Platform competition and market definition in the US Amex case: lessons for economics and law

ABSTRACT The rise of multi-sided platforms in the marketplace has spawned a vast amount of research to understand their implications for competition and welfare. This paper presents the scrutiny of one such academic work that classifies multi-sided platforms into “transaction” and “non-transaction” platforms for the purpose of relevant market definition. It has been posited that in the case of “transaction” platforms, there is one all-encompassing relevant market comprising of all sides of a platform. And such a “transaction” platform can compete only with another “transaction” platform. The U.S. Supreme Court in its Amex decision relied upon this classification and elevated the same into law. This paper identifies flaws in this concept by demonstrating that the relevant academic work defines “transaction” too narrowly. The paper takes the swift adoption of the “transaction” platform approach by the Supreme Court as an opportunity to also provide lessons for economics and law.


Introduction
The research on two-sided platforms that began predominantly in the early 2000s is two decades old now. 1 During this period competition law has benefited immensely from the economic research on multi-sided platforms. For instance, it is known how indirect network effects influence competition, 2 what adjustments are needed in the SSNIP test, 3 the effect of product differentiation on the market power of a platform, 4 how multihoming constrains the power of multi-sided platforms, 5 and non-neutrality of prices that makes below cost pricing on one side legitimate when seen on the aggregate level. 6 This learning process is still not over and we are yet to learn a lot more about platforms that will shape the way we apply competition law to such business models. For instance, there is still no unanimously agreed upon definition of a two-sided platform. 7 This paper, however, is aimed at a subsequent concept once a two-sided market has already been identified. It is aimed at analysing the further categorization of a two-sided platform into the so-called "transaction" and "non-transaction" platforms, first proposed by Filistrucchi et al 8 and later crystallised into law by the US Supreme Court in its Amex decision. 9 In the American Express case (hereafter Amex case) in the US, while determining the legality of the "anti-steering provision" that American Express imposed on merchants, the US Supreme Court defined credit card market as one relevant market encompassing both cardholders and merchants as its respective two sides. 10 The Supreme Court defined one encompassing relevant market instead of two separate yet interrelated markets, arguing that a credit card platform is akin to a "two-sided transaction market"using the reasoning and nomenclature proposed by some academics. 11 In a 2014 paper Filistrucchi et al classify two-sided platforms into "two-sided transaction market" and "two-sided non-transaction market" by proposing a test and argue that in the case of the former antitrust authorities need to define the platform as one all-encompassing relevant market. In the case of the latter, two separate relevant markets need to be defined. 12 Admittedly, the SC while explicitly used the term "transaction platform" 13 vis-à-vis the credit-card market and referred to the paper by Filistrucchi et al., it did not look into the test that sets out the criteria to identify a "transaction platform" generally.
There are at least two consequences of defining one market instead of two. First, as the US Supreme Court observed, a platform as a whole can only compete with another similar platform. This means that single-sided firms are not substitutes. Second, defining one market or defining multiple markets also has bearing on the plaintiff's burden of proof. If an integrated market is defined, a plaintiff will have a tough task proving net anticompetitive effects, if there are corresponding pro-competitive effects on the other sides(s). 14 The Supreme Court also acknowledged that the plaintiff has a higher burden of proving net harm to competition. 15 While much ink was spilled in rejecting the integrated market approach with respect to credit cards by the dissenting part of the judgment delivered by Justice Breyer, the test proposed by Filistrucchi et al. to identify "two-sided transaction" markets remains uncontested so far, even in the academic literature. This paper points out flaws in the test proposed by Filistrucchi et al. and argues that instead of attempting to pigeonhole two-sided platforms based on certain characteristics, which cannot be generalized, into different types for the purpose of market definition, Bus. L. Rev. 667, while observing "credit-card networks are a special type of two-sided platform known as a 'transaction' platform" page 2, Ohio v American Express (n 9). However, Filistrucchi et al. have more prominently dealt with the concept of "transaction" platform and have attempted to develop it as a sub-category of two-sided markets. 12 Filistrucchi et al. (n 8). 13 This paper uses the terms "transaction platform" and "transaction market", and "two-sided" platforms and "multi-sided" platforms interchangeably. regard should be had to the competitive conditions measured by product interchangeability. Not only do such "bright-line" tests have restricted utility for they are extrapolated based on limited observation, such tests also remain prone to the pace of innovation that is known to obliterate market boundaries.
How we approach market definition in cases of multi-sided platforms essentially reflects upon our understanding of how platforms compete in the marketplace. At a time when our understanding of platform competition is still developing, it is too early to propose pre-defined criteria that run the risk of leading to an erroneous conclusion about the state of competition in the relevant market. The Amex case also provides lessons for the law and economics discipline in general. The quick adoption of a flawed economic test that has not had the occasion to be contested by other academics and courts, begs the critical question about the adoption process and standards of an economic theory before it matures into law. The paper thus also illustrates that the imprecision of economic science amplifies manifold when courts translate little contested theories into law.
Part 1 of this paper briefly summarizes the adoption of the "transaction platform" test by the US Supreme Court in the Amex case. Part 2 then by way of practical examples illustrates the shortcomings in the test. This part also shows that even alternative methods to categorize platforms for the purpose of market definition are incoherent and thus legally unpalatable. Based on this part, Part 3 presents lessons for economics. Part 4 by referring to the quick adoption and subsequent disenchantment with the "Areeda-Turner" test argues that a new economic insight must go through a "natural aging process" before it starts informing the law. This part also analyses the extent to which the EU competition law framework allows a new economic theory to be contested before it is translated into law. Part 5 concludes.

The decision in the Amex case and "transaction platforms"
The US Amex case was primarily centred on the antitrust legality of the "anti-steering provisions" employed by American Express with respect to its merchant partners. Amex adopted a business model that charged merchants higher fees as compared to its rivals. Thus, to prevent the merchants from steering the Amex card users to other cheaper options, Amex used contractual restrictions on its merchants. The United States and seventeen States collectively sued Amex arguing that the "anti-steering" was anticompetitive inasmuch as it violated § 1 of the Sherman Act. The proper examination of this allegation under the rule of reason using the burdenshifting framework required the courts to define a relevant market.
At the District Court, American Express argued that the relevant market should be decided based on "transactions" encompassing the entire multi-sided platform. 16 The District Court, however, rejected to define an all-encompassing relevant market citing past precedents. 17 The Court also observed that American Express provided no "compelling reason" to depart from the past precedence. 18 Accordingly, the District Court held that the credit-card market should be treated as two separate, yet deeply interrelated markets with merchants on one side and cardholders on the other side. 19 On appeal, the Second Circuit without referring to the "transaction market" approach defined a single relevant market by observing "[s]eparating the two markets here-analysing the effect of Amex's vertical restraints on the market for network services while ignoring their effect on the market for general purpose cards-ignores the two markets' interdependence". 20 The SC in the majority opinion delivered by Justice Thomas held that the credit-card market is a type of "transaction" market and thus this network is "best understood as supplying only one productthe transactionthat is jointly consumed by a cardholder and a merchant. Accordingly, the two-sided market for credit-card transactions should be analysed as a whole". 21 The SC explained that "transaction platforms" facilitate a single simultaneous transaction between two sides. Thus, when a credit card network sells its services to a merchant, it simultaneously needs to cater to a cardholder. The transaction in this operation, therefore, is indivisible and concurrent. 22 The SC treats "transaction" markets as a special case, in that the pro- two-sided platforms), they are mandatory for the existence of a transaction. It is true that while referring to the "transaction" market, the SC refers to other academics as well 23 ; however, Filistrucchi et al. have more prominently dealt with the concept of "transaction" platform and have attempted to develop it as a sub-category of two-sided markets. Aside from the nomenclature, the majority judgment frequently refers to Filistrucchi et al. while explaining the features of a "transaction" platform.
In the paper relied upon by the SC, Filistrucchi et al. define the twosided transaction platform based on three criteria. 24 First, presence of a transaction between the two groups of platform users; second, observability of a transaction between the two groups of platform users; and third, because of observability, a per-transaction (or per-interaction) fee or a two-part tariff is possible. Judged against this criteria, Filistrucchi et al. consider payment cards services, auction houses, video game consoles, operating systems and property rental agency as two-sided transaction market, noting that the product that is offered is "the possibility to transact through the platform". 25 Some scholars are of the view that in the case of multisided platforms, every side should be defined as a different relevant market, instead of defining a single multisided relevant market. 26 Also, there are others who believe that it is better to define one integrated market always. 27 Admittedly, there can be cases when it is prudent to define a single market. 28 Defining one market in the case of a particular two-sided platform is a matter of practical wisdom, in that the condition of In sum, while there is some divergence in views regarding the need to define one or two relevant product markets, this second school is generally consistent in the view that an assessment of whether there is anticompetitive harm must be incorporated at the prima facie burden stage of a rule-of-reason analysis. Notably, the Supreme Court in American Express (and the Second Circuit) endorsed this approach. page 726. 28 " … [b]oth approaches seem to be in line with the concept of demand-side substitutability; in particular, defining one single market does not conflict with this concept as a platform can be understood as a provider of an intermediation service, serving linked user groups with essentially the same service. All in all, and given the role of market definition as a tool that supports competitive analysis, neither of the two approaches seems right or wrong in absolute terms as long as the analysis appropriately accounts for interdependencies-such as indirect network effects-and for all competitive forces on each 'side' of the market.", Sebastian Wismer and Arno Rasek, 'Market Definition in Multi-Sided Markets', OECD (2018) Rethinking Antitrust Tools for Multi-Sided Platforms, 57. competition can be gauged at the level of the platform, as any competitive constraints may come from a similar structure. However, as this paper shows, this is a factual enquiry and pre-defined criteria about the platform type can lead to an erroneous conclusion about the relevant market.
It is noteworthy that even before the US Supreme Court adopted this nomenclature, the German competition authority (the Bundeskartellamt) had already resorted to the concept of "transaction" platforms. 29 However, the criteria chosen to characterize a platform as "transaction" platform was the presence of "pronounced bilateral positive indirect network effects between the two user groups". 30 In these merger cases, however, the Bundeskartellamt left the precise market definition open. In P7S1/ Verivox it was not conclusively decided if a "transaction platform", online comparison platform that also acted as an intermediary for contracts for various services, competed with alternative channels such as offline sales or supplier's own websiteboth single-sided markets. 31

Limitations of the concept of "transaction" platforms
This part scrutinizes the test proposed by Filistrucchi et al. and illustrates the glaring loopholes with their sweeping generalization of "transaction" markets.

Narrow view of "transaction"
Fillistrucchi et al. distinguish between transaction platforms, such as credit card networks, for which there is an observable transaction and nontransaction platforms, such as ad-supported media markets, for which there is no observable transaction. Speaking of interaction between the two sides of the market, the authors note that in media markets this interaction may be present, albeit only a delayed transaction is present, "and this transaction is usually not identifiable (as it is impossible to say 29 BKartA -Immonet/Immowelt, B6-39/15, case summary of 20.04.2015, page 2; also in, BKartA -P7S1/ Verivox, B8-67/15, case summary of 05.08.2015; In a subsequent paper, however, the Bundeskartellamt instead of using the term "transaction platform" uses the term "match making platform". Bundeskartellamt, Working Paper -The Market Power of Platforms and Networks, Ref. B6-113/15, June 2016. 30 "In this constellation of a so-called transaction platform there is a typical two-sided market with pronounced bilateral positive indirect network effects between the two user groups. In the case of transaction platformsin contrast to advertising-based two-sided marketsthe Bundeskartellamt considers it possible for definition purposes not to separate the different market sides." BKartA -Immonet/Immowelt, (n 29), page 2. 31 BKartA -P7S1/Verivox, (n 29).
whether someone has bought a product because he or she has seen an ad) so that the platform is unable to charge a fee for it". 32 The view taken by the authors interprets "transaction" narrowly, in that a transaction may not necessarily involve the exchange of money for goods/services. It may also involve cases where one side of the market only seeks the attention of the other side of the market. Different ads serve different purposes. Not every ad is aimed to make a consumer purchase a product or sign up for a service. Search and contextual advertisements that are typically subjected to a cost-per-click payment model are intended to elicit a response from consumers in the form of the purchase of a product or signing up for a service. Another kind of advertisement is a display ad that is generally used for brand advertising rather than having the consumers act on it. 33 Another shortcoming with the "observability" of a transaction is that the present technology makes it possible for the online newspapers to use advertising much like how the credit card market operates inasmuch as an online news platform may pay the advertiser based on the number of impressions that a particular advertisement receives. The US Supreme Court never got into examining this commonality between newspapers and credit card markets. 34 Evans also notes that online attention platforms facilitate observable transaction when a user clicks on an advertisement. 35 Thus, attention platforms have a changing nature. Interestingly, the possibility of the online newspaper portals using technology to calculate impressions was mentioned by Filistruchhi et al who remarked that Only recently, using online tracking technology, has it become possible for a platform to charge advertisers for a transaction in which an internet user buys a product online from the seller after having seen their online advertisement. Such technological developments may eventually push some media markets to become two-sided transaction markets. 36 Thus, the authors were aware that technological developments make the relevant market boundaries porous. In view of this, the test they proposed cannot be termed technology-neutral and hence will lead to erroneous results, especially at a time when the technology is evolving rapidly. 3.2. "Transaction" platforms also compete with "non-transaction" platforms The core idea behind defining a relevant market is to "identify those actual competitors of the undertakings involved that are capable of constraining those undertakings' behaviour and of preventing them from behaving independently of effective competitive pressure". 37 A relevant market is defined by identifying all those products that provide a "sufficient degree of interchangeability". 38 So far as the "transaction" platforms are concerned, it implies that similar "transactions" can be made only by bringing two different groups of users together. Overall, it implies that a multi-sided "transaction" market can compete only with another multi-sided "transaction" market. This is confirmed by the SC in the Amex case. The SC observed: Only other two-sided platforms can compete with a two-sided platform for transaction. A credit-card company that processed transactions for merchants, but has no cardholders willing to use its card, could not compete with Amex … Only a company that had both cardholders and merchants willing to use its network could sell transactions and compete in the credit-card market. 39 In footnote 9, the Supreme Court distinguishes "transaction" and "nontransaction" platforms by noting "[n]ontransaction platforms, by contrast, often do compete with companies that do not operate on both sides of their platform". 40 Filistrucchi et al also note that "[o]ne of the consequences of defining only one market is that a firm would be either on both sides of the market or on none". 41 This means that in order to complete a "transaction" a firm needs to be present on both sides. Thus, the SC and the academic paper that it relies upon indicate that a "transaction" platform will compete with a similar "transaction" platform only. This is a critical point regarding the business model of the two-sided "transaction" markets.
In general, not only do platforms compete with similar platforms, they also compete with one-sided markets. For instance, Free-to-air TV that is financed by advertising and Pay-TV may be substitutes as far as the Ohio et al. v American Express (n 9), footnote 9; However, the example of television network competing with newspapers to sell advertising that it chooses to substantiate this point is flawed, as a television network may also operate on both sides. 41 Ibid 301. viewers are concerned. 42 However, there are glaring exceptions to the proposition that a "transaction platform" competes only with another "transaction platform". So far, we do not have a large body of competition law cases on platforms. The following text presents some cases where competition authorities found a "transaction" platform to be competing with a one-sided market. This part is not comprehensive and is aimed at illustrating the limitations of the "transaction market" approach. If a "transaction" market competes with a one-sided market, then it makes no sense to define a "transaction" market as one all-encompassing relevant market. 43 Two-sided online portals that bring users and service providers often satisfy the test proposed by Filistrucchi et al. inasmuch as two sides engage in an observable transaction that can facilitate a two-part tariff by the platform. It has been found that hotel-booking portals may not be competing with hotels' own website 44 or other offline channels. 45 Also, travel ticket booking platforms that connect suppliers such as airlines and travel agents do not compete with airline websites. 46 However, another "transaction" platform, taxi aggregators, have been found to be competing with traditional taxi services that is a single-sided market. 47 These taxi services could either be owned by one company or owned by several individuals. In one Indian example, Meru owns a fleet of taxis that the Indian competition authority (CCI) found to be competing with OLA and Uber that are both "transaction" markets. The CCI also found that individually owned taxis (black-and-yellow taxis in Kolkata) can also compete with OLA and Uber. 48 Interestingly, it has been found that the competition in the taxi service market exists in the opposite direction as well. Thus, it is not asymmetrical. In a merger between Sheffield City Taxis Limited and Mercury Taxis (Sheffield) Limited, the UK competition agency, the Competition and Market Authority (CMA), observed that the merging parties, both onesided markets, would get close competition from taxi aggregators such Tim Wu also has also criticised the "transaction market" theory by pointing to some practical cases where a two-sided "transaction market" competes with a single-sided market. Tim Wu, 'The American Express Opinion, the Rule of Reason, and Tech Platforms' (2019) 7 Journal of Antitrust Enforcement, 104-127. 44 BKartA -HRS, B9-66/10, decision of 20.12.2013. 45 Such as sales of hotel rooms by telephone and e-mails, reservation forms on the hotels' own websites and sales via offline travel agencies and tourism organisations. Ibid 26. 46 Case No COMP/M.4523-Travelport/Worldspan, 21/08/2007, paragraphs 39-59. 47 Case No. 25-28 of 2017, Meru Travel Solutions Pvt. Ltd. v M/s ANI Technologies Pvt. Ltd. and others. 48 Ibid.
as Uber and Gett. 49 This is, however, not a uniform approach as the Spanish competition authority considered taxi app intermediaries falling in a different relevant market as compared to 1. Taxi radio/phone based intermediaries, 2. Taxi hail operators. 50 The Singapore competition authority also decided that there was no substitution between a two-sided taxi booking platform and street-hailed taxi service. 51 If consumers can switch from an e-commerce platform to brick-andmortar shops, then the relevant market cannot be e-commerce services. Instead, the relevant market will be a broader retail service to consumers. In one Indian case, which was a merger decision between two online retail "transaction" platforms viz. eBay and Flipkart, the Indian competition authority, indicated that an overall B2C market or its sub-segment online B2C market may exist, thus defining relevant market at the platform level. 52 However, the relationship between online retailing and traditional retailingwhether complements or substitutesis still not known and arguably depends on national preferences, product characteristics, and technological and business innovation. 53 "Transaction" market approach is problematic also for the reason that innovation may make it possible for a one-sided market to compete with such platforms. Disruptive innovation is known to blur market boundaries. 54 Especially at a time when technology is experiencing rapid changes, the law needs to be flexible enough to accommodate changes. It appears that the conceptual understanding of "transaction" and "nontransaction" platforms has been developed keeping in mind the differences between the media market and the credit card market. This conceptual understanding is not universal, however.

Alternative approaches
There are several other attempts as well to categorize platforms into predefined categories for the purpose of market definition. For instance, the Bundeskartellamt proposed that in the presence of two-sided positive network effects it is reasonable to define only one market. In a 2015 paper, the Bundeskartellamt observed that "[b]ilateral positive network effects occur in particular where the platform serves to connect two or more user groups for the purpose of direct interaction. Such platforms can be referred to as matching platforms". 55 The Bundeskartellamt terms platforms that witness only one-sided positive indirect network effects, which is the case with advertisement driven platforms, as "audience providing" platforms. 56 The Bundeskartellamt proposed in its paper that "at least in the case of matching platforms where the connection of the two user groups is the actual product offered by the platform it can be reasonable to define only one market". 57 However, noting that there can be exceptions, the Bundeskartellamt observed that where one side does not depend on a successful matching result to satisfy its demand. In these cases, a definition of just one market would fail to cover competitive relations that are potentially important. The question of whether one should define one or two markets, therefore, needs to be decided on a case-by-case basis. 58 In the Immonet/Immowelt merger decision, the Bundeskartellamt defined one single market because of the presence of bilateral positive indirect network effects, which lead to a largely uniform demand. 59 It is crucial to note that e-commerce platforms may display two-sided positive indirect network effects. But as shown above, it is possible that ecommerce platforms may compete with one-sided markets. Likewise, in some cases, viewers may also value advertisements on media platforms that bring together advertisers and consumers, triggering indirect network effects in both directions. Still, the platform does not have an observable transaction for the purpose of the test proposed by Filistrucchi et al. Thus, different tests lead to incoherent results.
platforms may be launched with one side only, and the second side may be added at a later stage". 60 This proposition is in conflict with the test proposed by Filistrucchi et al. Amazon or for that matter any other e-commerce platform may not be a "transaction" platform if one goes by the proposition of Wismer and Rasek. There is a possibility to operate the buyer side of the platform without having the seller side. This possibility is not just theoretical. Amazon first was catering to the buyer side only when it started in 1995. After gaining the "critical mass" it opened its platform to the sellers as well. 61 Also, the original Atari VCS videogame console functioned as a VI (Vertically Integrated) model and so did Yellow Cab. 62

Lessons for economics
An attempt to categorize platforms into "transaction" and "non-transaction" platforms that cannot be generalized is not very helpful for competition enforcement. To the extent we have understood multisided platforms, the presence of indirect network effects is the primordial element that influences competitive assessment. 63 Only due to network effects, the price structure and price level differ from a one-sided market. Additionally, for this reason, it is crucial to take all sides into account while assessing the effects of a firm's impugned conduct. Network effects can exist in one direction or both directions, resulting in different competitive dynamics. All in all, two-sided markets should be approached as a concept, not as a theory.
There are examples that do not fit into the defined pigeonholes of "transaction" and "non-transaction" platforms. There can be problems with a pedantic application of this theory in enforcement. Competition cases are becoming highly technical and time consuming. Working against tight deadlines, especially in merger cases, a dogmatic application of such concepts is bound to lead to errors. Thus, demand-side substitution (also supply-side substitution in some cases) is the right way to define relevant markets even with respect to multi-sided platforms. Although in the Immonet/Immowelt merger decision the 60 Sebastian Wismer & Arno Rasek, (n 28) 57. 61 Amazon Annual Report, 2014 <https://ir.aboutamazon.com/static-files/d6263104-b6fa-401a-aa29-3b66ec713f76> (accessed on 11 May 2019). 62 Andrei Hagiu and Julian Wright (n 7). 63 Wright and Yun also note that even though there is a lack of consensus on the definition of two-sided platforms, generally they are characterized by cross-group effects, or indirect network effects. Wright and Yun, (n 27), footnote 1.
Bundeskartellamt referred to the online real estate platforms as a type of "so called transaction platforms" with pronounced bilateral effects, it also resorted to the demand-side substitutability test to define the relevant market. 64 The Bundeskartellamt deemed it proper to define a single market if demand on the part of both user groups is largely uniform and the possibilities of the user groups to switch provider do not essentially differ. Here the opposite market side consists of property providers and property seekers which both use a property intermediation service. With all the feasible possibilities to switch to an alternative intermediary, both user groups would inevitably be brought together again. 65 Interestingly, in its 2019 Facebook decision, the Bundeskartellamt refrained from using the previous terminology of "transaction platform" or "matchmaking platform" while defining the relevant market and instead relied on demand-side substitution. The Bundeskartellamt observed that "[b]ased on the concept of demand-side substitutability, the Bundeskartellamt defines the product market as a private social network market with private users as the relevant opposite market side". 66 The recently issued Abuse of Dominance Enforcement Guidelines by the Competition Bureau Canada also avoids a predefined approach to the relevant market definition in the case of multi-sided platforms. 67 In order for the courts and antitrust authorities to properly undertake a factual enquiry, it is important not to have any preconceived notions about platforms.
Evans and Schmalensee, while avoiding the nomenclature and conceptual foundations used by Filistrucchi et al. for "transaction markets", observe "platforms that provide services that, by their very nature, are jointly and unseverably consumed by two different types of consumers" offer essentially a single service that competes with another similar service provided by a different platform. 68 Even Evans and Schmalensee suggest that platforms that "jointly and unseverably consumed by two types of customers" do 64 BKartA -Immonet/Immowelt, B6-39/15 (n 29). 65 Ibid 2. 66 The Bundeskartellamt, Facebook case summary (n 5) 3. 67 In paragraph 16, the Competition Bureau Canada observes, "Depending on the facts of a case, the Bureau may define a product market as one side of a multi-sided platform (i.e. consider the effects of a price increase on one side of the platform)". … In other cases, the Bureau may view it appropriate to define a market to include multiple sides of the platform <https://www.competitionbureau.gc.ca/eic/ site/cb-bc.nsf/vwapj/CB-ADEG-Eng.pdf/$file/CB-ADEG-Eng.pdf> (accessed on 13 May 2019). 68 The authors provide the example of an online reservation service provided to diners and restaurants. They note that the service these two groups consumer is reasonably interchangeable with services pro- not compete with single-sided businesses. 69 This approach invites the competition authorities to take a factual enquiry to identify if a service is actually "jointly and unseverably consumed by two types of customers" and refrains from devising pigeonholes that prejudice the highly factual relevant market enquiry that can reveal commercial realities. 70 In general, too much formalization by suggesting a "bright-line" test to explain market phenomena may complicate legal enforcement, rather than introducing legal certainty. Instead, the effort of economics should be to explain the underlying economic reasons that have bearing on legal enforcement. 71 As argued above, with respect to multi-sided platforms the presence of indirect network effects is the prime factor that influences competition law.
In general, a pedantic approach to two-sided markets may also create frictions between different laws. Advocate General Spuzner, in his Opinion with respect to Uber, stated that as a platform Uber exercises "decisive influence" by way of "indirect control" over the drivers, unlike intermediation platforms such as those that are used to make hotel bookings or purchase flight tickets, and hence the platform and drivers are in an employer-employee relationship. 72 By this rationale, Uber is a onesided market, as there are no two different groups of consumers. This illustrates that the economic classification of market phenomena into certain categories with pre-defined features and hence calling for a particular treatment in law is not the right approach. While for the purpose of facts at hand in Spuzner's Opinion, the element of control was decisive in order to decide the relationship between Uber and drivers, for the purpose of deciding market definition and measuring market power in competition law basic elements such as substitutability, presence of indirect network effects and multihoming are relevant.
Indeed, the concepts and definition related to two-sided platforms are still unsettled and it may be too early to devise a general framework with respect to two-sided platforms. 73 Any economic test needs to be strength tested first keeping in mind different enforcement scenarios. In view of this, the Majority's attempt to accommodate the "transaction" market approach is premature at best.

Lessons for law
The US Amex case also raises critical questions about the nature of optimal interaction between economics and law. As an organic discipline law borrows insights from multiple streams including economics. It is still, however, not clear as to the standards that an economic theory must conform to before it starts informing (at times shaping) the legal enforcement. So far this paper showed the limitations in the "transaction" platform approach adopted by the US Supreme Court that had originally been proposed by an academic paper. The following part demonstrates the problems that a hasty adoption of a particular economic theory can cause for legal enforcement.

"[H]ow courts should use new developments in economic or scientific theory" 74
The adoption of the "transaction" market approach by the US Supreme Court raises the same question that Brodley and Hay raised almost forty years ago with respect to the swift adoption and subsequent disenchantment with the Areeda-Turner test by several courts. 75 Professors Areeda and Turner proposed their test to identify predatory pricing in their seminal paper in 1975. 76 Areeda and Turner proposed Average Variable Cost (AVC) as the proxy for marginal cost and posited that prices below reasonably anticipated average variable cost are predatory and prices at or above reasonably anticipated average variable cost are non-predatory. 77 Several leading contemporary commenters had critiqued the test. 78 This, however, did not 73 Janusz A. Ordover, 'Comments on Evans and Schmalensee's "The Industrial Organization of Markets with Two-Sided Platforms"' (2007) 3(1) Competition Policy International 181, Ordover observes, "Thus-like free-riding or network effects were before-2SPs [two-sided platforms] may be a passing concept which calls for analytical vigilance but does not require a policy revolution". prevent some lower courts from adopting the test. 79 The SC, however, never adopted this test and in the landmark Brooke Group case in 1993, instead of relying on AVC, as proposed by the Areeda-Turner test, the Court instead held that "a plaintiff seeking to establish competitive injury resulting from a rival's low prices must prove that the prices complained of are below an appropriate measure of its rival's costs [emphasis added]". 80 The US Supreme Court has repeatedly declined to benchmark the type of cost below which the prices will be considered predatory. 81 Consequently, lower courts have been following different standards of cost to determine predatory pricing. 82 This is not to say that an economic test has to be foolproof. As is the case, the Areeda-Turner test has received criticism even after being adopted by several courts. 83 The early adoption of the Areeda-Turner test was premature for the reasons that first, it did not reflect an economic consensus; second, as opposed to what appeared initially, the test was difficult to apply legally. For these reasons, some courts chose not to rely upon the test. 84 Moreover, this test was too demanding for the plaintiffs and made proving predation extremely difficult. 85 Incidentally, proving harm on the aggregate if an allencompassing relevant market is defined for a platform also plunges the plaintiff into a herculean task. Here again, a wrongly defined single market based on a flawed "transaction" market test will make it difficult for the plaintiff to prevail before antitrust authorities and courts.
In the EU, the Court of Justice endorsed the Areeda-Turner test in 1991 in AKZO holding that prices below AVC are irrational and thus unlawful. 86 The Court also developed the test further and held that prices above AVC but below Average Total Cost (ATC) are unlawful if they are proved to be part of a plan to eliminate a competitor. 87 Subsequently, the Commission in its Guidance Paper on Article 102 TFEU has shown the preference for choosing Average Avoidable Cost (AAC) as the appropriate lower benchmark cost for determining predation. 88 Indeed, in Post Danmark 89 the CJEU appeared to adopt the AAC benchmark. 90 It is also important to note that as opposed to the Areeda-Turner test that provided the long desired cost benchmark that restricted judicial discretion, the "transaction" platform test fulfils no such purpose. For this reason, it is even more important to refrain from crafting legal rules based on economic theory that do not drastically help the administration of competition law. 91 The above discussion raises an important question about the process through which a scientific or economic thought must pass before it matures into law and starts affecting social welfare. At this stage, it is better to reproduce Professor Stigler's statement made around fiftyyears ago that even Brodley and Hay had cited with respect to the Areeda-Turner test.
A new idea does not come forth in its mature scientific form. It contains logical ambiguities or errors; the evidence on which it rests is incomplete or indecisive; and its domain of applicability is exaggerated in certain directions and overlooked in others. These deficiencies are gradually diminished by a peculiar scientific aging process, which consists of having the theory "worked over" from many directions by many men. This process of scientific fermentation can be speeded up, and it has speeded up in the modern age of innumerable economists. But even today it takes a considerable amount of time, and when the rate of output of original work gets too large, theories are not properly aged. They are rejected without extracting their residue of truth, or they are accepted before their content is tidied up and their range of applicability 87 Ibid, paragraph 72. In general legal rules based on economic theories pose challenges.
… while technical economic discussion helps to inform the antitrust laws, those laws cannot precisely replicate the economists' (sometimes conflicting) views. For, unlike economics, law is an administrative system the effects of which depend upon the content of rules and precedents only as they are applied by judges and juries in courts and by lawyers advising their clients. Rules that seek to embody every economic complexity and qualification may well, through the vagaries of administration, prove counterproductive, undercutting the very economic ends they seek to serve. ascertained with tolerable correctness. A cumulative slovenliness results, and is not likely to be eliminated until a more quiescent period allows a full resumption of the aging process. 92 While there cannot be mathematically calibrated standards to determine a "peculiar scientific aging process", it is safe to say that a reasonable passage of time allows a theory to be critiqued and reformed. The academic work that the SC relies upon to support the "transaction" platform approach to market definition appeared in 2014. 93 Indeed, the concept of "transaction" platform is new to competition law that was incorporated into law without critical scrutiny either by academics or by courts. 94

Review standards for new economic theories and tests
The following part presents an enquiry into the learning process that the EU competition law institutional framework adopts vis-à-vis new economic insights based on academic work. This exercise is aimed at exploring if new economic theories pass through a "scientific aging process" before they become law in the EU.

The European Commission
There are two channels through which economic insights enter EU competition law. First, the EU competition law framework incorporates economic teachings through hard law (block exemption regulations 95 ) and soft law (guidelines 96 ) instruments. Before regulations and guidelines are finalized they go through extensive debate and discussion process. Second, through the decisions of the Commission. Even in the second channel, economic insights are well contested both by the Commission and by opposite parties before they become part of the law. Ever since the changes brought by Commissioner Monti in 2003, including the creation of the post of the Chief Competition Economist (CET), the Commission has strived to ensure that EU competition law is fully compatible with economic learning. 97 DG Competition is the first stage where the economic assessment is done. DG Competition is an administrative authority and arguably is better at complex economic assessments and empirical analysis compared to a judicial system as it has in-house expertise, a general acceptance of economics and the opportunity for lawyers and economists to work together. 98 It has been suggested that an adversarial system is more suited to the assessment of economic evidence where it may go through the process of assertion and refutation. 99 Remarkably, the procedure before the Commission cannot be termed as purely inquisitorial in that parties involved can request meetings with the case team and submit documents making particular claims. 100 Over time, the Commission has gained experience in evaluating complex economic evidence. Various checks have been suggested in order to determine the relevance and significance of economic analysis presented before the Commission. 101 Parties are required to explain the methodology and data they use in their economic evidence. 102 It is also incumbent upon the parties to show that their technique (economic or econometric analysis) has been generally accepted in the scientific community. 103 A new economic theory, therefore, can be put through optimal scrutiny where the parties and the Commission have the opportunity to exchange views on the same.

Judicial review
The General Court is the second stage where the economic evidence and methodology interact with the legal process. In the EU, the Commission enjoys broad discretion in the matters involving complex economic assessment. 104 Thus, the Court limits its review to verifying whether the relevant rules on procedure and on the statement of reasons have been complied with, whether the facts have been accurately stated and whether there has been any manifest error of appraisal or a misuse of powers. 105 In general, economists and economics in court proceedings have been less influential in court proceedings as compared to the Commission. 106 Even in the cases where the Court examines economic arguments, it rarely performs its own independent economic analysis and instead relies mostly upon the economic assessment of the Commission or the economic arguments advanced by the parties. 107 Aside from economic evidence, what is the level of judicial review with respect to new economic theories? Clearly, economic theory or technique is not a piece of evidence. Rather, evidence is required to support the application of a particular economic theory. While the General Court can undertake a comprehensive relook at the facts that underpin economic theories advanced by the Commission, 108 this form of check-and-balance is seemingly unavailable for assessing the appropriateness of economic theory. The following comment by the CJEU underlines the nature and extent of review by the General Court with respect to complex economic matters. … The Court must therefore limit its review of such an appraisal to verifying whether the relevant procedural rules have been complied with, whether the statement of the reasons for the decision is adequate, whether the facts have been accurately stated and whether there has been any manifest error of appraisal or a misuse of powers. 109 The proceedings before the General Court are adversarial in nature where the parties are given the chance to contest the evidence. 110 In some cases, parties make references to economic literature before the General Court. Parties also have the possibility to use economic experts at the oral hearing stage before the General Court. 111 It is, however, not clear if the General Court will unpack and examine an economic test or technique based on an academic work that the Commission has relied upon. Bo Vesterdorf, who was the president of the General Court (then CFI) from 1998 to 2007, also suggests the duty and responsibility of the General Court with respect to new theories is limited to scrutinizing the nature of the evidence relied upon in support of such theories. 112 It appears that an extensive review of new economic theory already adopted by the Commission may be inhibited by the standard of limited review for complex economic assessments that the courts have been following. Any such review is limited to the test of "manifest error". 113 For instance, the General Court in Wanadoo upheld the choice of a particular method used by the Commission to calculate the rate of recovery of cost. 114 In this case, the appellant had argued that the Commission had modified its "cost of recovery" test. 115 The Court held that As a preliminary point, it should be recalled that, as the choice of method of calculation as to the rate of recovery of costs entails a complex economic assessment on the part of the Commission, the Commission must be afforded a broad discretion. The Courts review must therefore be limited to verifying whether the relevant rules on procedure and on the statement of reasons have been complied with, whether the facts have been accurately stated and whether there has been any manifest error of appraisal or a misuse of powers. 116 Some of the cases where the courts in the EU laid down economic tests are conditions for collective dominance in the Airtours case, predatory pricing test in Akzo and Tetra Pak II, the refusal to supply test in Oscar Bronner, IMS and Magill, and the rebate test in Michelin and British Airways. 117 Barring the test for predation, however, none of these tests owes its genesis to one particular academic work. The EU courts define, modify and alter economic tests based on the prevailing economic assumptions which command a common economic consensus of that time. These concepts, therefore, already go through the process of criticism and reform before they translate into law. Accordingly, legal tests based on such economic tests are less prone to quick changes. The Commission first mentioned collective (joint) dominance in Alcatel/AEG Kabel merger decision. 118 Since then the Community Courts clarified and developed the test to identify collective dominance. 119 Likewise, test for rebates 120 and refusal to supply 121 have also witnessed gradual evolution and clarification over a period of time.
In general, courts are more deferential to peer-reviewed academic study as compared to a study conducted for the purpose of litigation. 122 Moreover, the court may have a tough time analysing a recent economic theory based on academic work, as there may not be much academic literature critiquing that particular theory.
It seems that new economic insights and new theories more easily make entry into the US legal system as compared to the EU system. In contrast to the judicial review of matters involving economic analysis in the EU, where the General Court exercises restraint, judges in the US exercise far greater power and discretion while analysing economic theories and concepts. 123 The institutional design of the antitrust enforcement provides which has been able to attract only minimal support within the community may properly be viewed with skepticism". 133 Thus, if the expert testimony relies on a study or a particular technique, the same can be challenged as per the Daubert Standard. Indeed, the validity of an even famous study can be challenged. 134 In the AmEx case, there was no occasion to test the "transaction" market approach against the Daubert Standard either at the District Court or at the Appeal Court stage. It was argued back in 2006 that the EU could adopt a similar standard to Daubert to analyse economic evidence. 135 The above discussion illustrates that in the EU a new economic theory can be put through rigorous review at the Commission. Once, however, this theory is approved by the Commission, it is unlikely that the General Court will alter that theory in totality. Also, there may be an inclination on the part of economists working at antitrust bodies to leave the analysis of the merit of new theories to academic journals, where a new theory can go through rigorous peer review to ensure consistency and empirical relevance. 136 However, as this paper shows an academic study as well warrants proper scrutiny before its legal adoption, especially in those cases where a particular theory is still new. In the EU the Commission can carry out this task properly. In comparison to the EU, the US legal system gives enormous discretionary powers to judges who at times use their own thoughts and independent learning to incorporate economic theories. All in all, this makes the US system more prone to the vagaries of untested economic insights.

Conclusion
Multi-sided platforms are becoming ubiquitous in the modern economy. It is, therefore, crucial that the competition law community correctly understands and accordingly devises implementable rules to ensure consumer welfare in the platform economy. While economics has enriched our understanding of platforms, much still needs to be discovered. Indeed, the rising number of academic papers on multi-sided platforms indicates the heightened efforts to clearly understand the platform competition. This paper dealt with one such academic effort to classify platforms into "transaction" and "non-transaction" platforms for the purpose of market definition, which was swiftly elevated into law by the US Supreme Court in its Amex decision. It has been posited that in the case of "transaction" markets, there is one all-encompassing relevant market comprising of all sides of a platform. Whereas, every side of a "non-transaction" platform needs to be defined separately.
The paper demonstrated that the test underpinning the identification of a "transaction" platform is flawed for two reasons. Firstthe requirement for the observability of a transaction between two different sides of a "transaction" platform fails to account for the different purposes for which two sides come together on a platform. As shown in the paper, the interaction between two sides does not take place only for the reason of exchange of money for services/products. Some advertisers reach consumers through platforms merely for creating brand awareness. Secondas opposed to what the US Supreme Court and the academic work it relied upon posited, some "transaction" platforms also compete with a single-sided market. The paper cited some such real and also theoretical examples to identify holes in the "transaction" platform approach to market definition. The paper also demonstrated that some alternative methods to classify platforms into predefined groups for the purpose of market definition are mutually incoherent and thus legally repugnant. Overall, any predefined categories for the purpose of market definition also fail to account for disruptive innovation that is known to blur the market boundaries.
Based on the above, the paper suggested lessons for economics and law. In view of the imperfections of pre-defined approaches to correctly identify relevant markets with respect to platforms, regard should be had to demand-side (and at times supply-side) substitution to identify the relevant market. Additionally, such imperfect tests also illustrate the limitations of "bright-line" economic tests often proposed to help antitrust authorities and courts. Instead, the optimal solution is to explain the underlying economic reasons that can facilitate a case-bycase analysis.
The final part of the paper showed the problems that the hasty adoption of an economic test can cause and argued that a new economic theory must go through a "scientific aging process" before it starts shaping law. A gradual process of criticism and reform strengthens a theory and increases its utility. An examination of the EU competition law institutional framework suggested that a new economic theory can be subjected to rigorous scrutiny only at the Commission. The US system, in contrast, allows a broader doorway to an uncontested economic theory to inform the law.