Introduction
In recent years, leadership at the Antitrust Division of the Department of Justice (DOJ) and the United States Federal Trade Commission (FTC) have discussed the agencies’ increased focus on competition issues raised by the use of data, including by digital platforms. As stated in 2022 by the then Assistant Attorney General for Antitrust, ‘new market realities demand new approaches to competition enforcement’. In addition, the 2023 Merger Guidelines, which provide the framework used by the DOJ and FTC in evaluating antitrust markets and competition, outline more explicitly the agencies’ focus on potential harm to competition in addition to increased prices. This includes assessments of potential ‘worsening terms along any dimension of competition’, including ‘quality, service, capacity investment, choice of product variety or features, or innovative effort’. This change in focus is important when assessing potential theories of harm involving digital platforms.
In addition, federal regulators have pursued several high-profile lawsuits against data-focused companies, such as Google, Meta (formerly Facebook), UnitedHealth Group, Amazon and Apple, alleging unilateral conduct that substantially lessened competition; for example, in Federal Trade Commission v. Amazon.com, the FTC and 17 state attorneys general alleged that the online retailer is a monopolist and engaged in illegal practices to maintain its market power, and that its ‘access to valuable shopper data’ allowed it to strengthen its alleged dominant position as a platform between sellers and consumers while ‘overcharging its customers’ and ‘degrading the services it provides them’.
In cases that allege monopolisation, before assessing a firm’s alleged market power, it is necessary to understand the nature of competition and define the antitrust market at issue. Below, we discuss how economic analysis, and the definition of the relevant markets, affects an assessment of competitive harm in data-focused markets.
The federal agencies have alleged that firms used data to enhance their market power, thereby lessening competition and harming consumers through both higher prices and non-price terms, such as quality of service. In doing so, the federal agencies have adopted theories of harm focused on access to specific types of data. We discuss below the specific data-focused theories of harm presented by the agencies and how economic analysis was used to assess these theories of harm in United States of America, et al. v. UnitedHealth Group.
Last, we look forward and consider the implications of emerging data-driven technologies on antitrust enforcement. The federal agencies have discussed their focus on investigating potential harm to competition in the context of the digital revolution, including increased data collection and the use of automated decision-making. We discuss algorithmic price-setting and the implications in assessing alleged coordinated conduct.
Analysis of antitrust markets in data-focused industries
Before a firm can be deemed to be a monopolist (or to have exercised monopolistic power), it is critical first to determine the relevant antitrust market for the product (or service) at issue. There are two dimensions to assess an antitrust market: (1) the product and potential substitutes customers can turn to in response to a price increase (or quality decrease); and (2) the geographical area in which competition for the relevant product (or service) takes place. The relevant market drastically affects the assessment of alleged competitive harm.
Two-sided technology platforms have unique features that should be considered when evaluating the relevant antitrust markets in these cases. These platforms provide products or services to multiple groups, or ‘sides’, that ‘benefit from each other’s participation’. Because there are two distinct sets of customers on each side of the platform, the analysis should consider the supply and demand conditions that affect each side of the platform and the nature of competition for each side. In addition, nascent competition and product development are important features in technology markets. The threat of entry through new technology or innovation can pose a partial competitive threat that needs to be considered in a market definition analysis.
In October 2020, the DOJ and 11 state attorneys general sued Google for violating Section 2 of the Sherman Act by ‘unlawfully maintaining monopolies in the markets for general search services, search advertising, and general search text advertising in the United States through anticompetitive and exclusionary practices’. The complaint alleges that Google entered into exclusionary agreements with wireless device manufacturers, carriers and browser developers, which resulted in Google becoming the default search engine for their products.
The government alleged the exclusionary agreements ‘deny rivals the scale to compete effectively with Google’ as ‘the volume, variety, and velocity of data gathered through active queries accelerates the automated learning of search and search advertising algorithms’. In other words, greater scale improves the quality of Google search and search advertising services over rivals, which in turn reinforces Google’s alleged market dominance. The government further claimed that Google’s conduct harmed consumers by reducing the quality of general search services (across dimensions such as privacy and data protection), thereby lessening their choices and impeding innovation.
A key area of debate in this case was the definition of the relevant markets:
- On the user side of the platform, the government proposed an antitrust market of general search services that includes general (horizontal) search engines (e.g., Google, Bing, Yahoo! and DuckDuckGo).
- On the advertiser side of the platform, the government proposed two antitrust markets: (1) general search text advertisements, which are primarily text that appears on the search engine results page; and (2) broader search advertisements, which encompass any advertisements shown on a search results page in response to a search query, including text advertisements, shopping advertisements and travel advertisements.
These relevant markets informed the government’s analysis of Google’s alleged market power; for example, the government contended that Google controlled almost ‘90 percent of all general search engine queries in the United States, and almost 95 percent of queries on mobile devices’. In addition, the government argued that ‘Google has 16 times more fresh search data than [Microsoft’s] Bing, its nearest competitor’.
Google’s economic experts criticised the US government’s market definition analyses as being overly narrow by omitting important competitors for both users and advertisers, namely specialised vertical providers such as Amazon, Yelp and Expedia, as well as social media sites. For users, Google argued that the government ‘distort[ed] the commercial reality that users routinely substitute other search providers for general search engines—such as Amazon when they shop, or Expedia when they travel’. That is, the relevant competitive conditions to assess the alleged conduct would be different if, instead of general search services, the relevant markets analysed included specialised search services, which Google argued were ‘many of [their] strongest competitors’.
On the advertising side of the market, Google argued that economic analysis and substitution patterns show that advertisers allocate their spending across Google, Meta, Amazon and others. As a result, Google’s expert concluded that these other options must be included in the relevant markets for advertisers.
Judge Amit P Mehta, who presided over the case, found that Google maintained an illegal monopoly in two antitrust markets: (1) general search services on the user side of the platform; and (2) general text advertising on the advertiser side of the platform. The court’s consideration of the relevant markets for each side of the platform highlights the unique characteristics of two-sided technology platforms that should be considered when determining relevant antitrust markets. On the user side of the platform, the judge relied on evidence in the case that ‘users can access Google’s rivals by switching the default search access point or by downloading a rival search app or browser. But the market reality is that users rarely do so’.
While assessing the search advertisement market proposed, the court concluded there was sufficient evidence showing that the strength and growth of entry by competitors in search advertising (e.g., Meta, Amazon, Apple) supported the conclusion that Google did not have monopoly power in this market. Judge Mehta claimed that the government did not prove that Google had ‘sufficient power in that market’ to support a violation of the monopolisation laws. However, the court did find that Google maintained an illegal monopoly in the proposed advertising market for general text advertising (e.g., text advertisements that are found on the results page from a user’s query).
The decision underscores the need for relevant market definition to be fact-specific.
Data-focused theories of harm involving two-sided platforms
The methodology for assessing competition between multi-sided platforms is a key change to the 2023 Merger Guidelines. The Guidelines lay out the agencies’ approach to assessing potential harm to competition from access to or control of data by these platforms. Specifically, the Guidelines point to four types of mergers involving platforms that would potentially lessen competition:
- mergers between a ‘dominant’ platform operator and ‘smaller competing platforms’;
- acquisitions that deprive rivals of network effects, which ‘may weaken rival operators or increase barriers to entry and expansion’;
- acquisitions of companies that ‘help sellers manage listings on multiple platforms, or software that helps users switch among platforms’; and
- mergers involving a firm that provides inputs (e.g., data) that ‘may enable the platform to weaken rival platforms by denying them that data’.
An economic analysis of alleged anticompetitive conduct is a fact-specific inquiry. In assessing data-focused theories of harm, it is critical to understand first how the firms at issue compete, including how they use (or potentially use) the relevant data. Below, we discuss how data-focused theories of harm involving two-sided platforms were assessed in the government’s case against UnitedHealth Group (United).
In this case, the DOJ claimed that United’s acquisition of healthcare billing and payment processing service provider Change Healthcare (Change) would allow United ‘to use its rivals’ [electronic claims and payment data] to gain an unfair advantage and harm competition in health insurance markets’. In February 2022, the DOJ filed a lawsuit to block the merger under Section 7 of the Clayton Act, alleging two vertical theories of harm:
- the combined firm would enable United to use its rival’s claims data tracked by Change’s digital platforms ‘to extract intelligence about its health insurance rivals’; and
- the combined firm would ‘disadvantage its health insurance rivals by raising their costs and reducing or withholding quality improvements and innovations from rivals that rely on Change’s technologies’.
The key disagreements between the government and the merging parties centred on (1) the incremental value of competitive intelligence that Change’s claims data provided United, and (2) whether United had the economic incentive to withhold innovative technologies from rival insurers or otherwise degrade the quality of these products.
To support the first theory of harm, the government relied on an economic analysis of healthcare insurance claims data. This analysis showed that, through the acquisition, United would gain access to data on competitor claims that make up approximately 40 per cent of all commercial claims, potentially giving United access to detailed information about a large number of rival insurers’ customers. In response, the defendants argued that Change’s electronic claims data was not unique, and that United could access key elements of the claims data at issue from publicly available sources.
To support the second theory of harm, the government argued that the combined firm would be incentivised to limit sales of Change’s claims-transmission technology services to rival insurers, putting forward an economic analysis of the combined firm’s potential profits after the merger to show that it would purportedly need to gain a de minimis share of the insurance market to offset any profits it might lose from restricting sales of these services to rivals. In response, the defendants’ economic expert opined that no rival insurers relied exclusively on Change, such that these insurers could (or did) turn to competitors for similar claims-transmission services. Their expert further concluded that ‘selling products that include innovations would be more profitable for [United] than withholding them’.
Ultimately, the court ruled in favour of the merging parties on these issues and allowed the merger to proceed. Specifically, the court found that:
- the claims data at issue can have competitive value, but there was sufficient overlap between the types of data to which United already had access prior to the merger and the data to which it would have access through the acquisition of Change; and
- the evidence presented by the merging parties demonstrated that United had strong incentives to maintain a ‘multi-payer business strategy’ rather than withhold innovative technologies from rival insurers.
Based on the defendants’ analysis, the court concluded that the competitive intelligence gained by United from access to additional data post-merger would be similar to the intelligence it would have absent the proposed acquisition. The court concluded that the testimony from United executives on the strategic direction of the company was ‘far more probative of post-merger behavior than [the plaintiff’s expert’s] independent weighing of costs and benefits’.
As this decision shows, to directly test the merits of an allegation and be probative to the court, an economic analysis must be able to reliably isolate the effects of purported anticompetitive conduct. In data-focused markets, it is important for this type of economic analysis to account for how the firms actually use (or would use) the data at issue. In this case, the court found that United had access to similar data such that access to additional data from Change would not result in United gaining materially different potential business intelligence post-merger. An economic analysis should account for these business realities so as to reliably model the outcome from alleged conduct.
Implications of emerging technologies for antitrust enforcement
Artificial intelligence (AI) is a diverse domain that incorporates various technologies, such as machine learning and natural language processing. These can be used in various capacities, such as chatbots (which can process and generate humanlike text in real time) and healthcare tools (which can be used to assist with diagnosing and monitoring patients). The increased use of AI and other data-driven technologies raises potential competitive concerns. Below, we consider the concerns surrounding potential collusion facilitated by AI models, specifically with respect to pricing decisions.
As these technologies have advanced, firms have adopted AI-powered pricing algorithms to recommend the price that would optimise profits. These algorithms are programmed to derive this price by analysing data such as production cost, overhead charges, consumer demand, inventories and, most critically in this context, competitor pricing.
Many economists and lawyers have discussed how the use of AI pricing algorithms may facilitate collusion among competitors, even if there is no explicit agreement between the companies; for example, one publication noted that advancements in data processing power that enables real-time observation of the market and the use of AI to engage in autonomous decision-making can ‘amplify tacit collusion to a new level of stability and scope’.
This type of tacit collusion can be a result of firms that ‘unilaterally create[] an algorithm’ while knowing that ‘the industry-wide use of pricing algorithms will facilitate tacit collusion’ or when firms unintentionally align prices with those of competitors by using similar algorithms to monitor prices. Specifically:
- Firms may try to use real-time data to ‘anticipate and react to competitive threats well before any pricing change’, thus gaining a competitive advantage that enables them to react more quickly and acquire more data. This can result in the industry being dominated by a few ‘self-learning algorithms’ that tacitly collude.
- As AI models are programmed to optimise profits, they may use information about customers’ purchases of competitors’ products to set their own prices, even if the firms themselves did not share pricing information.
According to academic research, the use of such algorithms may create more situations where an anticompetitive outcome is a ‘side effect of the rise of the machines and their quest to optimize and serve’.
The updated Merger Guidelines outline several factors that the DOJ and FTC consider when assessing the extent to which competition may be harmed by tacit collusion through algorithms, including where (1) pricing algorithms that ‘track or predict competitor prices or actions’ can increase risk of collusion as companies’ pricing and strategies are easily observable across the market and (2) faster pricing algorithms can result in more predictable strategic responses from rivals. It is challenging for regulators to identify collusive conduct, however, as unless there is strong evidence that there is an agreement, it is unclear whether the algorithmic price-setting is distinct from conscious parallelism in an oligopolistic market (which is legal).
Ultimately, it is important for firms to actively understand the AI models they use. Empirical questions, such as whether there are barriers to access relevant data in certain marketplaces and whether the use of AI models by competitors facilitates collusive behaviour, can only be answered by studying the specific AI tools and relevant data.
Economic analysis can be used to provide an understanding of the algorithms deployed by companies, including the data inputs that feed into the pricing algorithms and how the algorithm leverages various data inputs to determine a price recommendation. Existing economic tools that have been widely used in traditional competition matters are still applicable in assessing the competitive effects of pricing decisions influenced by AI; for example, economic modelling can be used to compare the actual prices determined by companies using pricing algorithms to those in a ‘but-for’ world where no pricing algorithms are deployed. Another example is using natural experiments to examine whether sales data shows whether competitors in a market are systematically raising the price offered to a certain group of customers using similar pricing models.
Acknowledgements
The authors wish to thank Dr Ashley Zhou, formerly managing principal at Edgeworth Economics, for her contributions to previous editions of this chapter. The authors are also grateful for the research assistance conducted by Anav Singh and the collegial support from Stephanie Cheng and Michael Kheyfets.
