Does it Matter if Competition is “Fair” or “on the Merits”? An Application to Platform Self-Preferencing

Abstract

Platform self-preferencing is often attacked as being “unfair.” Proponents of the consumer welfare standard complain that a fairness standard is too vague and too untethered from competitive effects to be a useful guide for antitrust enforcement. However, the consumer welfare standard relies on assessment of whether competition is “on the merits,” and the criteria for “merits” substantially overlap with those for “fairness.” Both standards generally condemn: (a) deception; (b) raising rivals’ costs; and (c) gaining too much competitive advantage from past success, especially across markets. This article examines whether these criteria help identify when the competitive effects of self-preferencing are positive or negative. The discussion is framed in terms of a platform that facilitates the interaction of buyers and sellers, and chooses whether to preference certain sellers. As the literature has shown, whether it owns a seller or not, a platform may use preferencing to promote seller competition or to promote seller market power, depending on the circumstances. Given the varying effects of self-preferencing, a blanket prohibition is unwarranted. Unfortunately, the common criteria for fairness and merit do not reliably identify the direction of self-preferencing’s competitive effects. For example, deception by a lagging firm could strengthen competition in the sense of lowering equilibrium purchase prices and raising consumer welfare. And gaining “too much” competitive advantage from past successes could facilitate entry into new markets, thus increasing competition in those markets. Instead of applying criteria for fairness and merit, a case-by-case, fact-intensive analysis of actual competitive effects is needed.

“When I use a word,” Humpty Dumpty said in rather a scornful tone, “it means just what I choose it to mean—neither more nor less.”¹

1 Introduction

Many people are concerned that major online platforms serve as important means by which third-party sellers reach their customers while the platform owners also compete with these third-party sellers in various product markets. For example, Amazon provides an ecommerce platform on which third-parties sell products – sometimes in competition with Amazon’s own products.

The specific concern is that such a platform owner has a “conflict of interest” and will engage in “self-preferencing” by operating its platform in ways that favor its own products in competition with those of third-party platform users. One study, for example, found that Amazon favors its own products, such as Amazon Basics, in its product-search results.² Similarly, Apple and Google have been accused of preferencing their own products on Apple’s App Store and on Google’s internet search platform, respectively.

In his 2023 State of the Union address, President Biden called on Congress to “Pass bipartisan legislation to strengthen antitrust enforcement and prevent big online platforms from giving their own products an unfair advantage.”³ One example of this type of legislation, the proposed American Innovation and Choice Online Act, would make it unlawful for a covered platform to⁴:

• (1) Preference the products, services, or lines of business of the covered platform operator over those of another business user on the covered platform in a manner that would materially harm competition;

• (2) Limit the ability of the products, services, or lines of business of another business user to compete on the covered platform relative to the products, services, or lines of business of the covered platform operator in a manner that would materially harm competition.

These provisions outlaw self-preferencing practices that “materially harm competition.”⁵ Notably, however, the bill defines neither “competition” nor what it means to “materially harm” it.⁶ Moreover, these proposals to restrict self-preferencing are being made at a time when American antitrust policymakers, enforcers, and academic commenters are engaged in a broad debate with regard to what constitutes competition and how one can measure whether competition is stronger or weaker as the result of some change in market structure or market-participant conduct.⁷

Commencing in the late 1970s and for half a century afterwards, a broad, if somewhat murky, consensus emerged around the consumer welfare standard. Very loosely, under the consumer welfare standard, conduct harms competition if the conduct reduces consumer surplus.⁸ In recent years, this standard has come under attack on several fronts, one being the claim that fairness is fundamental to the notion of competition as used in the antitrust statutes but that the consumer welfare standard instead focuses solely on “efficiency.”⁹ In the quotation above, for example, President Biden focused on “unfair” competitive advantage, and Sect. 3(a)(9) of the American Innovation and Choice Online Act would make it unlawful for a covered platform to,¹⁰

in connection with any covered platform user interface, including search or ranking functionality offered by the covered platform, treat the products, services, or lines of business of the covered platform operator more favorably relative to those of another business user and in a manner that is inconsistent with the neutral, fair, and non-discriminatory treatment of all business users[.]

Proponents of a fairness standard judge whether conduct harms competition not by whether it reduces consumer surplus but instead by whether the conduct is unfair.

Many supporters of the consumer welfare standard find the appeal to fairness deeply problematic because proponents of a fairness standard have not put forth a coherent or comprehensive basis for assessing whether market conduct is fair. Proponents of the consumer welfare standard argue that its use of consumer welfare as a yardstick provides a coherent and (at least conceptually) well-defined means of determining whether conduct harms competition.

This argument misses an important fact: Although consumer welfare plays a central role in the eponymous standard, the standard as applied by U.S. courts is not one based solely on the measure of consumer welfare. Instead, there is a separate requirement that the conduct in question “harms the competitive process” or is “not competition on the merits.”

Consequently, the fairness and consumer welfare standards have more in common than their respective champions acknowledge. Leading conceptions of fairness seek to identify when conduct constitutes “fair” competition, while the consumer welfare standard seeks to identify when conduct constitutes competition on the “merits.” As I discuss in Sects. 2 and 3 below, criteria by which to determine what is fair have a high degree of overlap with criteria by which to determine what is on the merits.

Although this overlap might appear to be good news that suggests the possibility of finding common ground, it is bad news in that what the approaches have in common is vagueness and ambiguity. What the OECD said about competition on the merits can also be said of fairness: “it has served too often as a shortcut that glosses over the difficult work of defining clear principles and standards that embody sound competition policy.”¹¹

I illustrate the common shortcomings of the two approaches by applying them to the assessment of preferencing among third parties by a platform owner that is not also a platform user (in Sect. 4) and self-preferencing (in Sect. 5).

2 When is Competition “Unfair”?

If one is going to use fairness to define competition or harm to it, then one needs either: (a) a set of principles by which to determine whether specific conduct is fair; or (b) a comprehensive list that enumerates what conduct is fair and what is not. The U.S. Congress rejected approach (b) on the grounds that business practices are too numerous and too subject to change for it to be possible to construct a comprehensive and enduring list;¹² instead, it falls on the Federal Trade Commission (FTC) and the courts to determine what practices constitute unfair methods of competition.¹³ Unfortunately, neither Congress, the FTC, the courts, nor various proponents of a fairness standard for antitrust enforcement have provided a clear statement of principles. For example, Sect. 5 of the FTC Act and the FTC’s recent statement on unfair competition refer to “unfair competition” but do little to define fairness with any specificity.¹⁴

One important distinction among principles of fairness is whether the assessment of an outcome is made by reference to the characteristics of the outcome itself (e.g., whether there is income inequality among households or whether small businesses are profitable) or by examining whether the outcome is the result of a fair process or procedure (e.g., asking whether there was equality of opportunity regardless of the actual outcome).¹⁵

There are several problems that are associated with defining harm to competition by appealing to intuitive principles of what constitutes a fair outcome: One is that policies that are intended to promote outcomes that satisfy some of these principles can be in direct conflict with any reasonable notions of competition and rivalry. For example, policies that seek to promote the welfare of smaller suppliers can create incentives for more-successful suppliers to curtail their efforts to attract buyers so as not to be accused of leading to an unfair outcome. But most people would recognize increased efforts to attract buyers as the essence of competition.

Another problem is that these principles may conflict with one another – which, at a minimum, raises the question of how they should be weighed against one another.¹⁶ Trading the realization of one measure of fairness off against another is made difficult by the fact that these principles generally do not offer precise metrics. In fact, even the application of a single principle can raise questions of tradeoffs without providing precise answers. For example, in evaluating an outcome, how much weight should be given to the welfare of market participants in the poorest decile versus those in the second-poorest decile?

Faced with a set of ill-defined and possibly conflicting principles of fairness, business decision makers may be unable to determine what they can and cannot do, which can be expected both to reduce the effectiveness of antitrust policy and to lead to adverse unintended consequences.¹⁷

A policy that defines harm to competition through the application of outcome-based notions of fairness can also be hard to administer. Under a policy that defines competition as fair only if it promotes the welfare of low-income households, for example, antitrust enforcers would have to determine what constitutes a fair societal distribution of income and then measure the income of the employees, business owners, and customers to evaluate the effects of any given enforcement decision. The evaluation could also turn on factors that had little or nothing to do with the industry in question (e.g., local real-estate prices could affect employees’ real incomes) and would raise issues with respect to the interaction and division of labor among different government policies (e.g., income taxation).

In the remainder of this paper, I will focus on the second approach, which focuses on characteristics of market conduct rather than on the outcomes to which that conduct gives rise: An outcome that is the result of a fair process is itself fair.¹⁸ A then-Acting Assistant Attorney General of the U.S. Antitrust Division stated that “[a]nimating the beliefs of ordinary Americans who demand vigorous antitrust enforcement are the value of fairness and the belief that properly functioning competitive markets are themselves fair.”¹⁹

Of course, stating that an outcome is fair if it is generated by a properly functioning competitive market does not provide a workable definition of fair competition, and raises at least as many questions as it answers: One is how much rivalry is enough for a market to be “competitive”? Economists have developed the notion of workably or effectively competitive markets, which provides at least a partial answer.²⁰ Here, I will focus on a second question: Do some forms of rivalry constitute “unfair competition” and thus fail to generate fair outcomes regardless of the apparent degree of rivalrous conduct?

Many people have concluded that the answer is “yes,” and several principles regarding what constitutes a fair process have been suggested.²¹ One way of consolidating several of them is the following²²:

A party should not gain competitive advantage through deception. This is the oldest and possibly least contentious criterion for fairness. The concept of unfair competition in English common law arose from the idea that infringing a rival’s trade-mark by passing off one’s own goods as being those of the rival was unfair.²³ The idea of unfair competition was then expanded to include competing based on deception more broadly.²⁴ More recently, Hughes (1994, p. 299) has argued that one “component of fairness entails the protection of legitimate expectations”, which is “consistent with the precept that rules should not be changed in the middle of the game.” This notion of fairness could be conceived as applying the deception criterion to the dynamic treatment of business partners and rival suppliers, as well as consumers.²⁵ It provides one basis for criticizing what has been referred to as an “open early, closed late” strategy, whereby a firm unilaterally restricts the interoperability of its platform with rival platforms after it has attracted complementary suppliers that had been expecting interoperability.²⁶

It is unfair to compete by taking actions that raise rivals' costs or diminish rivals’ quality. An early application of this criterion was to competition to attract waterfowl to be hunted. In a mid-nineteenth-century case in England, “[t]he court said that it was lawful for the plaintiff to attract his feathered customers away from the Duke (of Rutland) with birdseed, but not for the defendant to attract them back again with rockets and combustibles.”²⁷

A party should not be able to take too much advantage of its past investments and resulting success.²⁸ Proponents of a fairness standard do not characterize their concerns in this way but it encapsulates what they do state. There are both within- and cross-market versions of this criterion, and the sources of advantage that are derived from past success can include money, a customer base, or a large range of complementary products (e.g., apps for a smartphone). Within-market examples include the claim that it is unfair for a large firm to bargain for quantity discounts that allow it profitably to charge lower prices than can its smaller rivals.²⁹ With respect to cross-market concerns, some people would deem it unfair for a large firm to engage in vertical integration when doing so creates a competitive advantage. And there is a long history of considering tied-sales to be unfair.³⁰ The ability to finance predatory pricing might also be condemned by this principle – whether within a market or across markets.

Competition is unfair if a supplier relies on practices that are immoral, unethical, or contrary to public policy (even if not illegal).³¹ For example, a candy company was found to have engaged in unfair competition when it sold candy with prices and prizes that were revealed only after purchase, thus encouraging “gambling among children.”³² Presumably, some people consider the use of sexual images to sell products to be immoral and thus an unfair method of competition.

Market outcomes should not entail the imposition of coercive or oppressive terms on trading partners. The legislative history of the Federal Trade Commission Act contains several expressions of concern with “oppressive competition,” while offering little clarity as to what makes conduct oppressive.^{33, 34}

Competition is unfair unless it constitutes “competition on the merits.”³⁵ The definition of “competition on the merits” will be discussed below.

3 The Consumer Welfare Standard and competition on the “Merits”

Under a pure consumer-welfare standard, a court assesses the overall effects of the challenged conduct on consumer welfare (as measured by consumer surplus) and deems the conduct to be illegal if it reduces consumer welfare.³⁶

There are several problems with the use of a pure consumer-welfare standard. Perhaps the most significant one is that such a standard would impose extremely strong requirements on firm behavior. For example, the standard would make it illegal for firms to earn expected economic profits – consumers would be better off if the firm earned the minimum amount necessary to remain in business while pursuing consumers’ preferred innovation and investment strategies. The standard would also ban sellers from investing to improve their bargaining positions, such as engaging in forward integration to offset the market power of a monopsony buyer.

Another objection to using changes in consumer welfare to define harm to competition is that changes in consumer welfare need not correspond to any intuitively reasonable notion of a change in competition. First, changes in consumer welfare can arise under monopoly. For example, innovation by a monopolist may increase or decrease consumer welfare due to changes in the extraction of consumer surplus.³⁷ There is no coherent sense in which the degree of competition is changing as consumer welfare changes in a monopolized market. And a drastic innovation might lead to a firm’s monopolizing a previously rivalrous market as rivals are driven to exit. Even if consumers are better off due to the innovation, it seems paradoxical to say that competition has increased.

Yet another example is provided by a seller that holds an auction and that first engages in advertising to attract additional bidders. If the resulting increase in rivalry among bidders increased the winning bid without changing the identity of the winning bidder, a pure consumer-surplus standard would condemn the seller’s conduct as harming competition. To say that such advertising harms competition because it increases rivalry and leads to a higher winning bid is to give “competition” a meaning that is the opposite of common usage.³⁸ Moreover, such advertising sometimes would lead to greater buyer welfare by promoting a better match. The firm that holds the auction thus might be hard-pressed to make a credible prediction of the expected effects of its advertising on consumer welfare.

More generally, a pure consumer-surplus standard would often be very hard to apply. For example, the ban on expected economic profits would be tantamount to an extreme form of regulation, and antitrust enforcers would face many of the information asymmetries that can make effective regulation difficult, especially in innovative industries.

The difficulty of applying the standard can also undermine the ability of policy to deter anticompetitive behavior. Commenting on the use of this approach to assess allegations of exclusionary conduct, Melamed (2005, p. 1254) argues that prospective defendants generally do not have sufficient information to predict the net effects of their conduct on consumer welfare and that, consequently,this test for anticompetitive conduct “would likely either be ignored, impose excessive transaction costs (a kind of tax on entrepreneurship), or result in excessive caution.”³⁹

Whatever the merits of a pure consumer-surplus approach, the consumer welfare standard as applied by U.S. courts is not such an approach. For example, under the consumer welfare standard as applied by U.S. courts, charging monopoly prices is not illegal if a firm has lawfully obtained monopoly power.⁴⁰ This is true even though – at least in the short run – consumer welfare would be higher if such a firm charged lower-but-still-profitable prices.

Instead of a pure consumer-surplus standard, U.S. courts apply a standard under which there is also a separate determination of whether there is harm to the competitive process: to be illegal, conduct that harms consumer welfare must be conduct that does not constitute competition on the merits.⁴¹ Even advocates of “the true consumer welfare standard” often have in mind some notion of effects on the competitive process in addition to changes in welfare levels.⁴²

Although merger policy, for which advocates of the consumer welfare standard focus solely on how a merger affects consumer surplus, might appear to be a counterexample to the claim that the consumer welfare standard has two components, it is not; there is not a separate investigation into whether a merger harms the competitive process because it is taken as a given that any merger that decreases consumer surplus also harms the competitive process by eliminating competition between the merging parties.

Many commenters now interpret the “consumer” welfare standard as encompassing harm to the welfare of trading-partners (whether buyers or sellers) due conduct that does not constitute competition on the merits. The critique of the consumer welfare standard below applies to this interpretation as well.

Whether in conjunction with a consumer-surplus test or as a standalone concept, one must define what constitutes rivalrous conduct that is not competition on the merits. Examples of proposed criteria include:

A party should not gain competitive advantage through deception. The abuse of government process – e.g., obtaining a patent through deception – has been identified as conduct that is not competition on the merits.⁴³ And as Carrier and Tushnet (2021, pp. 1852–1853, emphasis in original) have stated, “by definition, false advertising is not competition ‘on the merits’ because it is deceptive about the merits.”^{44, 45}

A party should not take actions that raise rivals’ costs or diminish rivals’ quality. Stated affirmatively, conduct is competition on the merits if it makes a party a more attractive trading partner relative to some absolute standard without making rival parties less attractive trading partners relative to some absolute standard.⁴⁶

A party should not be able to take too much advantage of its past investments and resulting success. For example, the U.S. Supreme Court has long held that a product should stand on its own and that tying prevents a product from being chosen on its merits – the price and quality of the product – rather than because its sale was tied to that of another product.,^{47, 48} Predatory pricing can also be condemned under this principle, where past success allows a seller to fund the period of predatory losses that are incurred from pricing below cost.⁴⁹

4 Application to Preferencing by an Unintegrated Platform

The concepts of both fair competition and competition on the merits share several criteria in that both condemn: deception; raising rivals’ costs; and taking inappropriate advantage of past successes. These concepts also share several shortcomings as enforcement guidelines, as I will now illustrate in the context of a platform that can choose to preference certain users.

By definition, a platform facilitates the interactions of users on its different sides. Often, users on one side are sellers and users on the other are buyers, and it useful to adopt the buyer–seller terminology to discuss preferencing.⁵⁰ Platforms generally offer buyers some means of discovering sellers. For instance, Amazon.com, the Apple App Store, and Google Play Store offer search functions. Typically, a platform can choose to treat a seller better or worse in the discovery process. For example, a product might appear higher or lower in an ordered list of search results in settings where consumers are known to favor products with higher placement. Or a platform might draw attention to certain sellers by featuring them in highlighted sections of its results page, such as the Amazon “Buy Box.”

The discovery process can be broader than traditional search. For example, Apple and Samsung smartphones currently are shipped with Google as the default search provider at key access points, which makes it easier for smartphone users to “discover” Google Search.

Preferencing can also entail differential access to platform features and functions: For example, a mobile operating system might expose certain application programming interfaces (APIs) to some applications developers but not others. An important difference between preferred treatment in a discovery process and preferred access to features and functions is that typically the former is necessarily rivalrous in that the lead search position, say, can be awarded to only one seller. By contrast, access to a feature or function may be non-rivalrous in that it is feasible to grant multiple sellers access to it.

Is a platform’s favoring its own products over those of third parties fair and/or competition on the merits, and does the answer to this question inform enforcers about competitive effects? Before examining these issues, I consider preferencing by a platform owner that is not also a seller on its platform. I do this both: (a) because prominent parties have condemned preferential treatment of third parties as well as self-preferencing;⁵¹ and (b) to provide context for assessing the effects of self-preferencing. As I will now discuss, although the effects of preferencing strike some people as self-evident, they are anything but.

At the outset, it should be noted that, unless the arguments are limited to certain market conditions, attacking the fairness or merit of purchasing preferential treatment runs the risk of being arbitrary and inconsistent. For instance, placement in a platform’s discovery process is just one of many inputs that sellers utilize to compete. If paid placement is unfair or is not competition on the merits, then why isn’t it also unfair or unmeritorious for other media to sell advertising and for product sellers to purchase it? For that matter, why isn’t it anticompetitive for a firm to purchase any inputs that make its output more attractive to buyers? As far as I am aware, no proponents of a fairness approach nor those of the consumer welfare standard have called for it to be illegal for restaurant chains to advertise or to use higher-quality ingredients, although such practices might make it more difficult for smaller, rival restaurants to compete.

One approach is to limit concerns with regard to preferencing to situations in which: (a) the platform has substantial market power in the supply of the input at issue; and (b) preferred access to the input provides substantial competitive benefits to the recipient. In the remainder of this article, I will assume that the platform is a very important form of distribution for the sellers, and that preferential treatment significantly boosts demand for the recipient’s product. That is, I will restrict attention to situations in which preferencing has substantial effects on competition and ask whether various fairness and merits criteria help one determine whether the effects are positive or negative.

4.1 Competitive Effects

As I will now discuss, when conditions (a) and (b) are satisfied, preferencing can harm competition but also can lead to greater competition by any reasonable interpretation of competition as rivalry:⁵² Under some conditions, blocking preferencing makes entry by new sellers more difficult and softens price competition among incumbent sellers on a platform.⁵³

The potential effects of preferencing on competition can be seen most starkly in the case of entry: Suppose that there is a physical product that is sold through a monopoly platform that provides buyers a ranked list of sellers of the product. Assume further that there is a single incumbent seller and a single potential entrant, where the latter offers a differentiated product. The platform owner does not own either the incumbent seller or the potential entrant.

One of the reasons that the effects of preferencing are complex is that preferencing can affect consumer behavior in several ways, such as by reducing consumers’ search costs or by raising their expectations of the quality of the product that receives preferential treatment. In the present example, assume that past sales and preferred placement are substitute forms of consumer exposure and that increased exposure has diminishing marginal returns with respect to current sales. Specifically, suppose that, if the entrant does not receive preferred placement, then a consumer will ignore the existence of its product, and the entrant will make no sales. By contrast, if the entrant’s product does receive preferred placement, then a consumer will consider both firms’ products: the entrant’s because of preferred placement; and the incumbents’ because of exposures through past sales.

Under these assumptions, if the incumbent receives preferred placement, then: (a) the entrant will make no sales; and (b) the incumbent will price at the monopoly level. By contrast, if the entrant receives preferred placement, then the two firms will compete, resulting in differentiated-Bertrand prices.⁵⁴ To analyze the competitive effects of allowing platforms to engage in preferencing, it is necessary both to: (a) solve for equilibrium when preferential treatment is permitted; and (b) identify the but-for world in which preferential treatment is not allowed.

When preferencing is allowed, the platform owner chooses its preferencing policy to maximize the platform’s profits, which generally depend on how preferencing affects both buyers and sellers because buyer and seller welfare affects the fees that the platform can charge those parties. For simplicity, ignore the effects on other platform fees and suppose that the platform runs a second-price auction for placement in the lead position in its search results.

As is well known from bidding models of patent races, there is a bias toward the incumbent producer’s outbidding the prospective entrant because of the benefits of exclusion.⁵⁵ Indeed, in the present example, the only benefits to the incumbent from purchasing preferred placement is that it prevents the entrant from gaining exposure. There are, however, scenarios in which the direct benefits to an entrant can be sufficient to overcome any exclusionary benefits that an incumbent might realize from outbidding the entrant. For example, the entrant may gain more than the incumbent would lose because the former offers a differentiated or lower-cost product that expands the market.⁵⁶ In general, either the incumbent or the entrant might win the bidding.

Critics of preferencing focus on cases in which the incumbent wins and blocks entry. Suppose that, because of this concern, preferencing is forbidden and the platform is required to rank results in a “neutral” or “unbiased” order. One interpretation of such a requirement would be that the ranking algorithm must display first the product that a consumer is most likely to purchase if it is placed first. In this case, a neutral algorithm might well place an incumbent’s product first. Knowing this, the potential entrant would anticipate making no sales and would not enter. Even equally randomized placement might not be enough to make entry financially attractive. In short, there are situations in which, when preferencing is allowed, the entrant will purchase it and successfully compete in the product market but, when preferencing is prohibited, the market will be monopolized by the incumbent.

Next, consider the effects of paid placement on competition among incumbents. Critics of preferencing focus on situations in which preferred placement strengthens the position of a dominant seller, which can lead to higher prices in the short run and have adverse long-run effects because the weaker seller is denied scale or cumulative sales that would otherwise improve its product.⁵⁷ However, there are also conditions under which “neutrality” can serve to soften competition by creating a set of “captive” customers for each firm, which thus gives all firms a reason to set relatively high prices.

The following highly stylized example illustrates the general mechanism:⁵⁸ Suppose that there are two incumbent sellers – which are denoted by $i = 1,2$ – that each have constant marginal costs of $c$ per unit. Suppose further that, if seller $i$’s product is in the preferred position, then a consumer chooses product $i$ if and only if $b - p_{i} \ge {\text{max}}\left\{ {b - t - p_{ - i} ,0} \right\}$ and product $- i$ if and only if $b - t - p_{ - i} > {\text{max}}\left\{ {b - p_{i} ,0} \right\}$, where: $b$ is the benefit of the product; $t$ is the (possibly psychic) transaction cost of considering the non-preferenced product; and $p_{i}$ is the price of product $i$. If seller $i$ receives preferred placement for all consumers, then the equilibrium prices are $p_{i} = c + t$ and $p_{ - i} = c$, and all consumers purchase product $i$.

Now, suppose that the platform is forced to have neutral placement, which consists of giving each seller preferred placement half of the time, and that a seller cannot set its price contingent on its search-results position. Then, as long as $t \ge \frac{1}{2}\left( {b - c} \right)$, the equilibrium prices are $p_{1} = b = p_{2}$.

Comparison of the formulas for the equilibrium prices reveals that, when $b > c + t > b - t$, the equal-treatment regime leads to higher equilibrium prices than does the preferencing regime. Intuitively, when each seller has a set of consumers for which that seller is the default, each seller finds it more profitable to charge a high price to take advantage of those consumers’ greater willingness to pay for its product than to charge a low price to compete for all customers. By contrast, under preferencing, the disadvantaged firm has no such group of consumers, and the firm attempts to overcome its disadvantage by reducing its price to marginal cost.⁵⁹

Observe that, in this example, the platform may find equal treatment more profitable than preferencing: When the platform uses equal treatment to soften downstream competition, the platform may be able to appropriate the resulting downstream profits through fees that are levied on the sellers.⁶⁰

In general, whether a platform will adopt a preferencing policy that promotes or weakens seller competition depends on the many factors that can affect the platform’s ability to extract surplus from buyers and sellers.⁶¹ For example, a platform might want to promote perfect competition among sellers when it is able fully to extract the resulting buyer surplus though platform fees that are charged to consumers. In other situations, such as the example above, where the platform can better extract surplus from sellers, it might want to promote monopolistic seller behavior.⁶²

4.2 Application of the Common Criteria

I next examine whether the three principal criteria for fair competition and competition on the merits are useful in identifying whether preferencing harms entry and/or softens competition. Of course, under the consumer welfare standard, whether conduct constitutes competition on the merits is not the sole basis for determining whether it is illegal: The conduct must also harm consumers.⁶³ And even under the fairness approach, there can be a separate condition that includes harm to consumers.⁶⁴ Reliance on changes in consumer welfare has the problems that were identified in the discussion of the pure consumer-surplus standard in Sect. 3 above. The analysis below examines whether the three criteria can serve to avoid or limit these problems.

4.2.1 Deception

As was discussed above, the notions of both fairness and competition on the merits indicate that it is problematic when consumers make incorrect inferences from a platform’s decision to grant a product preferential treatment and those inferences lead to material changes in the market outcome. However, at least as measured by equilibrium margins and entry, such deception can lead to more intense rivalry.

Consider a platform that sells preferred placement in its search results: If the platform does not disclose its preferencing policy, a consumer may believe that a product that receives preferred placement is, in the platform’s view, the best match from the consumer’s perspective, and the consumer may believe that this is an informative signal. As the following example demonstrates, this deception can benefit both consumers and weaker sellers.

Consider a product that is sold by five Cournot oligopolists: A representative consumer derives gross dollar benefits $B = \mathop \sum \limits_{j = 1}^{5} \alpha_{j} x_{j} - \frac{1}{2}\left( {\mathop \sum \limits_{j = 1}^{5} x_{j} } \right)^{2} ,$ where $x_{i}$ is seller $i$’s output level. Assume that $\alpha_{i} = \alpha$ for $i = 1,2 \ldots ,4$, and $\alpha_{5} = \frac{6}{7}\alpha$, where is $\alpha$ a positive constant. Each seller incurs a fixed cost of $F = \left( {\frac{\alpha }{21}} \right)^{2}$ if it produces output. The common, constant marginal cost of output is subsumed in $\alpha_{i}$.

In this example, if seller 5 does not appear in the first search position, then consumers correctly perceive the values of $\alpha_{i}$, and the resulting inverse demand curves are $p_{i} = \alpha_{i} - \mathop \sum \limits_{j = 1}^{5} x_{j}$. By contrast, if seller 5 does appear in the first search position, then buyers make consumption decisions as if $\alpha_{i} = \alpha$ for all five sellers, but realized consumption benefits are as given by the formula for $B$ above: Placing seller 5 in the first position deceives buyers into believing that seller 5’s product generates greater consumption benefits than it actually does.

Standard calculations show that, absent deception, only sellers 1 through 4 are active in equilibrium; given consumers’ lower willingness to pay for its product, seller 5 would be unable to cover its fixed costs and chooses not to produce output.⁶⁵ By contrast, the equilibrium under deception is symmetric, and all five sellers are active.⁶⁶

If we compare the sellers’ profits under the two equilibria, seller 5 is willing to pay more to purchase the preferred position than is any of sellers 1 through 4.⁶⁷ As long as the platform is unable to charge seller 5 its full willingness to pay for the preferred position, seller 5 gains from deception. Notably, even though they are the ones who are deceived, consumers also collectively gain from deception.⁶⁸

In general, there are two broad types of consumer-welfare effects that arise from deception: (a) changes in the intensity of competition; and (b) consumer-product matching effects. As the example above demonstrates, deception that favors a firm with a less-attractive product can intensify price competition while harming matching.⁶⁹ Thus, preferred placement can deceive buyers in a way that “levels the playing field” for a smaller or weaker seller, which can benefit that seller and, depending on parameter values, lead to higher or lower consumer welfare.⁷⁰ In summary, the deception criterion by itself does little or nothing to identify the effects of preferencing on entry, rivalry, small-seller welfare, or consumer welfare. Instead, one must determine these effects by examining the specific facts.⁷¹

4.2.2 Raising Rival’s Costs

Both approaches to assessing competition might condemn preferencing on the grounds that it raises the costs that are incurred by rivals of a seller that receives preferential treatment. Indeed, the U.S. Department of Justice argued that Google’s purchase of default positions on iPhones substantially raised rivals’ costs of obtaining data and search traffic, which led these rivals to offer lower-quality search results than they otherwise would have.⁷²

A major weakness of this criterion is that, in the presence of various benefits of scale, any action that a supplier takes to make its product more attractive will tend to divert sales from rivals’ products, which will make those products less attractive (in the presence of network or data effects) or more costly (in the presence of declining marginal costs of production). Hence, on its own, the raising-rivals’-costs criterion does little to help determine whether conduct softens competition or harms entry.⁷³

A policy under which the courts concluded that any attempt by a firm to increase its sales is unfair or unmeritorious under these conditions, which are common in many industries, could stifle competition rather than promote it. In the short run, current market leaders might be incentivized to compete less vigorously to avoid triggering liability. And in the long run, firms might be less willing to invest in becoming market leaders in the first place. There is a need for a principle that limits and refines the use of this criterion to condemn preferencing (or other conduct).⁷⁴

One approach is to apply the logic of the “no economic sense test” to the firm purchasing preferred placement. Under this test, conduct is not exclusionary unless it makes no business or economic sense but for the likelihood of harming competition.⁷⁵ However, it will often be the case that conduct will both strengthen a seller’s competitive position and weaken the positions of its rivals, which makes the necessary calculations difficult.⁷⁶

There is also a more troubling issue: Suppose that a firm undertakes an innovation that reduces its production costs or improves its product quality and finds that – having made the innovation – it is profitable to make such attractive offers to consumers that at least some other sellers are driven to exit the market. In addition, suppose that making these attractive offers maximizes the firm’s profits even after these rivals exit, but that investing in the innovation would not be profitable if, instead of exiting the market, the rival firms continued to sell output at prices that are above their marginal costs but are below their average costs. To apply the no economic sense test, one would need a means of determining whether the induced exit represented harm to competition or was the result of vigorous competition. In this sense, the sense test begs the fundamental question.

Another approach is to condition liability on whether the firm that receives preferred placement is currently a market leader. However, this approach could stifle competition, as I will discuss next.

4.2.3 A Firm Should Not Benefit Too Much from Past Success

Preferential treatment has been attacked on the grounds that firms that have been successful in the past may have greater abilities to purchase preferential treatment and thus “entrench” their positions. As was shown above, preferential treatment has ambiguous effects on the intensity of competition among incumbents and the ease of entry. Preferential treatment can harm competition in some circumstances, but it can promote entry when the entrant is the one that receives preferential treatment and can intensify competition if it allows a weaker incumbent at least partially to offset advantages that are enjoyed by rival sellers.

These possibilities suggest that the past-success criterion might be a useful means of distinguishing between the pro- and anticompetitive cases. Upon examination, however, it is far from evident that an enforcement policy under which preferential treatment is objectionable only when granted to already successful incumbents would promote competition.

One problem is that granting preferential treatment to a weaker incumbent can soften price competition for reasons that are similar to those that were discussed in Sect. 4.1 above: The formerly weak firm might have incentives to abandon its low-price strategy so as to exploit a set of captive or preferred customers that is created by the preferential treatment.

Another problem arises when the platform’s treatment of a seller influences consumer decision making and generates matching benefits: The dominant firm might be the one that generates better matches, so that granting preferential treatment to the weaker product might be a form of deception, while having no preferential treatment of any firm (in those cases where it is feasible) could deny consumers potentially useful information.

Limiting the availability of preferential treatment to weaker sellers is also problematical because it undermines investment incentives. For example, it might be more profitable to be classified as a weak incumbent – and thus be eligible for preferential treatment – than to invest in obtaining competitive advantages that then render the seller ineligible for preferential treatment. In short, the policy of favoring weaker sellers can be seen as rewarding sellers for offering unattractive products.

There is also a more-limited application of the past-advantage criterion: one that can be viewed as a cross-product application of the criterion. Specifically, testifying as a witness for the U.S. Department of Justice, the CEO of DuckDuckGo alleged that, had Google not demanded to be the exclusive default search engine for all users of the Safari web browser on iPhones, his company’s search engine might have been able successfully to compete to be the default search engine on the iPhone Safari web browser for the segment of consumers who are particularly concerned with privacy.⁷⁷ Under either a fairness or merits approach, one might object to such exclusivity on the grounds that it prevents a product from standing alone for a given customer segment and being fully considered on its own merits. This practice is similar to a tie-in sale, although here it is a tie-in purchase. In this regard, it is notable that, in general, the competitive effects of ties are complicated and can be positive or negative.⁷⁸

In short, the past-advantage criterion does little to distinguish cases in which preferential treatment harms competition from those in which it promotes competition.

5 Self-Preferencing

It is frequently alleged that a platform owner has an inherent conflict of interest when it also offers products that compete with those of third-party platform users, so that self-preferencing – here, I use the term to refer to a platform that treats its own first-party seller more favorably than some or all third-party sellers – is greeted with even more skepticism than is platform preferencing generally.⁷⁹ In this section, I address three questions: (1) What effects does integration have on a platform owner’s incentives to preferentially treat certain sellers, including its own, first-party seller? (2) Does self-preferencing strengthen or weaken competition? and (3) To what extent do the concepts of fairness and merit allow one to distinguish procompetitive from anticompetitive instances of self-preferencing?

5.1 The Effects of Integration on Preferencing Incentives

In comparing the preferencing incentives of integrated and non-integrated platform owners, an important initial observation is that a platform bears an opportunity cost of favoring itself. Suppose that a platform grants its own, first-party seller preferential treatment. When preferential treatment can be granted to only a limited number of sellers – e.g., only one product can be first in a set of search results – the platform will be able to sell preferential treatment to fewer third-party sellers than otherwise. Even when an unlimited number of sellers can receive preferential treatment, third parties may be less willing to pay for preferential treatment when the first-party seller receives it too.⁸⁰

Nevertheless, there are at least two important differences between an integrated firm’s preferencing incentives and those of an independent platform and downstream providers:

The integrated firm may more fully internalize the effects of preferencing on the first-party seller than on third-party sellers. In theory at least, an integrated platform completely internalizes the effects of its preferencing decisions on its first-party seller. By contrast, the platform may or may not be able to extract all of the benefits that a third-party seller enjoys from preferential treatment. For example, if there are multiple third-party sellers that value preferential treatment equally to one another, then the platform could run an auction and fully extract the value. In other cases, however, it would be infeasible to extract all of a third-party’s surplus that is associated with preferential treatment, and there could be various forms of the traditional monopoly distortion in the sale of that treatment: While trying to extract surplus from third parties, the platform might restrict the sale of preferential treatment to them, which means that, all else equal, the platform would have a tendency to favor its own product.⁸¹

For example, suppose that the third parties have heterogenous values of preferred placement and the platform runs a second-price auction. Although third parties would have incentives to bid their true values of preferred placement, the first-party seller could have an incentive to shade its bid upward so as to extract more surplus from the other firms when it was the second-highest bidder. Or, if the platform set a posted price for preferred placement, it would have an incentive to set the price higher than its own value of preferred placement, trading off the possibility of losing a sale to a third party against the gains from obtaining a higher price when such a sale did occur. Either practice would create a bias toward the platform’s downstream division’s winning preferred placement.

The first-party seller internalizes its effects on the platform and, thus, will, to some degree, indirectly internalize the effects of its conduct on third-party sellers and buyers. Internalizing the effects on third-party sellers could lead the first-party seller to behave more or less competitively than third-party sellers, depending on the circumstances.

The first-party seller will behave more competitively than would a similar third-party seller if doing so helps the platform extract rents from third-party providers, as happens when competition from the first-party seller drives down third-party sellers’ profits while strongly increasing buyers’ willingness to pay for the platform’s services.

However, the first-party seller will behave less competitively than a similar third-party seller if behaving competitively would reduce third-party seller’s willingness to pay the platform for its services by more than it would increase buyers’ willingness to pay for platform services. Consideration of deadweight loss suggests that more intense seller competition should increase the sum of buyer and seller benefits that are potentially available for the platform to extract; however, the platform may have different abilities to extract incremental surplus from the two sides, so that there is no guarantee that the platform acts to minimize deadweight loss.

As a result of these two differences in the degrees of internalization, an integrated platform owner can have incentives to choose its first-party seller over third-party sellers when it grants preferential treatment.

5.2 Competitive Effects of Self-Preferencing

The implications of these incentives to self-preference are complex for at least two reasons: First, as was discussed in Sect. 4.1 above, preferential treatment of a third-party seller can have positive or negative effects on competition. Second, although internalization has been called a conflict of interest, it can have beneficial effects too; as was just discussed, a first-party seller may behave more competitively than would a third-party seller. Hence, it should come as no surprise that self-preferencing can strengthen or weaken competition in comparison with a neutrality regime.

Zou and Zhou (2023) show that, in the short run, search neutrality can soften price competition between third-party sellers and an integrated firm’s first-party seller. Zou and Zhou (2023, pp. 25–26) also identify conditions under which search neutrality can lead the platform to preempt the entry of third-party sellers because additional sellers can be less valuable to the platform under search neutrality than under paid placement. This entry effect arises in their model because, under search neutrality, price competition is softened, which has the effect that consumers do not consider options other than the seller that receives prominent placement. By contrast, when the platform favors itself, the entrant will charge a lower price, which leads some consumers to search for the non-prominent product, which can lead to better matching of seller and buyers, thus increasing the volume of commerce on which the platform collects a commission.

In short, as with an unintegrated platform, there are conditions under which a prohibition of preferencing can soften price competition and make entry more difficult.

As was discussed in Sect. 4.1 above, an independent platform may not choose the optimal pattern of preferencing from the perspective of promoting competition. The difference between the incentives of an integrated platform and an independent platform can make this distortion better or worse. An integrated firm may be more likely to monopolize the downstream market for the following reason: The integrated firm fully appropriates the downstream profits that result from the market power that preferencing generates for its first-party seller, but an independent platform may be unable to appropriate the benefits of the increased market power enjoyed by a third-party seller that is given preferential treatment.

In the other direction, because an integrated platform/seller may be able to internalize certain benefits of increased downstream competition to a degree that independent firms cannot, downstream competition may be more attractive to the integrated firm. For example, an integrated platform owner may self-preference to increase competition by facilitating the entry of a first-party seller that creates more competitive pressure on rival sellers than would a third-party entrant.

A final difficulty in assessing competitive effects is that they may vary over time. For example, some people would argue that Amazon Basics are examples of procompetitive entry, while others would argue that the long-run effects will be to induce rivals to exit.

5.3 Application of the Common Principles

Self-preferencing, like preferencing generally, can strengthen or weaken competition. Do the criteria with regard to raising rivals’ costs and inappropriate benefits from past success help discern the sign of the competitive effects?

With respect to the raising-rivals’-costs criterion, it is useful to consider the non-rivalrous-preference and rivalrous-preference cases separately:

The non-rivalrous case is a form of refusal to deal, and a prohibition of preferencing in these cases raises the issue of investment incentives that is central to U.S. courts’ reluctance to impose a duty to deal: Forcing a platform to share with rival sellers the fruits of its costly investments will weaken the platform’s investment incentives.⁸²

The raising-rivals’-costs criterion raises particularly difficult issues with respect to intellectual property, which may have near-zero costs of dissemination. Does the low cost of dissemination mean that any refusal to license intellectual property to a competitor is a form of raising rivals’ costs that would fail the no economic sense test? The adverse investment incentive effects of such a policy would be profound and could change races to innovate into waiting games, with each firm hoping to free- or cheap-ride on its innovating rivals.⁸³ Alternatively, courts would have to determine appropriate licensing fees and essentially regulate prices, which is a difficult task for a specialized regulatory body, let alone for a generalist court.⁸⁴

In the rivalrous case, the issue of investment incentives can be especially strong: Not only would the platform have to offer something of value to third-party sellers; the platform would not be allowed to offer that value to its first-party seller.

Next, consider the criterion that a firm should not benefit too much from past successes. To some degree, self-preferencing can be characterized as a form of tying.⁸⁵ Like other forms of tying, self-preferencing can be condemned as being unfair and/or not on the merits because the firm uses success in one market to obtain advantage in another. But self-preferencing can have positive and negative effects on competition.⁸⁶

Moreover, self-preferencing can also be viewed as a means of allowing a platform to monetize past investments: The platform owner generally will have attained its current market position by having invested in the platform. Antitrust policy generally favors allowing a firm to earn a return on its past investments. If it is fair or competition on the merits for a firm to take advantage of past successes within a market (at least to some degree), why isn’t it also fair to do so across markets? An enforcement policy that limits the ability to benefit from success in other markets risks becoming an attack on systems-level competition and vertical integration.⁸⁷ Surely a firm is allowed to enjoy some benefit from vertical integration. But how much is too much? Here too, there is a need for a (currently missing) limiting principle.

6 Conclusion

The analysis above demonstrates that there is not a sound basis for declaring that self-preferencing should be either per se legal or per se illegal. Instead, the competitive effects of self-preferencing vary with the circumstances, which suggests that – given our current state of knowledge – a case-by-case evaluation is appropriate.

What principles should be applied in evaluating any given case? Determining whether competition is “fair” or on the “merits” is central to the fairness and consumer welfare approaches to antitrust, respectively. Alone, however, neither an appeal to “fairness” nor to “merit” is sufficient to evaluate whether conduct harms competition.⁸⁸ Rather than rely on vaguely defined words, additional work – in the forms of both academic research and fact-intensive court cases – should be undertaken to identify the linkages between market conditions and the equilibrium effects of self-preferencing.

“When I make a word do a lot of work like that,” said Humpty Dumpty, “I always pay it extra.”⁸⁹