The First Priority of Antitrust Analysis is Getting It Right, Not Making it Easier

Hurwitz & Stout —  5 March 2018 — 2 Comments

Excess is unflattering, no less when claiming that every evolution in legal doctrine is a slippery slope leading to damnation. In Friday’s New York Times, Lina Khan trots down this alarmist path while considering the implications for the pending Supreme Court case of Ohio v. American Express. One of the core issues in the case is the proper mode of antitrust analysis for credit card networks as two-sided markets. The Second Circuit Court of Appeals agreed with arguments, such as those that we have made, that it is important to consider the costs and benefits to both sides of a two-sided market when conducting an antitrust analysis. The Second Circuit’s opinion is under review in the American Express case.

Khan regards the Second Circuit approach of conducting a complete analysis of these markets as a mistake.

On her reading, the idea that an antitrust analysis of credit card networks should reflect their two-sided-ness would create “de facto antitrust immunity” for all platforms:

If affirmed, the Second Circuit decision would create de facto antitrust immunity for the most powerful companies in the economy. Since internet technologies have enabled the growth of platform companies that serve multiple groups of users, firms like Alphabet, Amazon, Apple, Facebook, and Uber are set to be prime beneficiaries of the Second Circuit’s warped analysis. Amazon, for example, could claim status as a two-sided platform because it connects buyers and sellers of goods; Google because it facilitates a market between advertisers and search users… Indeed, the reason that the tech giants are lining up behind the Second Circuit’s approach is that — if ratified — it would make it vastly more difficult to use antitrust laws against them.

This paragraph is breathtaking. First, its basic premise is wrong. Requiring a complete analysis of the complicated economic effects of conduct undertaken in two sided markets before imposing antitrust liability would not create “de facto antitrust immunity.” It would require that litigants present, and courts evaluate, credible evidence sufficient to establish a claim upon which an enforcement action can be taken — just like in any other judicial proceeding in any area of law. Novel market structures may require novel analytical models and novel evidence, but that is no different with two-sided markets than with any other complicated issue before a court.

Second, the paragraph’s prescribed response would be, in fact, de facto antitrust liability for any firm competing in a two-sided market — that is, as Kahn notes, almost every major tech firm.

A two-sided platform competes with other platforms by facilitating interactions between the two sides of the market. This often requires a careful balancing of the market: in most of these markets too many or too few participants on one side of the market reduces participation on the other side. So these markets play the role of matchmaker, charging one side of the market a premium in order to cross-subsidize a desirable level of participation on the other. This will be discussed more below, but the takeaway for now is that most of these platforms operate by charging one side of the market (or some participants on one side of the market) an above-cost price in order to charge the other side of the market a below-cost price. A platform’s strategy on either side of the market makes no sense without the other, and it does not adopt practices on one side without carefully calibrating them with the other. If one does not consider both sides of these markets, therefore, the simplistic approach that Kahn demands will systematically fail to capture both the intent and the effect of business practices in these markets. More importantly, such an approach could be used to find antitrust violations throughout these industries — no matter the state of competition, market share, or actual consumer effects.

What are two-sided markets?

Khan notes that there is some element of two-sidedness in many (if not most) markets:

Indeed, almost all markets can be understood as having two sides. Firms ranging from airlines to meatpackers could reasonably argue that they meet the definition of “two-sided,” thereby securing less stringent review.

This is true, as far as it goes, as any sale of goods likely involves the selling party acting as some form of intermediary between chains of production and consumption. But such a definition is unworkably broad from the point of view of economic or antitrust analysis. If two-sided markets exist as distinct from traditional markets there must be salient features that define those specialized markets.

Economists have been intensively studying two-sided markets (see, e.g., here, here, and here) for the past two decades (and had recognized many of their basic characteristics even before then). As Khan notes, multi-sided platforms have indeed existed for a long time in the economy. Newspapers, for example, provide a targeted outlet for advertisers and incentives for subscribers to view advertisements; shopping malls aggregate retailers in one physical location to lower search costs for customers, while also increasing the retailers’ sales volume. Relevant here, credit card networks are two-sided platforms, facilitating credit-based transactions between merchants and consumers.

One critical feature of multi-sided platforms is the interdependent demand of platform participants. Thus, these markets require a simultaneous critical mass of users on each side in order to ensure the viability of the platform. For instance, a credit card is unlikely to be attractive to consumers if few merchants accept it; and few merchants will accept a credit card that isn’t used by a sufficiently large group of consumers. To achieve critical mass, a multi-sided platform uses both pricing and design choices, and, without critical mass on all sides, the positive feedback effects that enable the platform’s unique matching abilities might not be achieved.

This highlights the key distinction between traditional markets and multi-sided markets. Most markets have two sides (e.g., buyers and sellers), but that alone doesn’t make them meaningfully multi-sided. In a multi-sided market a key function of the platform is to facilitate the relationship between the sides of the market in order to create and maintain an efficient relationship between them. The platform isn’t merely a reseller of a manufacturer’s goods, for instance, but is actively encouraging or discouraging participation by users on both sides of the platform in order to maximize the value of the platform itself — not the underlying transaction — for those users. Consumers, for instance, don’t really care how many pairs of jeans a clothier stocks; but a merchant does care how many cardholders an issuer has on its network. This is most often accomplished by using prices charged to each side (in the case of credit cards, so-called interchange fees) to keep each side an appropriate size.

Moreover, the pricing that occurs on a two-sided platform is secondary, to a varying extent, to the pricing of the subject of the transaction. In a two-sided market, the prices charged to either side of the market are an expression of the platform’s ability to control the terms on which the different sides meet to transact and is relatively indifferent to the thing about which the parties are transacting.

The nature of two-sided markets highlights the role of these markets as more like facilitators of transactions and less like traditional retailers of goods (though this distinction is a matter of degree, and different two-sided markets can be more-or-less two-sided). Because the platform uses prices charged to each side of the market in order to optimize overall use of the platform (that is, output or volume of transactions), pricing in these markets operates differently than pricing in traditional markets. In short, the pricing on one side of the platform is often used to subsidize participation on the other side of the market, because the overall value to both sides is increased as a result. Or, conversely, pricing to one side of the market may appear to be higher than the equilibrium level when viewed for that side alone, because this funds a subsidy to increase participation on another side of the market that, in turn, creates valuable network effects for the side of the market facing the higher fees.

The result of this dynamic is that it is more difficult to assess the price and output effects in multi-sided markets than in traditional markets. One cannot look at just one side of the platform — at the level of output and price charged to consumers of the underlying product, say — but must look at the combined pricing and output of both the underlying transaction as well as the platform’s service itself, across all sides of the platform.

Thus, as David Evans and Richard Schmalensee have observed, traditional antitrust reasoning is made more complicated in the presence of a multi-sided market:

[I]t is not possible to know whether standard economic models, often relied on for antitrust analysis, apply to multi-sided platforms without explicitly considering the existence of multiple customer groups with interdependent demand…. [A] number of results for single-sided firms, which are the focus of much of the applied antitrust economics literature, do not apply directly to multi-sided platforms.

The good news is that antitrust economists have been focusing significant attention on two- and multi-sided markets for a long while. Their work has included attention to modelling the dynamics and effects of competition in these markets, including how to think about traditional antitrust concepts such as market definition, market power and welfare analysis. What has been lacking, however, has been substantial incorporation of this analysis into judicial decisions. Indeed, this is one of the reasons that the Second Circuit’s opinion in this case was, and why the Supreme Court’s opinion will be, so important: this work has reached the point that courts are recognizing that these markets can and should be analyzed differently than traditional markets.

Getting the two-sided analysis wrong in American Express would harm consumers

Khan describes credit card networks as a “classic case of oligopoly,” and opines that American Express’s contractual anti-steering provision is, “[a]s one might expect, the credit card companies us[ing] their power to block competition.” The initial, inherent tension in this statement should be obvious: the assertion is simultaneously that this a non-competitive, oligopolistic market and that American Express is using the anti-steering provision to harm its competitors. Indeed, rather than demonstrating a classic case of oligopoly, this demonstrates the competitive purpose that the anti-steering provision serves: facilitating competition between American Express and other card issuers.

The reality of American Express’s anti-steering provision, which prohibits merchants who choose to accept AmEx cards from “steering” their customers to pay for purchases with other cards, is that it is necessary in order for American Express to compete with other card issuers. Just like analysis of multi-sided markets needs to consider all sides of the market, platforms competing in these markets need to compete on all sides of the market.

But the use of complex pricing schemes to determine prices on each side of the market to maintain an appropriate volume of transactions in the overall market creates a unique opportunity for competitors to behave opportunistically. For instance, if one platform charges a high fee to one side of the market in order to subsidize another side of the market (say, by offering generous rewards), this creates an opportunity for a savvy competitor to undermine that balancing by charging the first side of the market a lower fee, thus attracting consumers from its competitor and, perhaps, making its pricing strategy unprofitable. This may appear to be mere price competition. But the effects of price competition on one side of a multi-sided market are more complicated to evaluate than those of traditional price competition.

Generally, price competition has the effect of lowering prices for goods, increasing output, decreasing deadweight losses, and benefiting consumers. But in a multi-sided market, the high prices charged to one side of the market can be used to benefit consumers on the other side of the market; and that consumer benefit can increase output on that side of the market in ways that create benefits for the first side of the market. When a competitor poaches a platform’s business on a single side of a multi-sided market, the effects can be negative for users on every side of that platform’s market.

This is most often seen in cases, like with credit cards, where platforms offer differentiated products. American Express credit cards are qualitatively different than Visa and Mastercard credit cards; they charge more (to both sides of the market) but offer consumers a more expansive rewards program (funded by the higher transaction fees charged to merchants) and offer merchants access to what are often higher-value customers (ensured by the higher fees charged to card holders).

If American Express did not require merchants to abide by its anti-steering rule, it wouldn’t be able to offer this form of differentiated product; it would instead be required to compete solely on price. Cardholders exist who prefer higher-status cards with a higher-tier of benefits, and there are merchants that prefer to attract a higher-value pool of customers.

But without the anti-steering provisions, the only competition available is on the price to merchants. The anti-steering rule is needed in order to prevent merchants from free-riding on American Express’s investment in attracting a unique group of card holders to its platform. American Express maintains that differentiation from other cards by providing its card holders with unique and valuable benefits — benefits that are subsidized in part by the fees charged to merchants. But merchants that attract customers by advertising that they accept American Express cards but who then steer those customers to other cards erode the basis of American Express’s product differentiation. Because of the interdependence of both sides of the platform, this ends up undermining the value that consumers receive from the platform as American Express ultimately withdraws consumer-side benefits. In the end, the merchants who valued American Express in the first place are made worse off by virtue of being permitted to selectively free-ride on American Express’s network investment.

At this point it is important to note that many merchants continue to accept American Express cards in light of both the cards’ higher merchant fees and these anti-steering provisions. Meanwhile, Visa and Mastercard have much larger market shares, and many merchants do not accept Amex. The fact that merchants who may be irritated by the anti-steering provision continue to accept Amex despite it being more costly, and the fact that they could readily drop Amex and rely on other, larger, and cheaper networks, suggests that American Express creates real value for these merchants. In other words, American Express, in fact, must offer merchants access to a group of consumers who are qualitatively different from those who use Visa or Mastercard cards — and access to this group of consumers must be valuable to those merchants.

An important irony in this case is that those who criticize American Express’s practices, who are arguing these practices limit price competition and that merchants should be able to steer customers to lower-fee cards, generally also argue that modern antitrust law focuses too myopically on prices and fails to account for competition over product quality. But that is precisely what American Express is trying to do: in exchange for a higher price it offers a higher quality card for consumers, and access to a different quality of consumers for merchants.

Anticompetitive conduct here, there, everywhere! Or nowhere.

The good news is that many on the court — and, for that matter, even Ohio’s own attorney — recognize that the effects of the anti-steering rule on the cardholder side of the market need to be considered alongside their effects on merchants:

JUSTICE KENNEDY: Does output include premiums or rewards to customers?
MR. MURPHY: Yeah. Output would include quality considerations as well.

The bad news is that several justices don’t seem to get it. Justice Kagan, for instance, suggested that “the effect of these anti-steering provisions means a market where we will only have high-cost/high-service products.” Justice Kagan’s assertion reveals the hubris of the would-be regulator, bringing to her evaluation of the market a preconception of what that market is supposed to look like. To wit: following her logic, one can say just as much that without the anti-steering provisions we would have a market with only low-cost/low-service products. Without an evaluation of the relative effects — which is more complicated than simple intuition suggests, especially since one can always pay cash — there is no reason to say that either of these would be a better outcome.

The reality, however, is that it is possible for the market to support both high- and low-cost, and high- and low-service products. In fact, this is the market in which we live today. As Justice Gorsuch said, “American Express’s agreements don’t affect MasterCard or Visa’s opportunity to cut their fees … or to advertise that American Express’s are higher. There is room for all kinds of competition here.” Indeed, one doesn’t need to be particularly creative to come up with competitive strategies that other card issuers could adopt, from those that Justice Gorsuch suggests, to strategies where card issuers are, in fact, “forced” to accept higher fees, which they in turn use to attract more card holders to their networks, such as through sign-up bonuses or awards for American Express customers who use non-American Express cards at merchants who accept them.

A standard response to such proposals is “if that idea is so good, why isn’t the market already doing it?” An important part of the answer in this case is that MasterCard and Visa know that American Express relies on the anti-steering provision in order to maintain its product differentiation.

Visa and Mastercard were initially defendants in this case, as well, as they used similar rules to differentiate some of their products. It’s telling that these larger market participants settled because, to some extent, harming American Express is worth more to them than their own product differentiation. After all, consumers steered away from American Express will generally use Visa or Mastercard (and their own high-priced cards may be cannibalizing from their own low-priced cards anyway, so reducing their value may not hurt so much). It is therefore a better strategy for them to try to use the courts to undermine that provision so that they don’t actually need to compete with American Express.

Without the anti-steering provision, American Express loses its competitive advantage compared to MasterCard and Visa and would be forced to compete against those much larger platforms on their preferred terms. What’s more, this would give those platforms access to American Express’s vaunted high-value card holders without the need to invest resources in competing for them. In other words, outlawing anti-steering provisions could in fact have both anti-competitive intent and effect.

Of course, card networks aren’t necessarily innocent of anticompetitive conduct, one way or the other. Showing that they are on either side of the anti-steering rule requires a sufficiently comprehensive analysis of the industry and its participants’ behavior. But liability cannot be simply determined based on behavior on one side of a two-sided market. These companies can certainly commit anticompetitive mischief, and they need to be held accountable when that happens. But this case is not about letting American Express or tech companies off the hook for committing anticompetitive conduct. This case is about how we evaluate such allegations, weigh them against possible beneficial effects, and put in place the proper thorough analysis for this particular form of business.

Over the last two decades, scholars have studied the nature of multi-sided platforms, and have made a good deal of progress. We should rely on this learning, and make sure that antitrust analysis is sound, not expedient.

Hurwitz & Stout


Justin "Gus" Hurwitz & Kristian Stout

2 responses to The First Priority of Antitrust Analysis is Getting It Right, Not Making it Easier


    You characterize Ms. Khan’s editorial as rejecting “conducting a complete analysis of these markets”, yet as I read her editorial, what she actually rejects is that “the Second Circuit held that the government would have to show that any price increases for merchants also harmed cardholders, or at least didn’t benefit them.” That appears to me to actually be calling out the placement of the burden of proof of harm across both sides of the market on plaintiffs. I do not see evidence anywhere in that editorial that she is making the broader rejection you characterize.

    Elsewhere, some prominent critics, such as Hal Singer, have very clearly focused specifically on the issue of where the burden should lie — i.e., not rejecting the use of a complete analysis, as you suggest, but rather, arguing that once the plaintiff has met the burden of showing harm on one side of the market, the burden should shift to defendants to demonstrate that those harms are fully offset by benefits on the other side of the market.

    That seems a far more sensible and defensible argument than the outright rejection of two-sided analysis that you are characterizing her position to be.

    The alternative of placing the additional burden on plaintiffs is effectively to presume that any harm on one side of the market is fully offset by benefits on the other side. Even as a non-economist, that sounds like a highly questionable proposition — perhaps even just plain crazy. How does that approach differ from saying, for example, that rather than self-defense being an affirmative defense to murder, the prosecution must not only prove the elements of murder, but also must affirmatively prove that the murder wasn’t in self-defense, without the defendant bearing any burden at all to first show actual evidence of self-defense?


      A further thought, that I think clearly weighs heavily in favor of the burden shifting to defendants, after plaintiffs have shown harm on one side of a market.

      In his book How To Regulate, your esteemed colleague Thomas Lambert observes that regulators’ goal — i.e., what it means to “Get it right” — is to minimize the sum of error costs and decision costs.

      How does that work out here?

      Well, first, decision costs — evaluating the impact across the entire market — are at best equal for plaintiffs and defendants — in which case, we should be indifferent to who bears the brunt of them, economically-speaking. On the other hand, some argue, probably reasonably, that defendants actually possess significantly better information respecting the total market impact, which would mean decision costs are significantly lower for defendants than for plaintiffs, and would argue for placing the burden on defendants.

      But even if we assume decisions costs are equal, and so could be placed either way, indifferently in terms of themselves, how does the allocation of the burden affect error costs?

      Well, if the assessment is difficult and plaintiffs bear the burden, then plaintiffs will take action to intervene in only a fraction — probably a small fraction — of cases where there is actual harm, because they will face difficulty assessing where there is actual harm, and potentially even greater difficulty meeting their burden to prove such. Since companies have a strong incentives to implement restrictions that benefit themselves, the harder it is to analyze the full market impact, the more likely it is that companies will implement restrictions that benefit themselves while harming consumers — they can feel comfortable that the move likely won’t be challenged. In other words, assigning the burden to the plaintiffs generates low Type I errors (inappropriate condemnation), but high Type II errors (inappropriate failure to condemn).

      On the other hand, if we assign the burden to defendants, then companies will have a strong incentive to assess the total market impact before enacting restrictions that harm one side of the market, and decline to proceed where that is not fully offset by gains on the other side of the market. That means low Type II errors. At the same time, where there is a significant opportunity for economic gain, companies have a very strong incentive to undertake the effort to conclude there is a net market benefit. So that means Type I errors almost certainly do not increase nearly as much as Type II errors decrease.

      Furthermore, since these two-sided markets are generally huge, errors costs almost certainly dominate decisions costs. That is, even if we are increasing decision costs by forcing companies to do the hard work of assessing the full market impact/defensibility of every restriction up front, that increase in costs is probably dwarfed by the savings due to error cost reductions.

      So it seems to me that minimizing the sum of the costs in this case — i.e., “Getting it right” — dictates shifting the burden of proof to defendants, once plaintiffs have successfully proven harm on one side of the market.

Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s