Tag: chapter-6

Author’s Note*

Introduction

There is an astounding amount of data generated every moment.[1] While the use of data has always been essential for a well-functioning market economy, the current scale, scope, and speed at which data is collected, organized, and analyzed is unprecedented—leading to the apt label of “big data.”[2] Importantly, an entire infrastructure has been built around data including cloud computing, machine learning, artificial intelligence, and the 5G wireless network.[3]

In this chapter, we discuss the role of big data in antitrust with a particular focus on the rise and success of large digital platforms—as the importance of data has caught the attention of various reports on the digital economy[4] and also raises issues involving privacy.[5] We begin with an economic overview of data and what sets it apart from other firm assets. In this discussion, we will also assess the question of whether big data represents a sizeable barrier to entry that hinders the ability of entrants to compete on equal footing.[6]

Next, we discuss data in the context of network effects. While network effects traditionally are associated with a network of people, some have hypothesized that the advantages that data confers to platforms can also be framed as type of network effect. The idea is that data creates a feedback loop of more data, which ultimately results in an impregnable bulwark against competition. We aim to demonstrate that, while there are commonalities between using data and network effects, there are important differences and distinctions worth highlighting.

We also explore a number of relevant legal considerations involving big data and antitrust. Generally, should courts administer cases that involve big data differently? Should there be a stronger presumption of market power when a large platform possesses big data? On the other side of the coin, can combining big data assets result in merger-specific efficiencies?

Finally, big data is increasingly discussed as a potential remedy for competition problems involving platforms. Proposals range from allowing users to more easily port their data across platforms to forced data sharing and interoperability.[7] We examine the incentive effects from imposing such remedies and potential unintended consequences.[8] Ultimately, these proposals are putting the cart before the horse, as remedies without an actual showing of an antitrust violation is not antitrust enforcement but sector regulation.

I. Understanding Big Data

There is a long tradition of using data to increase efficiency and create value. Mid-twentieth century grocery giant A&P used customer data, inter alia, to discern regional preferences and to forecast demand in order to reduce food waste.[9] An article from Harper’s Magazine in the late 1980s used the term “big data” to refer to the variety of customer data lists used to increase the effectiveness of direct mail advertisements.[10] The article found that direct mail services had access to data from credit-reporting agencies, magazine subscriptions, catalog purchases, customer questionnaires, real estate records, and other sources that could sketch a picture of the likely preferences of millions of individuals.

Today, continuing in the tradition of A&P, Walmart analyzes over 2.5 petabytes of data every hour in its Data Café in Bentonville.[11] In addition to helping it personalize advertisements, Walmart’s data analysis helps it run pharmacies more efficiently, speed up checkout, manage its supply chain, and optimize placement on store shelves.[12]

Yet, the sense that advances in data science are leading to an “overwhelming deluge of information” is not a new phenomenon.[13] That feeling traces back, at least, to the “avalanche of numbers” that occurred in the 1820s and 1830s, when national governments began accumulating more data than ever before to classify and tabulate information about the population in an attempt to improve governance.[14] Sociologist David Beer explains that, while “features of the current data movement are in some ways novel,” data has been scaling up for hundreds of years.[15]

Not only do firms have more data than ever, there are increasingly sophisticated methods for analyzing it including the use of machine learning and artificial intelligence. Routinely, firms collect and store data from incongruent sources such as website traffic, credit card purchases, smart phones and wearables, and numerous other data streams to create a profile of their user base.[16] These advances benefit consumers by, among other things, allowing firms to improve products and lower costs, and enabling the growth of the free-to-consumers model widely employed in the online space.[17] The growth in the importance of data has also led to concerns that incumbent firms’ access to data can create barriers to entry in many industries. We address these issues below.

A.  Role of Big Data in the Production Process

In digital markets, while there is little doubt that big data represents an important factor of production, it is not an end in and of itself.[18] Just as labor, innovation, capital and entrepreneurial skill differentiate firms, so too does the ability to turn “big data” into something useful. For example, a website might track and store every click that its users make while on the site. This data, in turn, could allow the website to better tailor advertisements to a specific user’s interests; determine which features to promote and which features to drop; and improve the overall design of the site.

As in other areas of business, some firms are more proficient than others to use and organize assets, as the value of big data can only be unlocked when combined with other inputs. Thus, rather than simply assuming that the sheer volume of data is what explains outcomes, it is also worth considering whether it is the skill and talent needed to combine the data with other inputs to produce something of value which differentiates firms.

Firms are also differentiated in the characteristics of their final product and in the mix of inputs used in their production processes. Thus, to produce a given level of quality, one firm might use a mix of big data, intellectual property, and highly skilled labor, while another firm might achieve a similar level of quality, using the same set of ingredients but in different proportions—relying on its particular comparative advantage. Professor Harold Demsetz observed that conditions frequently considered barriers to entry, such as scale economies, capital requirements, and advertising expenditures, are not the fundamental source of barriers; the fundamental barriers are rather the cost of information and the uncertainty that an entrant has to overcome.[19] In other words, it is not big data per se that represents the barrier to entry, but rather what big data helps a firm accomplish. This point is consistent with the observation that, over time, incumbents inevitably change how they combine their various inputs to achieve their levels of output and quality. For instance, an incumbent might have originally entered with a lower cost curve, a superior algorithm, or a valuable patent but, over time, to improve its product, it uses big data more extensively than when it first entered the market. Similarly, an entrant might initially operate with “small” or “medium” data but improve quality over time as its installed base of users grows.[20]

The larger point is that the use of big data is not, in of itself, an indication that big data is required to compete effectively or that having significantly more data results in an insurmountable competitive advantage.[21] Consequently, for the purpose of competition policy, it is important to consider why a product produced with the aid of big data might be successful. For example, an innovative algorithm could be the primary reason a product is successful in gaining and retaining consumers. Superior design can also make the difference between a successful product and an unsuccessful one. Even if the use of big data is the primary reason for a firm’s success, a relevant question is whether comparable, but not necessarily equivalent, data are costly to acquire.

B.  Is Big Data a Barrier to Entry?

The prior discussion leads one naturally to a question of whether big data is itself a “barrier to entry.” While a useful shorthand, it is not always entirely clear what constitutes a barrier to entry—as the term has a long history in economics with various scholars emphasizing different aspects of entry in order to create a definition. Below, we briefly summarize some of the key developments.[22] Ultimately, rather than focus on definitions per se, it is more pertinent to directly analyze the specific entry conditions for each relevant market; although, walking through the key developments can give useful insights into that analysis.

In the 1950s, Professor Joseph Bain defined barriers to entry as structural factors that allow incumbents to persistently price above the competitive level without incurring the threat of entry.[23] Thus, Bain considered the following to be examples of barriers to entry: economies of scale that require large capital expenditures, product differentiation, and absolute cost advantages. About a decade later, Professor George Stigler considered barriers to entry as costs that an entrant must incur but incumbents do not.[24] Examples include patents and grandfathered government regulations but not economies of scale to the extent that an entrant has access to the same cost function. The appeal of Stigler’s definition is its recognition that incumbents can earn supra-normal profits over the long-term only if they have some persistent advantage over potential rivals. What is missing from both Bain and Stigler, however, is an assessment of welfare.

About a decade after Stigler, both Professors Franklin Fisher and C.C. von Weizsäcker filled this void with normative definitions that incorporate social welfare.[25] Fisher found “a barrier to entry exists when entry would be socially beneficial but is somehow prevented.”[26] Similarly, von Weizsäcker explained, “a barrier to entry is a cost of producing which must be borne by a firm which seeks to enter an industry but is not borne by firms already in the industry and which implies a distortion in the allocation of resources from the social point of view.”[27] If economies of scale can increase overall welfare and we associate entry barriers with inefficiencies, then, von Weizsäcker asks, “in which sense can we speak of a barrier to entry?”[28] According to Fisher, “the right issue is not whether there are barriers to entry into the production of a particular mousetrap, but whether there are barriers to entry into innovation in mousetraps.”[29]

The tension in defining barriers to entry is that there are really two ways in which the term is discussed in the context of antitrust. As Professor Dennis Carlton clarifies, “Trying to use ‘barriers to entry’ to refer to both the factors that influence the time it takes to reach a new equilibrium and to whether there are excess long-run profits is confusing.”[30] Therefore, for the purpose of competition policy, Carlton recommends that “rather than focusing on whether an entry barrier exists according to some definition, analysts should explain how the industry will behave over the next several years . . .  [which] will force them to pay attention to uncertainty and adjustment costs.”[31]

Consequently, we find that it is best to avoid suggesting that big data is or is not a barrier to entry.^[32] Rather, the use of data is one potential factor when examining “the timeliness, likelihood, and sufficiency of entry efforts an entrant might practically employ.”[33] There are clearly impediments that an entrant must overcome in order to compete effectively. Common examples include regulatory compliance costs, expenditures on specialized equipment, developing intellectual property, and hiring skilled labor. Some obstacles are nominal. Some obstacles are substantial. Attempting to classify these impediments as entry barriers or not creates the confusion mentioned by Carlton.

It is also relevant to note that big data is not an exogenous factor that dictates the number of firms in a market, which in turn determines the degree of competition and the rate of return. Rather big data is endogenous, as are other dimensions of non-price competition.[34] For instance, if a firm invests heavily in research and development, which allows it to introduce a new product or to substantially improve an existing product, we would not normally view this as anticompetitive conduct or even conduct that ultimately leads to anticompetitive results. Rather, we would consider investment in innovation to be procompetitive. Similarly, investments in big data can create competitive distance between a firm and its rivals, including potential entrants, but this distance is the result of a competitive desire to improve one’s product. Moreover, the observation that a firm is making large margins gives no indication whether this reflects supra-competitive pricing if we properly consider the rate of return required over the whole production process, including its investment in big data.[35]

Even if big data represents an important component to the success of an incumbent, entrants can differentiate their products along other dimensions important to consumers. For instance, a grocery store entrant might focus more on carrying locally made produce or products that cater to specific diets. An online firm might focus more on building network effects or greater integration with complementary products rather than the use of big data. Additionally, a social media platform could focus narrowly on a particular format and demographic in order to expand its user base.[36] As Tucker & Wellford explain, “[t]he fact that some established online firms collect a large volume of data from their customers or other sources does not mean that new entrants must have the same quantity or type of data in order to enter and compete effectively . . . [L]ack of asset equivalence should not be a sufficient basis to define a barrier to entry.”[37]

In assessing the prospect of entry into any given market, an important consideration is the actual history of entry in that market.[38] The evidentiary value of prior instances of entry, however, depends upon the extent to which current entry conditions are similar to prior entry conditions. The fact that entry occurred previously does not establish that entry is currently easy. On the other hand, one has to be cautious before inferring that long periods without entry implies that entry would not occur if incumbents raised prices or reduced quality and innovation.

With these caveats in mind, we highlight some recent entry episodes, where the entrants successfully overtook incumbents with arguably big data. While the examples of Google disrupting Yahoo and Facebook displacing MySpace are well-documented,[39] they are far from the only instances of an incumbent with a seemingly significant big data advantage losing market share to a newcomer.

Apple’s iTunes is an example of a once powerful incumbent that was dethroned from its market leading position. Started in 2001, iTunes sold digital copies of songs, and eventually audiobooks, eBooks, movies, and television shows.[40] By 2010, Apple’s iTunes had roughly 70 percent of the U.S. online music market.[41] With years in the market, along with the massive sales of Apple iPhones, iPods, and Mac computers, the company inevitably amassed large amounts of data.

That, however, did not stop Spotify from entering the market in 2008.[42] Along with others, Spotify changed the online music industry by offering a music streaming service, where customers could listen free of charge with advertisements or pay for a premium service with the ability to download songs and make their own playlists.[43] One might argue that Apple’s iTunes is not exactly the same product as Spotify’s streaming service; consequently, it is not an apples-to-apples comparison. However, Spotify’s differentiation from iTunes actually proves the larger point that markets evolve in ways that are not easily forecasted. Apple’s recent decision to discontinue iTunes in favor of its Apple Music product, among others, further proves the point. Spotify’s pivot towards a streaming service caused rivals to launch similar services to more effectively compete, such as, Apple Music and Amazon Music. Thus, despite being a relative newcomer in the market, and taking on the incumbent Apple, Spotify currently has double the number of subscribers compared to Apple Music.[44]

Similarly, at the start of 2010, Internet Explorer and Firefox had a combined 86 percent market share in web browsers. Yet, between 2011 and 2012, Chrome overtook both Firefox and Internet Explorer as the predominant web browser even though Chrome had a 6 percent share at the start of 2010.[45] Given that web browsers collect a tremendous amount of data on user behavior,[46] under a big data-centric view of entry barriers, Firefox and Internet Explorer should never have relinquished their market leading positions.

These episodes should perhaps not be surprising as Professors Leslie Chiou and Catherine Tucker find little empirical evidence that the possession of historical data provides an advantage to firms, in terms of their market shares.[47] As Schepp & Wambach point out, “[T]he origin of many innovative start-ups illustrates that companies with smaller but possibly more specialized datasets and analytical expertise may be able to challenge established companies.”[48] Importantly, referencing these prior episodes and research are not intended to suggest that data cannot confer a competitive advantage to digital platforms; yet, there should be no presumption that it creates an insurmountable barrier to entry and the associated conclusion that there is a lack of competition due to data.

II. Data Driven Network Effects

We next move to a discussion of data in the context of network effects—as there is a view that the advantages of big data can be framed as a data-driven network effect.[49] The idea is that the collection and use of data creates a feedback loop of more data,[50] which ultimately insulates incumbent platforms from entrants who, but for their data disadvantage, might offer a better product. We aim to establish that, while there are commonalities between using data and network effects, there are important differences and distinctions that should be considered.

The data-driven network effect argument is summarized in numerous reports on the digital economy.[51] The central idea is that having more data begets more data—all the while improving the quality of the platform for all the groups involved including users and advertisers, which causes them to use the platform more. The apparent concern from the reports is that these data advantages create an unstoppable snowball effect, which prevents or hinders new entry since entrants cannot hope to match the quality of incumbents due to their data advantages.[52]

While this theory of incumbent strength is intuitively appealing, there are a number of relevant points to consider as it applies to antitrust enforcement and policy. First, assuming arguendo that this effect exists, it is entirely based on increased platform quality and innovation. While an increased level of quality and innovation makes life considerably more difficult for rivals, this is an example of meritorious competition that antitrust laws should not be disincentivizing. Further, the primary effect will be on undifferentiated entry, that is, entry based on trying to exactly replicate the incumbent’s product, rather than differentiated entry, which relies less on the type of data that the incumbent has an advantage over.

This also puts a finer point on von Weizsäcker’s work on barriers to entry.[53] What is the point of defining barriers to entry if welfare-enhancing activities like improving a product now constitutes a “barrier to entry”? What purpose does the definition serve? Such a broad, and indiscriminate, use of the term does not differentiate between welfare-enhancing or reducing activities. It would seem that all activities that create a competitive advantage become a barrier to entry including innovations from R&D; learning by doing and trade secrets;[54] and hiring skilled labor and training them to increase productivity. We could add the development and maintenance of strong brand names and lower marginal costs. It seems that every conduct or practice that separates a firm from the pack could be classified as a barrier to entry and, consequently, would imply a poor outcome for markets.[55]

Second, while no one would seriously dispute that having more data is better than having less, the idea of a data-driven network effect is focused too narrowly on a single factor improving quality. As mentioned in supra Section I.A, there are a variety of factors that enter a firm’s production function to improve quality. For example, by all accounts, Google’s own entry into search and its ability to displace the market leading incumbents Yahoo Search and Alta Vista was due to the superiority of its “PageRank” algorithm.[56] In a nutshell, the problem with data-driven network effects is that it presumes that having more data results in market success or, alternatively, incumbents enjoy their market position due to their size and data advantage.

Relatedly, even if data is primarily responsible for a platform’s quality improvements, these improvements do not simply materialize with the presence of more data—which differentiates the idea of data-driven network effects from direct network effects.[57] A firm needs to intentionally transform raw, collected data into something that provides analytical insights.[58] This transformation involves costs including those associated with data storage, organization, and analytics, which moves the idea of collecting more data away from a strict network effect to more of a “data opportunity.”[59] This data opportunity is a production opportunity based on residuals of consumption.[60] Firms that can take advantage of that opportunity and invest in innovation and quality achieve a competitive advantage—yet, as with all investments, there is no certain return.

Third, data has diminishing returns.[61] While it is ultimately an empirical and case-specific question regarding when diminishing returns sets in and its severity, this consideration should be a factor in all cases where data plays a major role. Additionally, depending on the specific market and platform, there are large, commercially available datasets on users,[62] which can supplement and bolster firm-specific data. Relatedly, there is an ever-increasing amount of “open data” that is being made available, which is public and private data that is available for use without an explicit license.[63] Perhaps one of the most important sources of open data is existing social media posts.[64] In short, there is a massive amount of data available, both free and paid, to complement firm-specific data.

III. Relevant Legal Aspects of Big Data

In this section, we explore a number of relevant legal considerations involving big data and antitrust. Generally, should courts administer cases that involve big data differently? Specifically, does the presence of big data imply a set of presumptions within the rule of reason framework—including a presumption of market power? For mergers, does the combination of big data assets result in cognizable efficiencies?[65]

Given the state of the current evidence—including a lack of a large body of prior cases and agency actions involving a central role for big data issues[66]—we believe there is an insufficient basis for having an antitrust presumption involving big data. Of course, this could change if the evidence mounts either in one direction or another. Yet, as has been previously discussed, big data is fundamentally about innovation—as a firm cannot gain a competitive advantage without some degree of effort and ingenuity to turn raw data into something that provides value. Thus, the presence of big data is more naturally categorized in a manner similar to R&D and other innovative activities. To the extent that these activities make it difficult for rivals to compete, then that is a potential consideration in terms of the degree of antitrust market power and the durability of that market power[67]—but it does not, in and of itself, suggest competitive harm.

Additionally, reflexive labelling of big data as a “barrier to entry” provides little guidance to courts as the term has no clear meaning in antitrust law.[68] A brief survey of recent antitrust cases confirms this observation that the current treatment of entry barriers remains relatively perfunctory and lacking in clarity. For instance, conclusions about what constitutes an entry barrier verge on the contradictory.^[69] Even in cases where discussion of entry barriers is more thorough, the lack of a universal approach leaves courts with a great deal of uncertainty when it comes to deciding the question.^[70] Thus, in terms of big data, courts should bypass labels and determine the precise role that big data plays in a given competition matter.[71]

In terms of efficiencies, there are compelling arguments that mergers that bring together various data sets and/or improve the ability to analyze data should be recognized as cognizable efficiencies. Luib & Cowie detail a number of recent competition matters involving the recognition of big data efficiencies.[72] The court in AT&T-Time Warner recognized that “the combined company can use information from AT&T’s mobile and video customers to better tailor Time Warner’s content and advertising to better compete with online platforms.”[73] In CVS Health-Aetna, “[w]e were able to show that the combination of CVS Health and Aetna would lead to better-integrated medical and pharmacy data.”[74] Also, in 2010, data efficiencies played a role in the Department of Justice’s decision to close its investigation into Microsoft and Yahoo’s search alliance agreement, which involved integrating, to a degree, their search algorithms, advertising assets, and data.[75] The DOJ explicitly invoked the role of data in providing efficiencies and improving Microsoft’s competitive position.[76]

In sum, the burden of production should remain with the plaintiff to demonstrate that the use of big data is a significant hindrance to the timeliness, likelihood, and sufficiency of entry and/or hinders the competitive process—as measured by the negative impact on consumer welfare. What should be avoided is a presumption that the mere presence of big data lowers the burden of production on the plaintiff to either meet its prima facie burden or its burden to rebut the defendant’s efficiency justifications. In contrast, what recent competition cases have illustrated is that big data can play an integral role in realizing efficiencies post-merger.

IV. Data Remedies

Finally, when it comes to data and antitrust, there is increasing momentum to use data and the related platform infrastructure as potential remedies for anticompetitive conduct.[77] These remedies typically include proposals for data portability, data sharing, and interoperability. While these proposals are putting the cart before the horse, in this section, we broadly examine the incentive effects from imposing such remedies and the potential unintended consequences.

The most basic proposal to alleviate the market leadership of various digital platforms is to allow platform participants, such as users and advertisers, to easily and seamlessly port their data to rival platforms. The idea is to lower switching costs and lock-in effects, which will facilitate the ability to multi-home or switch altogether.[78] Of course, there is a major difference between using this proposal as a remedy to address demonstrable anticompetitive harms versus as a proactive regulatory intervention in an effort to improve market outcomes. As a remedy, the benefits of this proposal are that it can be targeted at specific practices, users, and platforms. These benefits might not translate to a regulatory setting, where the statutory language will likely have to be broader. This potentially opens the door for creative means where a platform could adhere to a strict reading of the regulation but implement measures that undermine its effectiveness in actually lowering lock-in and switching costs. This concern is likely significantly mitigated when data portability is used as a remedy—again, as remedies are intended to solve a specific competitive problem and, thus, attempts to undermine the stated objective will more likely be flagged as an order violation. Finally, given that the industry has already moved towards data portability as a best practice, the value of this proposal as an antitrust remedy is more limited.[79]

Various digital reports also call for more aggressive remedies including forced data sharing and interoperability requirements. Data sharing is getting at the heart of the debate and involves taking a platform’s intellectual property and requiring them to share it with rivals. One initial observation is that data sharing raises at least a nominal concern that it would facilitate coordination and, consequently, would be detrimental to the welfare of consumers. Additionally, there are major questions whether data sharing would actually solve the problem it is trying to address—given that advantages in data do not necessarily translate into advantages in innovation and quality.[80] It would seem that a necessary condition to impose such a remedy is a finding that data is an essential facility.[81]

Relatedly, forcing platforms to be interoperable even further necessitates leading platforms to coordinate on the design and infrastructure of their platforms. While this is ostensibly appealing, it also raises concerns about “bad” coordination, that is, agreements that soften competition rather than sharpen it.[82] Further, “standardization” could be implemented in an overly technical and complex manner, where the intent is to actually hinder, rather than facilitate, entry. Finally, forced sharing—of any kind—has been demonstrated to dampen incentives and reduce innovation.[83]

In sum, the basic message is that regulations intended to fix the competitive deficiencies of a digital market are inevitably going to lead to unintended consequences, which can ultimately do more harm than good. Further, attempts to promote rival products through forced sharing and interoperability are highly likely to negatively impact dynamic incentives. Thus, these proposals, while well-intended, should only be seriously considered after a thorough analysis of the full benefits and costs. As a remedy, there certainly can be specific instances where these proposals could address the harm from proven anticompetitive conduct. Yet, even in those instances, agencies and courts should be aware of their inherent weaknesses and, thus, they should be implemented in the narrowest manner possible.

Conclusion

The sheer volume, velocity, and variety of data collected throughout an increasingly digital economy has given rise to the big data era. Big data is being used to innovate in many sectors of the economy including in healthcare, farming, and education.[84] Yet, the widespread use of big data has also given rise to antitrust concerns that big data is actually resulting in reductions in welfare by hindering competition and protecting big tech incumbents. In this chapter, we have examined the role of big data in antitrust with a particular focus on how it fits into the larger production process and whether it should be considered a barrier to entry. Further, we considered the idea that big data is part of a positive feedback loop where increased quality precipitates an increase in users which further increases quality and whether this is ultimately harmful. Finally, we considered whether there is a need to change our antitrust presumptions regarding market power and big data and the merits of various proposed remedies to the alleged big data problem. Ultimately, if big data alters the competitive process, then we should focus on the testable implications that emerge. What we should avoid, however, are shortcuts and regulation, or unwarranted antitrust intervention, based on the “bigness” of big data and perfunctory labels such as barriers to entry.