. Section 2 shows why the traditional substitutability, aftermarkets and connected markets approaches to market definition are insufficient when doing competition analysis of ecosystems and presents situations where a formal definition of ecosystem boundaries is necessary or, at least highly desirable in achieving accurate analysis. Section 3 presents methodologies to define an ecosystem’s boundaries; in particular, it focuses on four types of methodologies: clustering analysis, network of complementarities analysis, factor analysis, and snowball selection process. Section 4 synthesizes these methodologies and discusses when each of them is appropriate. Section 5 discusses the limitations and criticisms of our proposed approach and invites future and additional research. Section 6 concludes.

2. THE INSUFFICIENCY OF CURRENT APPROACHES AND THE NEED FOR FORMAL ‘MARKET’ DEFINITION IN ECOSYSTEMS

The insufficiency of the substitutability approach, aftermarkets, and connected markets

The meaning of ‘ecosystem’

There is a myriad of definitions of the term ‘ecosystems’ and related terms.²⁰ The literature has not converged to a common all-encompassing definition. This is not problematic, as divergent definitions might be useful to analyse different aspects of ecosystems from various disciplinary points of view. However, common features emerge across definitions. The objective of this article is neither to provide a novel or exhaustive definition of ecosystems nor to compare the existing ones. We instead focus on the two main common features that are present across all definitions, and which sufficiently capture the concept of ecosystem for the purposes of defining boundaries around them for competition analysis.

For the purpose of this article, and in accordance with most of the existing definitions in the literature, ecosystems are made of a set of interdependent products and/or services that jointly create value for end customers through designed complementarities. We can distinguish two types of ecosystems. Multi-actor ecosystems are organizational forms in which legally separate firms coordinate to create complementarities between their products or services. Accordingly, in multi-actor ecosystems, firms are said to work together to produce a ‘joint value proposition’.²¹ Multi-product or ‘experience’ ecosystems, in turn, are a collection of products and services offered by a single firm (or firms under common ownership/management) that exhibit complementarities. This is, for example, the case of the Apple ecosystem. In practice, multi-product and multi-actor ecosystems can be combined, which is commonly observed in digital ecosystems. The depiction is mirrored by the General Court in Google Android, where it described a (digital) ecosystem as an organizational structure ‘which brings together several categories of supplier, customer and consumer and causes them to interact within a platform, the products or services which form part of the relevant markets that make up that ecosystem may overlap or be connected to each other on the basis of their horizontal or vertical complementarity. Taken together, the relevant markets may also have a global dimension in the light of the system that brings its components together and of any competitive constraints within that system or from other systems’.²²

Within an ecosystem, firms produce different products and let consumers choose which ones to consume. Firms may also bundle some of them, as in Android and Google Play. This enhances the value of all the products in the ecosystem, and notably, the core product around which complements are built. For example, some camera manufacturers give consumers the possibility of combining their cameras with third-party lenses. Instagram, in turn, lets third parties produce complementary products. This makes Instagram more attractive to its users, which in turn increases the direct network effects between end users on this platform. Moreover, Instagram allows individuals to buy products on its platform by linking their accounts to third-party payment methods such as bank cards or digital payment methods like Google Pay and Apple Pay. Not only does this boost Instagram’s end users’ direct network effects, but it also intensifies indirect network effects between end and commercial users in Instagram and third-party payment methods’ platforms. Third-party apps such as Later allow planning and scheduling of Instagram posts in advance, as well as analysing performance and tracking hashtags. The photo and video editing app Visual Supply Company (VSCO) provides a range of filters and tools to enhance Instagram content. Instagram also gives users the possibility of re-posting their stories on Facebook. The popular Internet browser Chrome allows third parties to develop add-ons that enhance the user experience when using Chrome. Videogame producers coordinate through technical specifications to produce a game that works with a given console sold by another firm. In doing so, they make the game and the console complementary. Consumers, in turn, can decide on the bundle of console(s) and game(s) they want to purchase. At the same time, the more and higher-quality games that can be played with a console, the more valuable the console (i.e., the core product) becomes to consumers and game producers. In other words, the designed complementarities between the console (i.e., the platform) and the games boost indirect network effects between consumers and game producers. Importantly, as illustrated by these examples, firms need to coordinate for a joint value proposition to emerge. In the previous example, the camera producer needs to share some technical specifications with third parties so that they can produce compatible lenses. Instagram, in turn, has open application programming interfaces (APIs) and provides software development kits (SDKs) so that third parties can create complementarities with its core product, the Instagram platform. Then, in ecosystems, while “consumers have a say in the choice of complements” firms provide “the contours of free choice’.²³

Limitations of the substitutability approach

The fact that, in ecosystems, firms coordinate to generate complementarities between their products requires adapting the way in which market definition is carried out. In traditional markets, a firm’s competitive pressure only comes from ‘pure’ rivals producing substitute products. Therefore, the objective of the traditional (demand) ‘substitutability approach’ to market definition is to define a group of products that consumers consider to be substitutable.²⁴ The rationale behind this approach is twofold: (i) only firms producing substitutable products exert competitive pressure on each other and (ii) the welfare effects of conduct or structural changes that affect how and how much firms compete (e.g., a merger, anti-competitive behaviours, a regulatory intervention) are circumscribed to the firms producing substitute products, immediately upstream and downstream firms, and the consumers buying these products.²⁵ In this case, defining a single relevant market (and, eventually, a vertically-related market) in terms of substitutes is a pertinent and sufficient approach to identify the relevant agents and the scope of the analysis when doing competition analysis.

Conglomerate effects and ecosystems

The economic theory and case law on conglomerate theories of harm in mergers go beyond the substitutability approach in that they consider that firms that do not produce substitutes may exert competitive pressure on each other and hence that the welfare effects of a merger may affect how much firms compete. In this respect, conglomerate effects can accommodate to some extent the new challenges posed by ecosystems to competition authorities that the substitutability approach alone cannot meet. In particular, organizing complementary products within a digital ecosystem has been identified as a theory of harm resulting from conglomerate mergers. Post-merger, putting two products within the same ecosystem “may generate entry barriers by requiring firms to simultaneously enter several markets, or lead to co-ordinated effects and the softening of price competition by increasing symmetry and multi-market contact among firms”.²⁶

Conglomerate effects are useful for performing competition analysis involving ecosystems in that they provide lenses to look at competitive effects spanning across product markets as defined by the substitutability approach. At the heart of these theories is the concept of complementarity, which is also pivotal to ecosystems. However, this does not rule out the need for alternative methodologies to define the scope of competition analysis when ecosystems are to be considered for various reasons. Firstly, the investigator might be interested in analysing within-ecosystem conduct such as abuse of dominance between independent firms, not only mergers. Secondly, as pointed out above, in any analysis involving ecosystems, including conglomerate effects, methodologies allowing to more accurately define which products are to be considered to be part of the ecosystem can only bring rigour to the analysis. Thirdly, conglomerate theories of harm rely substantially on bundling and tying. While this certainly takes place in ecosystems, there are other competitive pressures and potential competitive harms of interest to competition authorities that span across antitrust markets in ecosystems. Below we provide some examples.

Competitive pressures across antitrust markets in ecosystems

Within an ecosystem, when firms choose to cooperate by creating complementarities between their products, they expose themselves to another form of competition: competition for value capture. Therefore, in ecosystems, coopetition, rather than pure rivalry, is the norm.²⁷ For example, app stores and app developers cooperate within the app ecosystem. However, they also compete for the revenue generated by this cooperation. This competition manifests in the commissions on apps and in-app purchases, as well as on non-price strategies, such as app stores requiring the use of their in-app purchase systems. The Epic Games/Apple case²⁸ is a good illustration of how firms cooperating within the same ecosystem can exert competitive pressure on each other despite being located in different relevant markets. Epic Games is a videogame developer behind some of the most popular and lucrative games available at Apple’s iOS App Store. Epic Games contested Apple’s 30 per cent commission fees and restriction from allowing consumers to use alternative purchasing methods to Apple’s for in-app purchases. In 2020, it introduced changes to its flagship game, Fortnite, to allow users to make in-app purchases directly through Epic Games, bypassing Apple’s walled garden ecosystem. Apple responded by blocking Fortnite from its app store, which in turn led Epic Games to an antitrust lawsuit. This dispute shows how a major complementor (Epic Games) can exert competitive pressure on the orchestrator (Apple) to lower its margins on the profits generated by the firms participating in the ecosystem.

In the same vein, anti-competitive behaviours can take place between two firms located in different relevant markets within the same ecosystem. Graef introduces the concept of ‘hybrid differentiation’ to define a “conduct whereby a platform differentiates among non-affiliated services in an effort to favor its own business”.²⁹ She uses the example of Google’s intention to remove the app Unlocked from the Play Store in 2018 to illustrate it. This app shows users advertising or other content when they unlock their phones and gives users points they can exchange for mobile credit, data, entertainment, or loyalty points. Although Google does not compete in markets in which Unlocked is present by selling a substitute (nor does it seem to intend to), it used its gatekeeper power in the app store market to exclude Unlocked from the Play Store because the app makes it harder for Google to monetize its activities through advertisement.

Moreover, for a firm, other firms located in adjacent markets within the same ecosystem can be active and potential competitors.³⁰ The literature on platform envelopment shows that sharing common components and having overlapping user bases increase the probability of a platform entering the market of another complementary platform.³¹ These two conditions are met in digital ecosystems. Therefore, complementors may pose competitive constraints to the extent that they are potential competitors.

As these examples illustrate, within ecosystems, competitive pressures and their concomitant welfare effects span across several relevant markets. This, however, does not imply that firms do not face competitive pressures from other firms producing substitute goods inside or outside of their ecosystems. It adds another type of competitive pressure coming from within-ecosystem coopetitors that cannot be identified by focusing on the degree of substitutability. Therefore, ecosystem analysis does not replace substitutability analysis; it complements it when the studied economic activity bears characteristics akin to ecosystems as described above. By defining an ecosystem, one can analyse how different relevant markets defined through the substitutability approach are interrelated through coopetitive relations. Therefore, when analysing conduct or a structural change (e.g., a merger, anti-competitive behaviours, a regulatory intervention) in a given market, one has to study how this will affect other related markets within the ecosystem through changes in the degrees or natures of competition and cooperation with firms located in other markets of the ecosystem.³² Moreover, when ecosystems are made the unit of analysis through a coherent definition of boundaries, one can then study how ecosystems compete with each other,³³ not just how their constituent parts (i.e., products, services, actors) compete among each other, or across other ecosystems. However, antitrust authorities and other practitioners usually lack the methodological tools to do this. In the remainder of the article, we present and discuss the pertinence of the use of several methodologies to define the boundaries of an ecosystem.

Limitations of aftermarkets and connected markets

Naturally, to overcome some of the above limitations, the case law has developed methods that ‘expand’ the methodology of defining single markets based on substitutes. The concept of aftermarkets was initially developed in Eastman Kodak to link a main market (photocopiers) to a secondary consumables market (repair parts).³⁴ Even though Kodak did not have market power in the primary market, it could monopolize the secondary market by locking in customers. By linking two non-substitute markets together, the court could develop a theory of harm. While such linking-together of markets resembles a rudimentary ecosystem structure, the doctrine of aftermarkets insufficiently reflects ecosystem structures because it is limited to two markets and requires the secondary product (market) to be a consumable which is purchased with a time delay such that consumers may not be abler to fully factor in the entire lifetime of a transaction. In its new Market Definition Guidelines, the Commission broadened the scope of aftermarkets to include any system consisting of a primary and secondary product and allows for market definition either as a single ‘system’ market or as two separate but inter-dependent markets.³⁵ The Commission refers to those arrangements as bundles or ecosystems, but does not give any indication as to the conditions under which the system or separate markets approach should be preferred. Nor does it provide much detail in terms of how to define a bundle or ecosystem.³⁶

In the European Union (EU), Tetra Pak gave rise to the related concept of connected markets, whereby conduct in a market where the undertaking is not dominant can have effects in the market where the undertaking is dominant.³⁷ Recognizing that normally the “application of Article 86 [102 TFEU] presupposes a link between the dominant position and the alleged abusive conduct, which is normally not present where conduct on a market distinct from the dominated market produces effects on that distinct market”,³⁸ the Court of Justice of the European Union(CJEU) required ‘special circumstances’ to be at play for a theory of harm between connected markets to work.³⁹ In Tetra Pak, the special circumstances were both the close affinity between the two markets and Tetra Pak’s superdominance in the dominated market.⁴⁰ While these conditions may be found in some ecosystems, the requirement for special circumstances makes the application of connected markets an extremely limited tool that cannot describe the complexity and multifariousness of ecosystems.

In all, the market definition exercise still revolves largely around substitutability in a single market. While concepts such as bundles and ecosystems have entered the debate, they are rarely used and with no analytical consistency. We proceed below to provide some analytical depth in defining ecosystems.

The need for and suitability of formal definition of ecosystem boundaries

Our argument is not that online products or services in digital markets always have to be defined as an ecosystem when there are interdependencies. Indeed, in many cases, while competition authorities acknowledged the existence of an ecosystem in a market, they did not use it as the unit of analysis and relied on traditional market definition instead. For example, in the high-profile approval of the Microsoft/Activision merger, the UK CMA repeatedly referred to the Microsoft gaming ecosystem, but only summarily and arbitrarily described what it includes. It eventually used it only to make a vague point: “the combination of Microsoft’s multi-product ecosystem gives it a stronger position in cloud gaming than would be suggested by assessing each of its products and services individually”.⁴¹ For the actual analysis, it relied on the traditional market definition as performed in merger analysis. Similarly, in the recent Apple App Store Practices case against Apple, the Commission delineates Apple’s ecosystem as consisting of Apple mobile devices, iOS, Apple App Store, and Apple/third-party apps, each of them in a vertical relationship with the previous layer, without any further detail as to how it determines these layers, structure, and relationships.⁴²

It may well be the case that, so far, authorities have considered that a fully detailed plotting and description of the ecosystems in question was not necessary in these cases. The Apple App Store Practices case concerned a very specific anti-steering provision, as it applied only in the context of Apple Music and its competitors.⁴³ The Microsoft/Activision merger, while sizeable in monetary value, only concerned a narrow subset of Microsoft’s overall ecosystem.

But there are situations where a formal definition of ecosystem boundaries (an ‘ecosystem definition’) will make a difference to the quality of the analysis, and so the ecosystem (rather than the market) as the main unit of analysis should be preferred. We propose four contexts where ecosystem definition has the potential to result in a more precise analysis, as precision is necessary to administer effective and substantiated enforcement and regulation. There are areas beyond those four where formal ecosystem definition can be required, but the state of antitrust development is not there yet. For example, the Greek competition law reform committee considered a proposal for a new provision on abuse of dominance in ecosystems, where the definition of an ecosystem, rather than the definition of a market, would set the scope of the analysis; the proposed Article 2A read: “Without prejudice to Articles 1 and 2 of this law [the “standard” abuse of dominance offence], it shall be prohibited for an undertaking to abuse its dominant position in an ecosystem of paramount importance with regards to competition in the Greek dominion”.⁴⁴ While the proposal did not make it into the final law, it serves as an example of potential future directions where formal definition of ecosystem boundaries will be required similar to how formal market definition is required to establish an offence. In the meantime, formal ecosystem definition already plays a key role in the contexts we analyze next.

Market inquiries, investigations, and pro-competition interventions

Market inquiries and investigations look into market segments for factors that distort competition, whether attributable to firm conduct or relating to general features of the market.⁴⁵ A new and similar tool is the Digital Markets, Competition and Consumers Act’s (DMCCA) pro-competition intervention, which allows the CMA to investigate factors that have an adverse effect on competition and then impose remedies to correct them.⁴⁶ While there is usually no obligation to perform formal market definition of the economic segment that is investigated, some methodology to carve out the outer boundaries of the investigation is still needed. In market segments that are structured as ecosystems (see “The insufficiency of the substitutability approach, aftermarkets, and connected markets” in this Section), a formal definition of ecosystem boundaries can help in two ways: (i) the methodologies which we present below that work with an open-ended population can ensure that all relevant types of actors (and the population within) are included in the investigation. For instance, if an authority wants to request information, send out questionnaires to competitors or complements, or make sure that it considers all types of potential competitors or complements, a formal ecosystem definition can provide much higher accuracy compared to an arbitrary designation of an ecosystem or market segment under study; (ii) the methodologies we present below that reveal how a selected population can be broken down into different ecosystems or reveal the internal organization of an ecosystem contribute to discovering the structure of the industry studied in the market investigation. This can help authorities better understand the industry, detect bottlenecks, and determine the appropriate course of action. While a formal methodology is not necessary to identify inter-relations between industry actors, it provides more confidence in the results and legitimizes authorities’ actions. For instance, we discuss in Section 4 how the CMA’s market study in mobile OS ecosystems blurs value chain layers, which a formal application of our methodologies would prevent. It also excludes actors such as advertisers or payment intermediaries, which weigh in heavily in the mobile ecosystem, and which our proposed methodologies would catch.

Designation of undertakings as having strategic market status

An increasing number of jurisdictions are adopting tools and instruments that allow them to designate undertakings as having strategic market status, which can then be used to impose obligations or prohibitions without having to find that they have technically abused their dominant position in the market. For example, two of the criteria the DMCCA uses is that the undertaking’s position in the relevant digital activity can allow it to extend its market power to other activities or to substantially influence the conduct of other undertakings.⁴⁷ Article 19a of the German Competition Law uses similar criteria to designate undertakings having ‘paramount significance for competition across markets’.⁴⁸ These descriptors of ‘strategic market status’, ‘significance across markets’, or similar are clear references to an ecosystem approach, since they place the interdependence and complementarity of undertaking activity at the core of their designation to that special category. Again, while a formal market (or ecosystem) definition is not required by law for such designations, properly-defined methodologies legitimize authorities’ designation decisions, especially given that these designation decisions are subject to judicial review. Competition authorities would then have to show that their decisions were not unreasonable or arbitrary within the standard of legal review applied in each jurisdiction, and formal methodologies would help them meet their legal standard.

Conglomerate mergers

Conglomerate theories of harm rely substantially on bundling and tying. While other conglomerate theories of harm exist and are gathering increasing interest, they are rarely used in merger analysis. This might be in part because doing so would require an ecosystem definition where market definition is difficult and/or insufficient to assess the theory of harm or the merger efficiencies.⁴⁹

Conglomerate mergers in digital markets may generate demand-side efficiencies by facilitating the joint use of complementary products (eg, single user profile, seamless content sharing across services, etc), offering users a common ‘look and feel’ across different digital services or establishing a one-stop-shop for multiple products.⁵⁰ They may also generate supply-side efficiencies through economies of scope in data aggregation and reuse across markets.⁵¹

The above-mentioned drivers of conglomerate merger efficiencies may also erect entry barriers by requiring entrants to simultaneously enter multiple markets to challenge the incumbent. They can also relax price competition by increasing multimarket contact. Another harm may come from a firm acting as an access point from which consumers can be steered to consume within-ecosystem products.⁵² For example, Alphabet, which integrated YouTube and Google Flights through acquisitions, has been suspected of giving more prominence to these two platforms over rival ones in Google Search results.⁵³ This concern has been raised by multiple competition policy reports⁵⁴ and by the Digital Markets Act with its prohibition of self-preferencing by gatekeepers. Finally, because of all the above-mentioned practices that facilitate users staying within a single ecosystem, conglomerate mergers might result in the elimination of potential competition (harm listed in merger guidelines from jurisdictions such as Germany, Japan, or Korea) or, in the Digital Markets Act parlance, reduce contestability. For example, many commentators consider Facebook’s acquisition of Instagram in 2012 a conglomerate merger (Instagram was then mostly a small complementor to Facebook allowing users to edit photos before posting them on Instagram or cross-posting them on Facebook) that eliminated what would have become Facebook’s main competitor in social media.

Methodologies to perform ecosystem definition can help competition authorities to refine the application of the ‘ability-incentive-effects’ framework to assess the existence and strength of conglomerate mergers efficiencies and harms. A more rigorous delimitation of which product markets are to be considered within the ecosystem would allow to better detect in which markets the merged entity may have incentives to engage in anti-competitive practices post-merger, or in practices benefiting consumers. It would also allow authorities to have a more precise vision of the potential competition the merger could eliminate. Moreover, some of the methodologies described below generate metrics that can be used to quantify the incentives and the expected effects of a conduct post-merger. Intuitively, the merged entity is more likely to engage in conduct involving more related markets or (potential) competitors within an ecosystem. By the same token, the conduct should have stronger effects if the markets/(potential) competitors are more closely related. To assess relatedness, a competition authority could, for example, analyse the correlation and factor loading matrices when applying the factor analysis methodology. Alternatively, it could assess the weight of the links connecting two markets and measure their connectedness when the network of complementarities methodology is applied. In essence, by defining the parties affected by and the expected strength of post-merger conduct, ecosystem definition would play within the ‘ability-incentive-effects’ framework a role akin to that of market definition when computing post-merger market shares.

Calculation of fines and quantification of harm/effects

It is well established that the determination of appropriate fines is related, among others, to ‘the value of sales, depending on the degree of gravity of the infringement’.⁵⁵ In turn, the value of sales depends on the markets that have been affected by the infringing behaviour, directly or indirectly.⁵⁶ When the infringing conduct and its effects are concentrated in a single market or in a small number of markets, the calculation is straightforward. However, it becomes more complex as the implementation and effects of the infringing behaviour are diffused across an indeterminate number of markets and/or economic segments. Because the Guidelines on Setting Fines are very broad, the Commission has enjoyed great latitude in setting fines, often with little explanation on the causal link between the fine and the generated revenue. Part of the reason is that the fine is supposed to have a deterrent effect, and even remotely linked activities can be counted. But this exercise cannot be limitless, and indeed it is common for companies to contest the validity and proportionality of fines on the ground that they were imposed on shares of sales that are (allegedly) irrelevant to the infringing conduct. In Google Android, for example, Google argued that “the contested decision breaches the principle of proportionality … [and that] the Commission wrongly calculated the relevant value of sales, applied an incorrect gravity multiplier, added an unwarranted additional amount and failed to take account of various mitigating circumstances”.⁵⁷ So far, the litigated cases that involved ecosystems and resulted in the imposition of a fine were not overly complex in terms of the outer boundaries of the economic activity that incurred the fine.⁵⁸ But to ensure the legitimacy of imposed fines and that fines withstand judicial review in more complex ecosystem activities, the footprint of the potentially infringing behaviour will have to be mapped more accurately. Without evidence of which products and services are affected by the infringing conduct, i.e., a provable causal link of how the conduct is connected to the products and services, the calculation of fines will justifiably be exposed to appeals. Formal methodologies to define ecosystems can help not only delineate boundaries that delimit the products and services to be included but can also reveal the internal structure of ecosystems, such that it becomes apparent how closely generated revenue is to the infringing behaviour.

3. METHODOLOGIES FOR ECOSYSTEM DEFINITION

As discussed above, ecosystems are made of firms that purposely coordinate to make their products complementary and, in that manner, produce a joint value proposition. An investigator interested in analysing an ecosystem from a competition perspective would first need to identify the products, services, and actors that constitute it. A methodology that does this would play a role analogous to the SSNIP test in market definition. Similarly to how the SSNIP test provides a degree of substitutability above which two products are to be considered to belong to the same relevant market, analogous methodologies should provide boundaries between the products, services, and actors that constitute an ecosystem.

In the remainder of this section, we present four types of methodologies that can be used to do so. They each have their own features and areas they perform better in, and we account for those in Section 4. As we will show below, the investigator can use one or more of these methodologies to define an ecosystem, depending on the nature of the ecosystem and the purpose of the investigation.

Cluster markets and cluster analysis

We begin with cluster markets, not because they are the paradigmatic ecosystem structuring method, but because they are already familiar to the antitrust community, and they represent an early attempt of competition law scholarship and case law to deal with groups of products and services where defining separate markets around each product or service did not seem conducive to the purpose of the investigation. We build on this early scholarship and case law to offer directions on how cluster markets, and cluster analysis more broadly, can help achieve more rigorousness in properly scoping out ecosystems for the purposes of competition law.

It is important to note that cluster analysis is more likely to develop its full potential not as a stand-alone methodology but one that is combined with others in the list below. This is because in cluster analysis cases fall in only one cluster, which may often not reflect how real markets work, that is many products or services can be part of different ecosystems at the same time. As we explain in Section 4, other methodologies may be more helpful to competition authorities, and clustering can serve as a secondary methodology once an initial version of an ecosystem has been constructed through other methodologies. That said, since at least one iteration of clustering (in the form of cluster markets) has been used in competition law analysis, we start here.

Cluster markets

Definition

Cluster markets, and the related one-stop markets and aggregator markets, are market organization concepts that have existed in competition law for decades but have been rarely used, and even then, it was with little consistency. While cluster markets predate the emergence of ecosystems, they describe groupings of business activities, products, and services similar to those reflected in the concept of ecosystems, and can therefore provide useful insights. Perhaps due to the disuse of the concept, a full methodology around cluster markets was never developed. However, we suggest that cluster analysis from statistics can be ported to help design a better methodology for defining ecosystem cluster markets.

Case law

Most cases that make use of cluster markets come from the USA. An early version of cluster markets was implied in Crown Zellerbach, where the Circuit Court attempted to define a ‘line of commerce’ (as was required by Section 7 of the Clayton Act applicable in the case) by looking for a product line that was ‘sufficiently inclusive to be meaningful in terms of trade realities’.⁵⁹ The product line did not have to be a single product, but any bundle which based on trade realities could be seen as the relevant market to be affected by the competitive conduct of the actors involved (in the case at hand, a merger). In Philadelphia National Bank, the Supreme Court refused to examine products and services offered by banking institutions as separate even though there was separate demand for them, but rather combined them in a ‘cluster of products (various kinds of credit) and services (such as checking accounts and trust administration) denoted by the term “commercial banking”’⁶⁰ because trade realities revealed that, while the products and services taken separately were insulated from competition, together they were not. In Grinnell, the SC lumped together various property protection services under the overall market for accredited central station services, recognizing again that even though there are substitutes for each particular service separately, “it would be unrealistic … to break down the market into the various kinds of central station protective services that are available. Central station companies recognize that, to compete effectively, they must offer all or nearly all types of service. The different forms of accredited central station service are provided from a single office and customers utilize different services in combination”.⁶¹

In the EU, ‘one-stop’ shopping markets have been treated as cluster markets, where “the broader experience of shopping at a supermarket or other type of food-retailing outlet where [the consumer] can find all the goods he needs on the spot (one-stop shopping)” is not comparable to ‘shopping at specialised outlets’⁶² and therefore, the specialized outlets do not compete with the one-stop outlets even though they may sell individually substitute products.⁶³

Aggregation markets

Aggregation markets are also similar in that they provide consumers with a curated selection of products or services accessible through the aggregator collectively or in combinations chosen by consumers.⁶⁴ Often, the aggregate product or service can be offered at a discount (bundle),⁶⁵ but this is not necessary. For example, while some telecom service packages (e.g., triple-play), when sold together, may be cheaper, other aggregation services, like Airbnb’s accommodation and experience offerings do not usually come with discounts when purchased together.

Existing methodological approaches

Despite the intuitiveness of cluster markets, defining their boundaries has remained elusive. One simple proposal has been to rely on complementarity between components, in the sense that they are used together.⁶⁶ However, on top of methodological questions in terms of how exactly one would measure complementarity, the proposal seems to run against the previous case law; the cluster and one-stop shop markets in the case law did not always involve complementarities between components—for instance, different types of banking products or items sold by supermarkets (both of which have been seen as cluster markets) are not necessarily complements.

Hovenkamp summarizes and specifies the case law criteria in a more restrictive approach recognizing that if clustering is to offer analytical value, it has to be distinct from simply lumping together related products or services.⁶⁷ He argues that cluster markets make sense if 1) “many customers” need or at least “prefer the convenience of receiving the defendant's grouping of products” rather than any single product, or (2) “economies of joint provision (economies of scope)” make it cheaper to distribute the cluster rather than each good separately, and (3) entering into competition with the cluster is difficult.⁶⁸ This definition includes both demand (first and second conditions) and supply (third condition).⁶⁹ It still, however, does not provide much guidance in terms of how to measure customer preferences or economies of joint provision. Customers may indeed find it convenient to fill up on gas at supermarkets alongside their other shopping, but it is questionable whether gas should be brought under the supermarket cluster. Similarly, large conglomerates like Rolls Royce or General Electric may enjoy a certain measure of economies of scope in producing, for instance, both aircraft engines and car engines, but one would need a more compelling case to cluster those two products together. Perhaps, if the first two criteria were cumulative and not disjunctive, the proffered conditions may constitute a better definition.

Ayres, in a more streamlined approach, draws the limits of cluster markets around the transactional costs of purchasing or using multiple products or services from a single provider.⁷⁰ Transactional costs here should be interpreted broadly to include search, purchase, use, and other costs involved in finding, using, or purchasing products and services. This can even include attention or cognitive aspects. For example, even when ‘competition is one click away’,⁷¹ consumers may find it preferable to use two services by the same provider than to make the effort to consider exiting the ‘zone’ of one provider and transitioning to another, even if the latter’s service might be superior (this is sometimes referred as ‘minimum cognitive spanning tree’ or ‘lowest cognitive load’).⁷² To identify whether and which products and services can be grouped together in a cluster, bundle, aggregation package, or one-stop shopping experience one should look at consumer behaviour to see whether consumers buy the group of products or services from individual firms and, secondarily, whether sellers offer or market them jointly (but not tie them).⁷³ When this is the case, the cluster has an economic significance beyond the individual products or services which it contains and it can therefore be seen as constituting a separate market of its own.⁷⁴

These approaches suffer two shortcomings: first, that they are limited to one variable around which clusters can be formed (e.g., for Hovenkamp it is the preference for purchasing together, for Ayres it is transactional costs), making clustering both over-inclusive and under-inclusive because any one variable will inevitably fail to capture all correct aspects of competitive relevance to include in an ecosystem. Secondly, they do not come with a defined methodology on how to actually form clusters based on the chosen parameter.

Cluster markets in ecosystems

To enhance the concept of cluster markets for use in ecosystems, so that courts and agencies can more confidently know both whether an ecosystem (or more) exists and what falls under it, they can incorporate cluster and factor analysis from statistics. Cluster analysis provides a structure and methodology for what courts and agencies could only do intuitively so far in attempting to delineate ecosystems, and it allows for more criteria to be used, as opposed to the single criterion methods proposed in the literature above.

The objective of cluster analysis is to group together observations/cases into clusters that are similar.⁷⁵ Clusters and sub-clusters can be seen as ecosystems, and for a given population of cases, the boundaries of clusters (or sub-clusters) are the boundaries of the ecosystem. Much like the use of cluster markets in existing antitrust case law, the ecosystem market definition based on cluster analysis is not built around substitutes but rather around a measure of commonality (but not necessarily complementarity).⁷⁶ Various methods exist, and their helpfulness depends on the purpose of the investigation.

Hierarchical clustering

Description

Hierarchical clustering is a basic clustering method that helps investigators form progressively more inclusive versions of an ecosystem from a given population of products/services.⁷⁷ In hierarchical clustering, one starts from a population of products/services and a criterion for determining similarity is selected, upon which the distance between them is measured. Products/services that are closer to each other are clustered together resulting in a first-order clustering of the population. Products/services whose distance is beyond a chosen threshold can be excluded as irrelevant to the studied ecosystem, which serves as the outer boundary of the ecosystem. Then, the exercise continues either with the same criterion or with a different criterion of similarity and the initially resulting clusters are merged together in bigger clusters. The exercise is repeated either until one big super-cluster is formed or until the investigator has applied all relevant criteria, as illustrated in Figure 1. As this process unfolds, fewer but more inclusive clusters (or one super-cluster) are formed that correspond to the selected criteria and reflect the commonalities among their constituent parts.

Use

Hierarchical clustering is most useful in competition law as a method of obtaining the boundaries and structure of a single ecosystem based on the selected criteria, although it can be used to derive different ecosystems across a population of observations/cases. Because the available observations/cases are gradually grouped together into bigger clusters, what we obtain is a version of an ecosystem that gets progressively more inclusive. The investigator can choose the ecosystem version that derives at any of the intermediate agglomeration levels or the final super-cluster. In Figure 1, for example, the resulting super-cluster/ecosystem is that of integrated office product and service providers. However, the investigator can stop at lower levels, for instance, the ecosystem that includes only office products and services, but excludes services offered to suppliers of office products and services. The boundaries of that narrower ecosystem are delineated around paper supplies, furniture, computers, peripherals, after-sale services, and maintenance services only.

K-means clustering

Description

K-means clustering is used to group together observations/cases into a pre-selected number of distinct non-overlapping clusters.⁷⁸ Such division of observations/cases into clusters can result to two types of definitions for ecosystem analysis: (i) each cluster can represent a sub-ecosystem and the boundaries of each cluster delimit the boundaries of sub-ecosystems or (ii) the resulting clusters together can make up the ecosystem, in which case K-means clustering reveals the internal structure of the ecosystem. In both cases, score thresholds are established beyond which observations are excluded and this serves as the outer boundary of the investigation.

Use

In K-means clustering, the number of clusters is pre-defined based on an initial determination (this can later be adjusted).⁷⁹ Then, a centroid for each cluster is selected and observations/cases are assigned to the clusters (randomly or based on a reasonable assumption). Similar to excluding observations that lie beyond a certain measure of distance in hierarchical clustering, observations that lie beyond a set score are excluded—this determines the outer boundaries of the ecosystem population. The threshold can be computed based on statistical analysis across scores so that observations beyond a degree of standard deviation are excluded. Then their fit in their assigned cluster is assessed, the goal being to minimize the sum of the squared distance between the cases/observations and the cluster’s centroid. The cluster’s centroid is the arithmetic mean of all the cases/observations that belong to that cluster. The exercise is run repeatedly until every case/observation finds its best fit in one of the clusters. Additional criteria can then be applied to re-cluster cases/observations to reflect the multiple criteria. Essentially, K-means clustering attempts to find which cases/observations belong better together in each of the pre-defined clusters, thereby giving us the boundaries of each cluster.⁸⁰ Cases/observations within each cluster compete more closely together, and neighbouring clusters also compete more closely together.

Example

For example, assume that a competition authority wants to study the national grocery stores ecosystem to deduce clusters of competitive intensity among them (this would be useful, for instance, in a merger assessment or a market inquiry). The Greek Competition Authority performed such a market investigation in basic consumer goods where it treated this economic segment as an ecosystem and applied a ‘betweenness centrality’ measurement to identify market power dynamics across actors, products, and services.⁸¹ To map the groceries ecosystem, the agency collects data on all stores that sell groceries and household items across parameters that it deems relevant; for example, two key parameters are the distance from a given location and the number of categories of products they sell (this would distinguish between express stores, supermarkets, hypermarkets, etc). A score threshold is established (e.g., two standard deviations from the mean across the two parameters), such that stores below the threshold are excluded. This is a data-cleaning step to remove from the population any stores that are too dissimilar from the mean across the selected parameters. The raw data are represented in Figure 2.⁸² K-means clustering analysis would then allocate them in a pre-defined number of clusters (which can vary depending on the preferences of the agency) as in Figure 3. If the agency wants to add more parameters to capture additional aspects of competition among grocery stores—such as a score for additional services (e.g., gas station or in-house restaurants)—the clusters can be represented as in Figure 4. The agency now has an overview of the grocery market, the degree of stores belonging together, and how closely they compete (or not) based on the selected parameters.

Network of complementarities

Description

This methodology is based on detecting firms or products that jointly create value for end users through their complementarities. Identifying and quantifying the intensity of these complementarities allows the investigator to map (i) the within-ecosystem competitive pressures, (ii) cooperative interrelations, and (iii) the scope of welfare effects of conducts or structural changes to be studied.

The methodology poses two practical challenges. The first one consists in identifying all the possible products that could be linked through complementarities. This would serve as the outer boundaries of the ecosystem(s), as firms that have no (significant) links with another firm in terms of complementarities between their products do not belong to any ecosystem. However, the markets linked by complementarities in which these products are located might be a priori unknown to antitrust authorities and regulators. The second challenge is how to measure the intensity of complementarities between two products.

In order to solve both problems, let us recall that, for complementarities to emerge within an ecosystem, (i) they have to be firm-designed and (ii) consumers have to choose to consume compatible products together. The first condition opens a way in terms of measurement. Depending on the nature of the product under analysis, different ways of observing coordination between firms to create complementarities between their products can be devised. In other words, the nature of the relevant complementarities to be analysed (e.g., joint consumption, joint specific investments, complementarities in demand, interoperability between products, etc.) will differ across ecosystems. Therefore, the investigator must make an informed qualitative assessment to correctly select them.

Examples of complementarities that can be measured

For example, Battistella and others⁸³ analyse, among other things, the flows of knowledge and information between firms. Similarly, Basole⁸⁴ studies the mobile ecosystem by building a network of actors (nodes) linked through knowledge exchanges, as well as more explicit coordination mechanisms such as commercial agreements (alliances, partnerships, etc). In a study of the Korean Information and Communication Technologies (ICT) ecosystem, Lee and Kim⁸⁵ measure how content (books, games, etc.) created by content providers flows from platform providers (iOS, Android, etc.) through network providers (e.g., AT&T) to personal devices manufactured by device sellers. In digital ecosystems, one can analyse whether an app used another platform’s SDK or if it simply has a version that is compatible with a given platform (e.g., a smart TV brand, an app store, a social network, etc.). In non-digital ecosystems, technical specifications or commercial agreements can be used to trace inter-firm coordination that makes joint consumption more attractive to consumers. Then, once product-specific indicators of inter-firm coordination for complementarity creation have been devised, tracing all coordination related to a product’s complementarity with other products allows one to ‘discover’ all the candidate members of an ecosystem.

A metric of joint consumption

However, as mentioned above, consumers have the ultimate choice of which complements to combine. Therefore, once complementarities between products have been traced, the intensity of joint consumption between these products that can be attributed to the existence of complementarities between the two products has to be measured. The metric of intensity of joint consumption chosen will also depend on the products under analysis. In some cases, the intensity of joint use of two complements can be measured directly. For example, one could measure the percentage of downloads of a given streaming app per smart TV brand to obtain a measure of how linked the app is to the ecosystem of each smart TV producer. If, for example, a streaming app was available for two smart TV brands but brand A concentrated 90 per cent of its downloads, one could safely conclude that the only relevant ecosystem of this app is that of brand A. This case is simple in that joint consumption is directly observable at the consumer level, provided that data about the number of downloads per brand exists. When this is not the case, econometric methods could help to determine the extent of joint consumption. Moreover, the threshold value of what percentage of the consumption of a good is joint with another good with which it has complementarities is also case-specific and can depend on the data available.

A graphical representation

The existence of complementarity-creating inter-firm coordination and the intensity of joint consumption of two products allows to represent an ecosystem as a network in which each product constitutes a node, as shown in Figure 5. The existence of complementarities between two products defines a link between them. The weight of these links, in turn, represents the intensity of joint consumption that can be attributed to the existence of complementarities between the products. Depending on the threshold value chosen, only links with a minimal weight will be considered to be part of the network. In addition to the ease of visual interpretation networks provide, this mathematical object is convenient to determine which products are sufficiently related to be considered as being part of the same ecosystem. In order to do so, clustering techniques can be applied. Clustering allows to identify sets of nodes (products) that are sufficiently related (be it directly or indirectly) to each other and defining subnetworks that can be assimilated to ecosystems. For example, suppose that platform A is only linked to complementor A1, and platform B is only linked to complementor B1. In that simple case, platform A and complementor A1 would be found to be part of the same cluster (ecosystem A) and platform B and complementor B1 part of another cluster (ecosystem B). One can interpret that the closer two products are within a network, the more likely they are to generate competitive pressures on each other because their markets are more adjacent. This can be true even if the products are not directly linked. For example, two complementors can be linked to the same platform through complementarities while not being directly linked to each other.

Example

Let us now exemplify with an application of this methodology taken from Carballa-Smichowski and others.⁸⁶ In this case, the objective of the investigation is to delimit the ecosystems that are created between platforms, while ignoring the ecosystems that each of these platforms might constitute along with their non-platform complementors. In order to measure the existence of complementarities-creating inter-firm coordination, they consider referral visits going from one platform to another. Referral traffic from website A to website B represents the visitors who arrive at website B by clicking on a link in website A, excluding search results, ads, and links within emails and generalist social media. Then, referral traffic represents a conscious choice of platform A to facilitate joint consumption with platform B in order to increase the joint value generated by both platforms. In other words, it represents the existence of complementarities between the two services (platforms).

For example, as shown in Figure 5 below, TripAdvisor includes links to Hotels.com and Google Maps because the joint consumption of its content (touristic information) with the services offered by these two platforms (lodging search and reservation and digital maps, respectively) enhances the utility consumers enjoy vis-à-vis that of consuming each of these services separately. In order to measure the intensity of joint consumption, Carballa-Smichowski and others take the percentage of referral traffic received by website B that comes from website A. If this percentage is above a certain threshold, they consider the intensity of joint consumption to be sufficiently high for the two platforms to be linked through complementarities. In this study, the threshold value is determined from the data by only including the right long tail of the distribution of the percentage of referral traffic received by each platform. For most pairs of platforms, this percentage is very close to zero, as rarely do consumers click on a referral link connecting two platforms. By looking at the distribution of this variable, they can find a threshold value (i.e., a minimal percentage) that distinguishes the signal from the noise. In that manner, they build a network of platforms linked through complementarities that have incited users to consume their services jointly. Once the network is built, they use a community detection algorithm to split the network into several sub-networks of nodes (platforms) that are densely connected through complementarities and joint consumptions (ecosystems).

Value of the methodology

The main advantage of using the network of complementarities methodology consists in obtaining a data-driven identification of markets linked by complementarities. This method can thus help identify out-of-market competitive constraints in cases involving ecosystems. As developed by Batra, Bijl, and Klein,⁸⁷ ecosystems face competitive constraints from other ecosystems that are not exerted on the basis of separate services, but on the basis of a substitute set of complementary products located in different markets. In that vein, in Booking/eTraveli,⁸⁸ the parties identified out-of-market competitive constraints that the European Commission acknowledged but assessed to be insufficient. Moreover, this method can complement established methodologies for defining markets in the case of aftermarkets. In doing so, it would strengthen the investigator’s capacity to detect relevant competition constraints. In its Revised Market Definition Notice, the European Commission notices that “(digital) ecosystems can, in certain circumstances, be thought of as consisting of a primary core product and several secondary (digital) products whose consumption is connected to the core product, for instance, by technological links or interoperability”.⁸⁹ Moreover, it argues that similar methodologies to those used to define relevant markets in the presence of after-markets could be used. These include, but are not limited to, the assessment of (in)compatibility or switching costs between primary and secondary goods. However, the mere existence of (in)compatibilities between two products does not imply that consumers consider one to be an aftermarket of the other.⁹⁰ The network of complementarities method could allow to empirically identify whether this is the case when used on variables showing consumers’ effective behaviour (as in the example of cross-traffic developed), and so to refine the delineation of competitive constraints.

In abuse of dominance cases involving ecosystems, in which market definition is a pivotal exercise, the network of complementarities methodology could thus help refine the definition of actual or potential competitors, competitive effects, and remedies.

In merger control cases, although a clear market definition is not as relevant as in Article 102 cases, using a network of complementarities methodology in cases involving ecosystems can help the investigator in identifying and quantifying the knock-on effects of the merger. For example, in its assessment of the competitive effects of the Booking/eTraveli merger, the Commission’s main ecosystem theory of harm was that the merger would “reinforce existing network effects and raise barriers to entry or expansion on the hotel OTA market as a result of the traffic that Booking would acquire through the Transaction thereby allowing it to further strengthen its already dominant position on the market for hotel OTA services in the EEA and enabling Booking to harm hotels and end customers”.⁹¹ Logically, the Commission focused on simulating the impact of increased, merger-driven cross-traffic from eTraveli brands on Booking’s market share in the hotel online travel agencies’ (OTA) market, and its corresponding welfare effects. However, by analysing all the sets of interdependencies between OTAs and metasearch engines involving Booking and eTraveli, a fuller picture in the light of Figure 3 could have allowed for detection of secondary or knock-on effects on other complementary services or to better model third-parties’ strategic responses to the merger. For example, suppose that, as argued by the Commission, post-merger, an eTraveli flight metasearch starts sending more traffic to Booking directly. This, in turn, might reduce the cross-traffic between an airline receiving significant traffic from the eTraveli flight metasearch, on the one hand, and a partner hotel OTA competing with Booking, on the other hand, as consumers would have already found accommodation before buying the flight ticket. This would further entrench Booking’s position in the hotel OTA market. If this airline in turn was sending significant traffic to a dominant car-rental company, it would in turn stimulate competition in the car-rental market.

Factor analysis

Factor analysis is related to but different from cluster analysis in that, whereas cluster analysis groups observations/cases that are similar, factor analysis groups variables that are similar. Factor analysis can be adapted to allow agencies and courts to obtain a richer picture of which variables indicate belonging to an ecosystem and then by measuring these variables across a population (of products, services, actors, etc), determine which individuals belong to the ecosystem.

Description

Factor analysis works by testing whether multiple observable variables covariate in response to an unobservable latent variable (the factor) across a set of cases.⁹² The variables that are found to covariate are the relevant ones that explain and can be attributed to the latent factor:

• The multiple observable variables would be the measurable parameters, such as whether products are bought, sold, or used together, whether there is joint investment or production, whether there is interoperability between them, or any other parameter agencies may think links products or services together for the purposes of each particular investigation.

• The individuals would be the products and services for which we want to test whether they form part of the same ecosystem.

• The unobservable latent variable would be whether the individuals form part of the hypothesized ecosystem (versus not belonging to it).

How closely an observable variable is correlated with the latent variable is indicated by a factor loading score. A factor loading with a higher value indicates a stronger correlation with the latent variable. In essence, factor analysis obtains data on multiple variables across individual products and services and determines whether they are influenced (covariate) by a more limited set of factors, which in ecosystem analysis is whether they belong in the same ecosystem. In more complex versions, the factors can be expanded to include two or even more ecosystems, in which case factor analysis would show which variables are influenced by which factors (i.e., belonging to a given ecosystem). This would be useful in cases where the enforcer wants to test different ecosystem versions or different ecosystems.

Example

To see factor analysis in action, assume, for instance, that an agency wants to test whether and which Apple’s products and services form an ecosystem (Figure 6).⁹³ A standard cluster market methodology as it currently stands in legal scholarship (see ‘Cluster markets’ section) would take a single variable such as ‘preference for joint purchasing’ (Hovenkamp) or ‘joint transactional costs’ (Ayres) and would look at trends of joint purchasing or transactional savings from joint purchasing respectively (regardless of whether the products and services are complements).

For example, Apple TV and iMacs are not complements; however, if consumer patterns reveal that consumers who purchase Apple TV also commonly purchase iMacs (at a rate higher than other combinations of TV streaming hardware and personal computer hardware), because the combination reduces their transactional costs,⁹⁴ then these products can be brought under the same ecosystem. But using these single variables separately may give an incomplete picture. Consumers may not necessarily purchase Apple TV and iMacs together, and perhaps there are no significant transactional savings from jointly purchasing those products together, but this should not conclusively rule out the existence of an Apple ecosystem in which both of these products belong based on other variables. Confirmatory factor analysis would hypothesize the existence of an Apple ecosystem and choose multiple variables that according to the investigator reveal a good fit with an Apple ecosystem. On top of the above variables, additional perceived relevant variables could include reliance on Apple software, seamless communication among Apple devices and services, product update cycles, pricing variations, compatibility and interoperability among Apple devices and services, etc. Then candidate products and services for inclusion in the hypothesized Apple ecosystem would be checked against these variables, meaning that the variables would be measured for each candidate product or service. If the measured values covariate, it means that the variables are linked back to the latent factor (belonging in the hypothesized Apple ecosystem); if not, then those variables are less indicative of belonging in the ecosystem. The value measurements of each of the products or services across the relevant variables indicate how closely each of the products or services is linked to the ecosystem (e.g., product A gets a score of 4 out of 5 in the variable ‘update cycle’). Then factor loadings are calculated (e.g., seamless communication gets a factor loading of 0.9 out of 1 indicating that seamless communication with Apple products is a crucial characteristic of belonging in the Apple ecosystem).

Once this process is repeated for all candidate products and services, and all variables, the investigator can decide thresholds for inclusion in the Apple ecosystem and observe which products and services make the cut based on their scores in the relevant variables. If desired, variations of the Apple ecosystem as a factor or the product/service variables can be tested, or, again depending on the purpose of the investigation, one may want to look for other ecosystems that compete with the Apple ecosystem, and for competitors on a per product/service included within the ecosystem.

Snowball selection process

Description

Snowball selection is a popular qualitative methodology in the social sciences, whereby the investigator starts from a convenience sample of subjects (seeds) that fit the investigation criteria, and which in turn refer to other subjects matching the criteria and so on until either a target sample size or a saturation point has been reached.⁹⁵ A variation of snowball selection can prove useful in defining ecosystems.

For use in ecosystem analysis, snowball selection is most suitable when the investigator wants to define the ecosystem around a conglomerate company. This will either be for the purpose of determining the boundaries of the investigation (similar to the function that market definition serves as well),⁹⁶ or because the ecosystem is the correct unit of competition analysis in the investigation, as opposed to individual products and services that belong in the ecosystem. Ecosystems, comprising a multitude of products and services, compete in the market both at the level of individual products and services and at the level of ecosystem.⁹⁷ Snowball selection can help delineate ecosystems thus allowing authorities to map how competition works at that system level rather than at the individual level.

The challenge with the snowball selection process is to devise appropriate rules of reference from one subject to the next so as to avoid both stalling and infinite expansion.⁹⁸ Stalling occurs when the reference criteria are so strict that the waves of reference from one subject to the next are exhausted quickly, in effect resulting in very narrow ecosystems. On the other hand, infinite expansion occurs when the criteria are so loose that the ecosystem is virtually never saturated and becomes all-encompassing. Too narrow ecosystems will not accurately reflect competitive forces and constraints, whereas all-encompassing ecosystems are devoid of analytical value.

Use

A proposed snowball selection process for ecosystems sees products/services as subjects and starts with a single product/service (or a set thereof) or undertaking based on a strong specific inclusion rule (see Figure 7).⁹⁹ Strong inclusion rules are those that define strict criteria of which products/services will be included in the initial selection, such that the initial selection comprises only products/services that the investigator thinks to form the core of the ecosystem. This may sound like an arbitrary judgment, but it is no more arbitrary than the judgments that need to be made in regular market definition, among other things on the comparator products and services, the threshold of interchangeability, etc.¹⁰⁰ To some extent, it is impossible to kickstart the market or ecosystem definition process without making some initial arbitrary choices (which, incidentally, is one of the reasons why market definition has been criticized as pointless).¹⁰¹

The next step is to establish reference criteria. These are the criteria that will move the investigation outward from one product/service group to the one at the next degree of separation. Two criteria become relevant here. First, because ecosystems encompass not only vertical but also diagonal relationships, the referred product/service categories can be either in an input or in a complement relationship with their referrer. Secondly, under the assumption that competition law prioritizes consumer welfare,¹⁰² only products/services that interface with the consumer can be referred to, regardless of the degree of separation. This way, the resulting ecosystem (and the boundaries of the ensuing competition law analysis) will encompass those products and services that have a more direct bearing on consumer welfare and are therefore more relevant to the competition law investigation. One should note that the term ‘consumer’ here need not be understood exclusively as the final consumer, but it may refer to intermediate consumers as well,¹⁰³ in which case the definition of the ecosystem boundaries will be refocused around them. This variation is discussed in more detail below.

Once the reference criteria have been established, the investigator identifies the first wave of subjects, that is inputs and complements to the core seed subject(s) that interface directly with consumers (end or intermediate). This step is repeated until no more input/complement waves matching the reference criteria can be added.

Example

An example can be instructive here. Suppose a competition authority wants to reconstruct Google’s ecosystem (cf. Figure 8). The seed subject can be Android. The first wave of products/services can include applications, payment services, phones, etc. These are products and services that serve as inputs or complements to Android and which interact directly with end users. Taking applications as one of the focal points for the second wave, the investigator now looks for inputs and complements to applications that interact directly with end users—these can include content producers and advertisers. An example of an input that would not be included is SDKs, because, even though they are an input to applications, they do not interact directly with consumers. Taking phones as another focal point for the second wave, inputs and complements that interact directly with end users can be network connectivity services, accessories such as headphones, etc. This exercise is repeated until the investigator runs out of waves of products/services that match the above criteria.

As mentioned, the reference criterion that the included subjects interact directly with consumers is meant to serve as a limiting factor of the ecosystem’s boundaries and was introduced under the assumption that the ultimate objective of competition law is to protect consumer welfare. It does, however, result in the exclusion of products and services that may still be relevant in an ecosystem but do not interact with consumers. For example, content delivery networks (CDNs) may have a significant influence on how consumers receive content and yet they are invisible to them. A potential solution would be to refocus the ecosystem around a different class of consumers, an intermediate consumer rather than the final consumer. For example, if application developers are chosen as the class of consumers around whom the ecosystem will be reconstructed, all previous inputs and complements remain, alongside CDNs, albeit in different waves.

4. SYNTHESIS: WHEN TO USE EACH METHODOLOGY

Ecosystem definition is not a singular exercise. Much like market definition, it can be used to delineate the boundaries within which products, services, and firms face competitive constraints, where competitive effects will be investigated, and where investigative efforts will be focused. Depending on the function that the definition of ecosystem(s) will be called to perform, the various methodologies presented previously can be of varying degrees of utility and deliver different results. In fact, it is likely, perhaps even advisable, that more than one way of defining ecosystems is used in the frame of an investigation, so as to increase the corroborating power of each ecosystem definition. Because ecosystem definition is a tool rather than an end in itself, it can be performed in different ways if it helps competition authorities gain a more complete picture of the factual setting of their investigation. Moreover, parts of the methodologies presented above can be combined to allow for a more customized and flexible approach to ecosystem definition. For example, one can combine snowball selection with cluster analysis; the snowball selection process would determine which products and services are included in each wave, and then clustering would help group similar products and services together to add more structure to the population in each wave.

Table 1 summarizes the features of each methodology and highlights their differences. Several insights emerge from comparing the different ways to define ecosystems. They point to the suitability of each methodology based on specific purpose and function of defining ecosystems.

Single versus multiple ecosystem definition

One first point concerns the question of whether one wants to determine the boundaries of a single ecosystem built around an undertaking’s product or service, or whether the goal is to derive how an economic sector can be organized in different ecosystems (eg, around different orchestrator firms). The former would be useful, for instance, in abuse of dominance investigations where the ecosystem sponsor is challenged for their role in the ecosystem built around their core products and services.¹⁰⁴ Quasi-regulatory competition law provisions, such as Article 19a of the new German Competition Law, which calls for obligations on undertakings which are of paramount significance for competition across markets,¹⁰⁵ can also benefit from such single-ecosystem definition tools, since an undertaking’s presence across markets can be defined as an ecosystem as per above. Single-ecosystem definitions can also be used in the assessment of concentrations, whereby the merging ecosystems are defined separately, followed by an analysis of the competitive effects of the resulting merged ecosystem.¹⁰⁶

All of the described methodologies can be used for single-ecosystem definition, with some useful points of differentiation: hierarchical clustering and K-means clustering start from the population and build up the ecosystem while also showing intermediate agglomeration (in a hierarchical relationship) or sub-clusters (in non-hierarchical relationship), such that the internal structure of the ecosystem is also visible; similarly, the network of complementarities method starts from a set of pre-selected firms or products and builds up the ecosystem from observing inter-firm (or inter-product) linkages and then clustering firms or products based on the density and the structure of these linkages; factor analysis, inversely, starts with an assumption of an ecosystem and tests what products, services, and actors fall in it after detecting the variables that indicate inclusion in the ecosystem; in that sense it is helpful when investigators already have an ecosystem in mind but are not sure which parameters indicate inclusion in that ecosystem; snowball selection does not start with a whole ecosystem hypothesis, but rather with seed product(s) or service(s) and builds the ecosystem outwards, and therefore it is helpful when there is only a core of an ecosystem in mind.

The definition of multiple ecosystems, on the other hand, can be required when competition authorities seek to study an entire economic sector to assess the competitive conditions within (and potentially introduce quasi-regulatory measures). This would be the case of market inquiries, such as those prescribed in competition laws of the UK,¹⁰⁷ Greece,¹⁰⁸ and the now abandoned new competition tool of the European Commission.¹⁰⁹ K-means clustering and network of complementarities are the only methodologies that specifically allow for the simultaneous reconstruction of multiple ecosystems.

Preselected versus open-ended population and ecosystem structure revelation

A second and third inter-related parameters that determine the suitability of each ecosystem definition methodology concern (i) whether the population or ecosystem are pre-selected and (ii) whether the investigator wishes to determine only the ecosystem’s boundaries or also its internal structure (and by extension the internal boundaries of the ecosystem’s sub-divisions). The reason why they are interrelated follows from their methodology: if the population is pre-selected, then the definition exercise is geared towards organizing the population into one or more ecosystems, thereby revealing how the pre-selected population is structured across one or more ecosystems. This illuminates which population members compete more closely with each other and how big each ecosystem subdivision is.

Hierarchical clustering falls under this use. Pre-selected observations that do not match the set criteria can be excluded, and, because the available observations/cases are gradually grouped together into bigger clusters, the investigator can choose the ecosystem version that derives at any of the intermediate agglomeration levels or the final super-cluster. An application of this could be the Staples/Office Depot merger, where authorities had to reconstruct the market in which Staples and Office Depot were active. Given that the products and services sold by Staples and Office Depot were finite and known in advance, each company could be structured as an ecosystem and then authorities could observe the extent to which they overlap, how closely they compete (at ecosystem level and at intermediate agglomeration levels), and what the resulting ecosystem could look like post-merger.¹¹⁰

K-means clustering also relies on a preselection of population instances (and score thresholds can be used to define the outer boundaries) but also pre-determines the number of clusters in which the population will be grouped. Each cluster can represent an ecosystem, or all clusters together can represent the (super) ecosystem, depending on the needs of the investigation. Therefore, K-means clustering is more appropriate when investigators already have an industry structure in mind and want to see how the relevant population is distributed across it.

The network of complementarities approach also begins with a pre-selection of a population (ie, firms or products and the variables that might constitute relations between them). Precisely because it adopts a relational perspective, in addition to delineating the boundaries of the ecosystems obtained, this approach reveals their structure. The structure of the links (relations) between the nodes (firms or products) shows how (and, if the links are weighted, also to what extent) firms or products connect to each other directly and indirectly. It allows to analyse which firms or products are critical to the ecosystem because of their centrality, which can be measured in multiple ways depending on the investigator’s focus.

If the definition process is open-ended, meaning that the investigation starts with a core group of products or services and expands outwards, then factor analysis and the snowball selection process are more suitable. Both result in the delineation of external boundaries but are not very helpful in revealing the internal structure of the ecosystem.

Factor analysis hypothesizes the existence of an ecosystem, then detects which variables indicate inclusion in the ecosystem, and then tests subjects for belonging by computing thresholds for inclusion. New subjects can be tested as needed until no new subjects pass the threshold. Therefore, factor analysis is most suitable when the main goal is to determine the boundaries of a presumed ecosystem, but not its internal structure, and when investigators are not sure which parameters are determinative of inclusion into/exclusion from the studied ecosystem.

The snowball selection process starts from a seed product or service and expands outwards based on the defined reference criteria. The investigation continues until no new products or services meet the reference criteria. Similar to factor analysis, the snowball selection process is best used to delineate an ecosystem’s outer boundaries, but not its internal structure: snowball selection reveals the different waves, but it is up to the investigator to group products and services in each wave into coherent groups, if so desired; as mentioned earlier, this grouping can be done using K-means clustering.

The combination of snowball selection and clustering can be used to study a single ecosystem or to compare ecosystems that are expected to have similar structures. The CMA’s investigation into the dominance of mobile operating ecosystems would be a good testbed for these methodologies.¹¹¹ The CMA set out to investigate the structure, business models, and dominance of Android and iOS, and while it observed that their revenue streams are different (which influences their conduct in the market), the basic ecosystem structure is similar. The CMA adopted an ecosystem representation shown in Figure 9, where it distinguishes four layers and their inter-relations. An application of the methodologies discussed herein would deliver both a clearer separation of groupings within and across layers and would reveal other forces at play. Consider a combination of clustering and snowball selection, similar to CMA’s approach, where the operating system is placed at the core and the investigator proceeds outwards. Snowball selection would reveal the layers and clustering would group products and services together per selected characteristics. The CMA’s approach shows native apps and pre-installed apps in two different groupings and yet an application of clustering would result in them being in the same group as they are often direct competitors. Conversely, web content and apps, which the CMA groups together would fall into different layers, as snowball selection would reveal that one is an input to the other. Similarly, app stores are in the same grouping as third-party apps even though they mostly serve as platforms for them. Furthermore, as the reference criteria of the snowball selection process continue to apply outwards, additional actors, such as advertisers or payment intermediaries, which weigh in heavily in the mobile ecosystem, would be added into the ecosystem.

Inclusion/exclusion criteria

One last consideration on the choice of suitable methodology revolves around the criteria used for the inclusion of products and services in the ecosystem. When constructing ecosystems, it is important to understand what the inclusion criteria represent: in hierarchical clustering and K-means clustering, the groupings express commonality among population cases, based on how the investigator understands commonality. In hierarchical clustering, products and services get progressively grouped in larger clusters, each level reflecting a grouping of similar products or services, until a super-cluster emerges. K-means clustering also groups together similar products or services, but the resulting clusters are not in a sequential hierarchical relationship. On the other hand, in factor analysis, inclusion in the ecosystem is based on criteria that indicate belonging in it, and the methodology allows investigators to try out different criteria to determine which of those indeed show belonging. In the network of complementarities approach, firms or products are grouped together according to their links to each other through complementarities. Lastly, the criteria used in the snowball selection process express the reference rules, which herein we defined as expressing vertical or diagonal input relationships (‘serving as input’) between products or services and a focus on consumer welfare (‘interact directly with consumers’).

Between closeness, belonging, the density of relations, and reference rules, there is no one superior criterion. They all express some sort of affinity among the products and services that make up an ecosystem, and the various criteria are indicative of the different ways by which the concept of an ecosystem can be understood.

5. LIMITATIONS AND FUTURE DIRECTIONS

In this section, we touch on some limitations and criticisms that we believe are important to keep in mind regarding our approach to ecosystem definition. As a first observation, we reiterate that the methodologies as presented herein are not meant to provide an exhaustive step-by-step vade mecum, but rather to bring insights from non-law disciplines into the radar of competition authorities on how to define ecosystems in a more rigorous way. This way, we increase the familiarity of enforcers with ecosystem definition methodologies, and we demonstrate that there are numerous scientific tools to define ecosystems, allowing ecosystems to play a more useful role in competition law analysis. Relatedly, and crucially, we acknowledge that not every case involving ecosystems necessitates full and detailed mapping of the ecosystem(s) in question. This is evident from extant case law, where the concept of ecosystem has been referred to in many cases, but only a few would have a full and detailed definition of material help. Notably, the CMA’s market investigation into mobile operating systems would have greatly benefited from more detail, and in the Google Android case, a more accurate mapping of Google’s ecosystem would have significantly helped the Commission and the General Court in their assessment of the anti-fragmentation rationale, which represented a key part of the case. As more economic segments structured around ecosystems form and are challenged by competition authorities, the cases where proper and rigorous definitions will add value will multiply too.

Furthermore, we appreciate that the methodologies we present are data- and resource-intensive. They require extensive data on candidate products, services, and actors that may be included in ecosystems, but competition authorities are no strangers to such resource-intensive exercises. Regarding data availability, competition authorities can request necessary data from the parties under investigation or involved in market inquiries. Moreover, private providers of fine-grained industry data are becoming widely available. As for resources, although these methods require investing in human resources with quantitative skills, the execution of these methodologies is not particularly complex for competition authorities, as the only novel element is the application to the competition analysis of ecosystems. For example, economists from the Competition Bureau of Canada have applied network centrality measures to study market power in the online travel ecosystem.¹¹² Hierarchical clustering has been applied since the 1980s to define commuting zones and labour markets, and recent research in consumption zones to establish zones of competitive intensity also extensively uses clustering.¹¹³ Some econometric methods used by competition authorities, such as demand estimation, are more complex and time-consuming compared to the methodologies presented herein. Finally, competition authorities are expanding their palette of quantitative tools in their daily operations by incorporating methods that are far more complex and time-consuming than the ones presented in this article, such as large language models and the analysis of text data with machine learning.¹¹⁴

Moreover, we should also recognize that the business practices adopted by undertakings can, in theory, affect how ecosystem boundaries are drawn in ways that escape the control of our suggested methodologies. For example, in a refusal-to-deal situation, whereby a complement is blocked from a platform around which a purported ecosystem is developed, the complement would be excluded from the boundaries of the ecosystem, the only reason being the refusal to deal and not the criteria applied as per our suggested methodologies. To clarify, this is not a problem that affects only ecosystem definition; it applies equally to market definition: the classic substitutability analysis should not include products or services that are unavailable for consumers to turn to on account of the business practices of the challenged undertaking. In any case, while possible, there is no reason why the otherwise excluded product(s) or service(s) cannot still be added into the population on which the methodologies will apply, assuming that investigators are aware that it is business conduct that excludes them. This is a reasonable assumption, since more often than not, the excluded product(s) or service(s) are the reason why enforcement procedures are initiated in the first place.

6. CONCLUSION

In recent years, there has been an increasing interest in ecosystems from academics and practitioners. Consequently, the concept of ‘ecosystem’ has been used in legal decisions, market inquiries, and legislative materials. However, despite the popularization of the term, ecosystems have so far failed to become a coherent unit of analysis in competition law.

We posit that the lack of operationalization of the concept of ecosystem in competition law analysis is due to the fact that traditional market definition, which relies on identifying a set of substitute products (ie, on the ‘substitutability approach’), is not sufficient to analyse ecosystems. Given that ecosystems link different markets together through complementarities, defining a single relevant market through the substitutability approach, although useful, is not sufficient to delimit sources of competitive pressure and the scope of welfare effects. Consequently, competition authorities and courts lack methodologies to define the boundaries of an ecosystem.

In this article, we propose a contribution that aims at rendering the use of ecosystems operational in competition law analysis as a unit of analysis. In particular, we present a set of methodologies to define the boundaries of an ecosystem and, in some cases, to reveal their structure. In a similar way to market definition, this exercise should allow us to look for actual or potential competitors, competitive effects, and remedies whenever competition analysis touches upon an ecosystem spanning across markets. We describe five non-exclusive methodologies and provide guidance on when each of them is pertinent: hierarchical clustering, K-means clustering, network of complementarities, factor analysis, and snowball selection.

We argue that ecosystem definition constitutes a useful and necessary complement to market definition whenever the conduct or market under analysis takes place within or between ecosystems. We hope this article will contribute to fill the gap between academia and practitioners by providing empirical methodologies that help operationalizing the use of ecosystems in competition law analysis. Moreover, we hope this article will spur further contributions that will refine the methodologies presented and develop new ones.

Footnotes