Big data, machine learning and artificial intelligence are rapidly changing the way businesses make strategic decisions. Antitrust enforcers and practitioners are particularly interested in how these tools are used in areas, such as pricing, that involve observation of, and reaction to, competitors' choices.

Anticompetitive use of automated pricing systems already has been the target of enforcement actions by the US Department of Justice as well as the subject of intense debate among antitrust practitioners and academics.1 US enforcers take the view that the legal standard for finding unlawful collusion does not need to change in the context of pricing algorithms, and that independent use of pricing algorithms that interact with each other, without an agreement to fix prices, should not give rise to liability.2

But increasing use of pricing algorithms does pose enforcement challenges. Algorithms may facilitate collusion and complicate detection of unlawful agreements. Technological advances provide a means for competitors to more easily engage in parallel pricing—potentially choosing specific pricing algorithms for that purpose. Third parties marketing these new pricing tools may be in a position to facilitate actual collusion. Widespread use of pricing algorithms also has the potential to magnify the competitive impact of a merger by facilitating tacit collusion among larger numbers of competitors. All of this has led some to call for rethinking whether current law adequately addresses the potential for anticompetitive outcomes in the digital age.3

In light of the uncertainty surrounding these new technologies, businesses using or considering the use of pricing algorithms may have concerns that doing so could expose them to costly litigation or intrusive investigations. As we discuss, while companies should be careful how they select, formulate and use pricing algorithms, there is no reason to forego newly-developed technologies for fear of antitrust liability as long as the adoption is unilateral and justified by a business rationale. At the same time, businesses should be aware that enforcers will likely continue to closely scrutinize use of these tools as technology evolves.

Background

An algorithm is a sequence of rules to be applied in precise order to carry out a task. Although people usually associate algorithms with formulas and computer code, a set of instructions for cooking a meal or a diagram for assembling furniture is also an algorithm. Barry Nigro, a Deputy Assistant Attorney General for Antitrust at DOJ, described algorithms at a recent GCR conference as "a bunch of if-then statements" written by humans that tell a computer what to do.4 Other senior US antitrust enforcers have emphasized that algorithms are tools to achieve particular goals, and it is people who determine the goals and then choose or design the appropriate algorithms to achieve the goals.5

Pricing algorithms are designed to collect and analyze a large amount of relevant market data and to price products and services taking into account a vast set of factors. With the help of a pricing algorithm, a business can react almost instantaneously to price movements by competitors. Moreover, because a computer program generates the actual prices for transactions and because the computer program evaluates complex data with speed and sophistication that a human cannot replicate, the use of pricing algorithms creates a perception that price-setting is entirely machine-driven. But it is the human element that creates potential antitrust liability.

There has been recent enforcement in this space. In 2015, the DOJ successfully prosecuted David Topkins—and charged others—for what could be characterized as algorithm-enhanced price-fixing. Topkins and his co-conspirators adopted specific pricing algorithms to coordinate prices for wall posters they sold through the Amazon Marketplace. In particular, Topkins and his co-conspirators wrote the computer code that instructed algorithm-based software to avoid price competition. Topkins pleaded guilty of violating Section 1 of the Sherman Act and agreed to pay a $20,000 criminal fine.6

While Topkins was the first criminal case filed by DOJ against an anti-competitive conspiracy in e-commerce, it was not the first case where US enforcers charged horizontal competitors with using electronic tools to collude. Back in 1994, DOJ settled charges that six airlines used a jointly owned computerized online booking system, the Airline Tariff Publishing Company (ATPCo.), to collude in setting airline fares. The original purpose of the ATPCo. was to disseminate fare information to travel agents and the public, but certain airlines used the system to signal each other about fare changes and effectively reach understandings that limited discounting.7

More recently, four major financial institutions and certain individuals they employed were charged criminally with colluding through an online chatroom to manipulate foreign currency exchange rates.8 The banks pleaded guilty to the charge and the individuals are awaiting trial. Again, the key dispositive allegation in these cases was that humans intentionally struck agreements. The use of a chat room to carry out the schemes was just part of the mechanism by which the scheme worked.

Where pricing algorithms are used to implement price-fixing conspiracies or to convey information between competitors—enforcement is fairly straightforward. In contrast, consider a case alleging that the selection of the pricing algorithm itself constitutes the anti-competitive agreement. In late 2015, private litigants sued Uber alleging that the pricing and payments mechanism at the heart of the Uber app violates the Sherman Act.9 They alleged that the Uber app supported a hub-and-spoke conspiracy, where each driver charged prices determined centrally by the algorithm, knowing that other drivers would not be undercutting that price. The court found the allegations in the complaint sufficient to withstand a motion to dismiss.10

Uber succeeded in removing the case to arbitration, so plaintiffs' contention that the algorithm played a central role in the conspiracy will not be tested through discovery and trial. Nonetheless, the court reached its decision to deny the motion to dismiss by applying established precedent to the alleged facts. The court pointed specifically to allegations that drivers would sign up for Uber on the understanding that all Uber drivers were agreeing to the same pricing algorithm. The court also considered the allegations that charging the algorithm-determined prices was against the drivers' own interest if they were working independently. The court was not persuaded by Uber's argument, at the motion to dismiss stage, that drivers had made independent decisions to enter in a vertical agreement with Uber in order to take advantage of the payment processing and rider matching services that the app enables. The court found that the plaintiffs had alleged a plausible conspiracy because the complaint contained sufficient allegations showing that the drivers and Uber knowingly agreed on a mechanism to coordinate prices. The fact that the mechanism involved a high-tech pricing algorithm did not determine the ruling. Ultimately, as Judge Rakoff wrote in his opinion in Uber, "automation is effected through a human design."11

Finally, scholars, including Professor Joseph Harrington, have described potential anti-competitive effects associated with pricing algorithms that employ artificial intelligence. He argues that, as these algorithms perform legitimate optimization calculations and operations, they can at the same time potentially reduce price competition.12 The implications of using such algorithms and their ability to affect markets in ways not necessarily intended by their designers or adopters are currently the subject to academic research and debate.

Concerted Action and Pricing Algorithms: No Liability Without Agreement

Algorithm-enabled collusion scenarios have dominated the discussion on antitrust aspects of digital pricing tools. What do US enforcers think about the issue? For the most part they take the view that the use of pricing algorithms are not novel from an antitrust perspective and do not call for new theories of competitive harm, even though they may raise concerns about the facilitation of collusion or the difficulty of detection. Particularly with regard to concerted action, they say that conduct in markets involving pricing algorithms can be evaluated within the existing analytical framework and guided by established laws and precedent.

Maureen Ohlhausen, Acting Chair of the FTC, made this point in her remarks to the Concurrences Conference on Antitrust in the Financial Sector with a simple but telling example. She described two situations: one where the owners of two gas stations on opposite sides of a road signal price increases to each other by changing the prices on the board; and the other, where each of the gas station owners determines the price to charge using a computer programed to take into account the price of the other. As long as the gas station owners are acting unilaterally, their actions are outside the scope of antitrust liability, but if they get together and agree on a course of action, they would be liable for price-fixing.13

Andrew Finch, the Principal Deputy Assistant Attorney General for Antitrust, and his colleagues at the DOJ have also reiterated that proof of agreement is key in determining whether parallel conduct amounts to an antitrust violation under US law.14 Enforcers say that they do not underestimate the unique ways in which pricing algorithms may support collusive arrangements or enable parallel pricing. In particular, the algorithms' market monitoring functions allow firms to observe competitors' price movements and react swiftly. In the context of a cartel, this ensures that cheaters are more readily detected and allows for almost instantaneous punishment. This in turn makes the cartel more stable and its detection more difficult.

Absent an anti-competitive agreement, however, the algorithm's market monitoring function may simply enable firms to observe and match competitors' prices. This type of parallel pricing is not illegal just as it would not be illegal if achieved without the help of a pricing algorithm. Enforcers suggest that they will focus, however, on whether firms reached an understanding in determining whether what appears to be parallel pricing is better characterized as collusion.

These observations by antitrust agency leadership echoed the conclusion in a joint memorandum that DOJ and FTC submitted to the OECD in June 2017. The memorandum lays out the analytical framework for analyzing concerted action under US antitrust laws and the experience with computerized pricing tools in that context, including the Topkins and ATPCo. cases. The agencies emphasized that US antitrust laws focus on challenging anticompetitive behavior rather than market outcomes. To prosecute companies for unilateral use of pricing algorithms would be contrary to this core principle even where the use of the pricing algorithm is deemed to have contributed to supra-competitive prices. The agencies reiterated that "[w]ithout proof of collusion or evidence that the knowing parallel adoption of pricing formulas narrowed the range of prices over time, parallel pricing conduct may be outside the reach of the antitrust laws."15

Pricing Algorithms and Merger Analysis: Closer Scrutiny Likely for Mergers in Industries Using Digital Pricing Tools

Pricing algorithms present a different issue in merger review. The ability of pricing algorithms to magnify the effects of increasing concentration in a market may become an increasing focus in merger analysis.

In a recent article, FTC Commissioner Terrell McSweeny and Brian O'Dea discuss how the use of pricing algorithms may result in more merger challenges.16 The ability of algorithms to analyze and process large amounts of data about consumer characteristics enables businesses to engage in sophisticated customer segmentation. This in turn leads to narrower antitrust relevant markets. A merger may result in a 6-to-5 market structure for some segments, a 3-to-2 market structure for other segments and even a merger to monopoly in others. Under the Horizontal Merger Guidelines, the agencies can challenge a merger if it is likely to be anticompetitive in any relevant market, even a relatively small part of an overall product market.17 Thus, the more dimensions the algorithm uses for differentiated pricing, the more market segments would constitute relevant antitrust markets and the greater the likelihood that the merger may substantially decrease competition in one or more of the smaller, price-discrimination markets.

Enforcers will face tough choices: exercise prosecutorial discretion and allow mergers with clear and sizable benefits to most of the consumers to proceed even though a small part of consumers would be harmed; or seek an injunction where there is no way to eliminate the harm other than blocking the transaction.18

Separately, companies and their counsel should anticipate that the agencies investigating mergers in industries where pricing algorithms are involved may look more closely at the risk of coordinated interaction post-merger. Indeed, McSweeny and O'Dea suggest that enforcers likely will need to assess coordinated effects at lower market concentration thresholds because algorithms potentially could support coordination and parallel pricing among larger groups of competitors.19

Minimizing Antitrust Risks Associated with Pricing Algorithms

The majority of reasonably foreseeable issues arising out of the use of pricing algorithms can be analogized to "low-tech" conduct for which the law is settled and guidance on risk mitigation is available from antitrust counsel. As a rule of thumb, using an algorithm or a formula will not make conduct lawful if that conduct is unlawful without the use of the algorithm. Below is some additional practical advice for companies using or considering the use of pricing algorithms:

  • Ensure that the company antitrust compliance program includes training specific to the use of pricing algorithms. Remind employees and decisionmakers that agreements with competitors on setting a particular price or market allocation are unlawful even if executed through an algorithm.
  • Instruct employees that the use of algorithms and the features of these algorithms constitute competitively sensitive information that should not be shared with competitors. If you choose to design a pricing algorithm in-house, the engineers and other staff involved must understand the importance of keeping the information about the algorithm within the company.
  • Beware of adopting a particular algorithm or software with the understanding that others in the industry are using or will be using it and that it will help coordinate or stabilize pricing. When choosing pricing software developed or sold by a third party, discourage representations by the software manufacturer about sales to your competitors or other information revealing how competitors use the software. This may be difficult because the sales pitch would naturally include successful implementation of the software within your industry. Consult antitrust counsel about mitigation and have them present in discussions with the software developer.
  • Carefully document the unilateral nature of the decision to adopt the particular algorithm or software and the business justifications for adopting the algorithm or software.
  • Seek advice by counsel about treatment of pricing algorithms in non-US jurisdictions where the company operates or sells products.
  • Expect heightened scrutiny of possible coordinated effects in mergers. In preparation:

    • Identify the effects of using a pricing algorithm.
    • Analyze the competitive landscape in each market segment defined by the pricing algorithm.
    • Present quantifiable merger-specific efficiencies in the market segments where there is no concentration issue to show how they would outweigh the potential harm in the market segments where concentration would increase to levels causing concern.

Footnotes

1. See, e.g., United States v. Topkins, No. CR 15-00201 (N.D. Cal. 2015); GCRI, US DOJ Deputy: Algorithmic Cartel Requires Agreement, (Miami, Feb. 3, 2018); Terrell McSweeny & Brian O'Dea, The Implications of Algorithmic Pricing for Coordinated Effects Analysis and Price Discrimination Markets in Antitrust Enforcement, Antitrust, Fall 2017 at 75; Maureen Ohlhausen, Remarks to the Concurrences Conference on Antitrust in the Financial Sector: "Should We Fear The Things That Go Beep In the Night? Some Initial Thoughts on the Intersection of Antitrust Law and Algorithmic Pricing" (New York, May 23, 2017); OECD Roundtable: Algorithms and Collusion (Paris, June 21-23, 2017) (information and documents).

2. See, e.g., Algorithms and Collusion—Note by the United States, May 26, 2017,official submission to the OECD Roundtable; Ohlhausen, supra note 1; GCRI, supranote 1.

3. See, e.g., Ariel Ezrachi & Maurice E. Stucke, Virtual Competition, The Promise And Perils Of The Algorithm-Driven Economy, (Harvard Univ. Press 2016).

4. GCRI, supra note 1.

5. Principal Deputy Assistant Attorney General for Antitrust Andrew Finch discussed pricing algorithms at a New York State Bar Association panel "Antitrust in High Speed: Colluding Through Algorithms and Other Technologies" (New York, January 8, 2017). Former Assistant Attorney General for Antitrust Bill Baer and Professor Joseph Harrington were among the panelists. See also Ohlhausen, supra note 1.

6. United States v. Topkins, No. CR 15-00201 (N.D. Cal. 2015) (Information and Plea Agreement). DOJ brought additional indictments against Daniel William Aston and a company he managed, Trod Ltd. (doing business as Buy 4 Less, Buy For Less, and Buy-For-Less-Online), a U.K. company headquartered in Birmingham, England. DOJ reached a plea agreement with Trod, Ltd. (Information and Plea Agreement). The UK Competition and Markets Authority (CMA) disqualified Aston from holding company directorships for 5 years. See Press Release.

7. United States v. Airline Tariff Publishing Co., 836 F. Supp. 9 (D.D.C. 1993).

8. See DOJ Press Release No. 17-030, Three Former Traders for Major Banks Indicted in Foreign Currency Exchange Antitrust Conspiracy (January 10, 2017); DOJ Press Release No. 15-643 Five Major Banks Agree to Parent-Level Guilty Pleas (May 20, 2015).

9. Meyer v. Kalanick, No. 1:15-cv-09796-JSR (S.D.N.Y.). The class plaintiffs also alleged that Uber's pricing algorithm was a vertical price maintenance arrangement and prevented Uber drivers from discounting the price for their services.

10. Id. Opinion and Order, ECF No. 37 at 12-13.

11. Id. at 16 (citing United States v. Ulbricht, 79 F. Supp. 3d 466, 559 (S.D.N.Y. 2014)).

12. See Joseph E. Harrington, Jr., Developing Competition Law for Collusion by Autonomous Price-Setting Agents (Aug 22, 2017).

13. Ohlhausen, supra note 1 at 3-4.

14. See Andrew Finch, Remarks at the 44th Annual Conference on International Antitrust Law and Policy (New York, September 14, 2017). See also GCRI, supra note 1.

15. Algorithms and Collusion - Note by the United States, supra note 1 at 4.

16. See McSweeny, supra note 1.

17. US Dep't of Justice & Fed. Trade Comm'n, Horizontal Merger Guidelines, § 3.

18. See McSweeny, supra note 1 at 79; see also Horizontal Merger Guidelines, supra note 16, § 10 n.14 (discussing exercise of prosecutorial discretion).

19. Id.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.