Financial Times: Policing the Digital Cartels

January 9, 2017

By David Lynch

Price-setting algorithms mean regulators must now tackle collusion among machines

David Topkins is no John D Rockefeller. But like the famed industrialist, the unheralded ecommerce executive has stirred fundamental concerns about the laws of economic competition in the digital age.

In the first criminal antitrust prosecution of its kind, Mr Topkins pleaded guilty in a San Francisco federal court in 2015 to rigging prices for classic cinema posters sold through Amazon’s online marketplace.

Even if the crime seems unremarkable, his method was revolutionary: Mr Topkins admitted to manipulating the market by programming customised algorithms to keep prices artificially high. Once his rivals agreed to the plan, the algorithm automatically maintained what prosecutors called “collusive, non-competitive prices” on printed wall art.

As Mr Topkins waits to learn his fate, US authorities and their European counterparts are beginning to grasp the implications of these increasingly powerful online tools. Mr Topkins’ modest poster sales pale alongside the mighty Rockefeller oil trust, whose 90 per cent share of the market prompted the nation’s first antitrust law in 1890. But by highlighting technologies capable of distorting markets in unfamiliar ways, his prosecution is a digital economy milestone.

“We will not tolerate anti-competitive conduct, whether it occurs in a smoke-filled room or over the internet using complex pricing algorithms,” William Baer, principal deputy associate attorney-general at the Department of Justice, insisted when Mr Topkins’ indictment was unsealed.

Yet existing antitrust laws, premised on human intent and action, may be inadequate to prevent companies from abusing their market power in the digital era, some experts say.

Markets dominated by “robo-sellers,” or automated pricing bots, will not respond to the same incentives or operate according to the same rules as those managed by people.

These concerns suggest that the digital economy’s promise of lower prices and plentiful choice for consumers could evaporate. The rise of artificial intelligence and powerful algorithms may instead create more durable cartels that are able to maintain higher prices at consumers’ expense and in defiance of traditional enforcement regimes.

“The stakes are a lot higher in the data-driven economy because of network effects,” says Maurice Stucke, a former federal antitrust prosecutor, who teaches at the University of Tennessee’s College of Law. “Competition as we know it is going to change.”

No evidence trail

For now, most regulators see this as a future concern. But as pricing systems become ever more autonomous, aspiring monopolists like Mr Topkins eventually will not even need to speak to their competitors to fix prices. Computers will do the colluding for them, either by using the same algorithm or learning from their interactions with other machines — all without leaving behind trails of incriminating emails or voicemails.

“Finding ways to prevent collusion between self-learning algorithms might be one of the biggest challenges that competition law enforcers have ever faced,” said a recent paper by the OECD, the Paris-based club of mostly rich nations.

These digital tools automatically calculate prices based on instantaneous assessments of supply and demand and a seller’s own instructions, such as specific profit or price targets.

“We’re talking about a velocity of decision-making that isn’t really human,” says Terrell McSweeny, a commissioner with the US Federal Trade Commission. “All of the economic models are based on human incentives and what we think humans rationally will do. It’s entirely possible that not all of that learning is necessarily applicable in some of these markets.”

First deployed in sectors such as quantitative finance, the use of algorithms is now entrenched in the airline and hotel industries and on online retailers such as Amazon. They are also spreading rapidly to other markets, including transportation, healthcare and consumer goods. Walmart’s $3.3bn acquisition of online retailer Jet.com in August was driven partly by a desire to improve its algorithmic capabilities.

The OECD report said the big data phenomenon “may pose serious challenges to competition authorities in the future, as it may be very difficult, if not impossible, to prove an intention to co-ordinate prices, at least using current antitrust tools”.

It added: “Particularly in the case of artificial intelligence, there is no legal basis to attribute liability to a computer engineer for having programmed a machine that eventually ‘self-learned’ to co-ordinate prices with other machines.”

President Barack Obama and leading Democrats such as Senator Elizabeth Warren have expressed concerns about a broader decline in competition in the US economy. Last April, Mr Obama’s Council of Economic Advisors released a report that found growing market share concentration in industries such as transportation, retail trade and insurance, which it blamed for slowing growth in living standards.

The White House also said new regulations may be needed to address specific antitrust concerns in the digital economy. “Price transparency can in some settings facilitate tacit collusion by enabling firms to see what other firms are charging, and hence easily detect any deviation from agreed-upon high prices,” the report said.

Smoke-filled rooms

The classic example of industrial-era price fixing dates back to a series of dinners hosted amid the 1907 financial panic by Elbert Gary, then chairman of US Steel.

In a narrow first-floor ballroom at New York’s Waldorf Astoria Hotel, men controlling 90 per cent of the nation’s steel output revealed to each other their respective wage rates, prices and “all information concerning their business”, one attendee recalled. Gary’s aim was to stabilise falling prices. The government later sued, saying that the dinner talks — the first of several over a four-year period — showed that US Steel was an illegal monopoly.

Algorithms render obsolete the need for such face-to-face plotting. Pricing tools scour the internet for competitors’ prices, prowl proprietary databases for relevant historical demand data, analyse digitised information and arrive at pricing solutions within milliseconds — far faster than any flesh-and-blood merchant could.

That should, in theory, result in lower prices and wider consumer choice. Algorithms raise antitrust concerns only in certain circumstances, such as when they are designed explicitly to facilitate collusion or parallel pricing moves by competitors.

“If the goal is to do bad things, automated systems and algorithms could help you do bad things faster,” says John Salch, technology leader with PROS Holdings Inc, a Houston-based pricing software company.

He does not believe such strategies are sustainable. But last year, German regulators warned that the competition-inhibiting effects of sophisticated algorithms could be difficult to prosecute. A report by the UK’s House of Lords last year, citing “the potential for anti-competitive behaviours” and “new forms of collusion”, called for the European Commission to conduct additional research on algorithms’ effects on competition.

“It’s an immediate concern,” says Mr Stucke, who along with his co-author Ariel Ezrachi of Oxford university, briefed antitrust officials in Washington and Brussels late last year. “We may have an anti-competitive result without necessarily having an antitrust fix.”

As an example, he cites a German software application that tracks petrol-pump prices. Preliminary results suggest that the app discourages price-cutting by retailers, keeping prices higher than they otherwise would have been. As the algorithm instantly detects a petrol station price cut, allowing competitors to match the new price before consumers can shift to the discounter, there is no incentive for any vendor to cut in the first place.

“Algorithms are sharing information so quickly that consumers are not aware of the competition,” says Mr Stucke. “Two gas stations that are across the street from each other are already familiar with this.”

This episode suggests that the availability of perfect information, a hallmark of free market theory, might harm rather than empower consumers. If the concern is borne out, a central assumption of the digital economy — that technology lowers prices and expands choices — could be upended.

An extreme example occurred in 2011, when a pair of pricing algorithms misfired over a developmental biology textbook sold on Amazon. Within days, The Making of A Fly: The Genetics of Animal Design, which sells for about $113, skyrocketed to more than $23m — thanks to the interaction between flawed pricing tools employed by two third-party sellers.

The first algorithm automatically priced its copy of the book at 1.27059 times the second seller’s price, while that vendor in turn fixed its price at 0.9983 times that of the first. The lack of a programmed price ceiling allowed the accidental spiral.

“What’s fascinating about all this is the seemingly endless possibilities for both chaos and mischief,” wrote Michael Eisen, the University of California, Berkeley, biologist on his blog.

Uber and price-surging

A $23m textbook by itself is an amusing anecdote, not an antitrust violation. But Mr Topkins’ case, along with a similar US prosecution of a UK online company, offers a glimpse of the digital economy’s emerging antitrust concerns.

The poster market should have been a poor candidate for price-fixing. The products are heterogeneous and thus hard to compare, sales were infrequent and keeping track of other sellers’ prices would have been time-consuming, says Salil Mehra, a professor at Temple University’s Beasley School of Law, who coined the term “robo-sellers.”

The ability to monitor other sellers’ prices through software, however, made a price-fixing agreement possible. A similar dynamic has been on display in a federal courtroom in New York, where Uber customer Spencer Meyer argued that the use of a price-setting algorithm by the service’s drivers, who are independent contractors, amounts to “classic anti-competitive behaviour”.

Mr Meyer argued in a proposed class-action lawsuit that Uber and its chief executive, Travis Kalanick, reaped artificially high profits through “surge pricing” — which raises fares at moments of high demand — generated by the Uber algorithm. “Kalanick’s pricing algorithm artificially manipulates supply and demand by imposing his surge pricing on drivers who would otherwise compete against one another on price,” Mr Meyer said in his court filing.

Ms McSweeny of the FTC has been at the forefront of those raising alarms over the algorithms’ impact. “It really, truly is a frontier now for us,” she says.

Mr Topkins, meanwhile, is scheduled to be sentenced on March 16. He will pay a fine of no more than $28,750 and may escape jail by co-operating with the investigation into the wall art market, according to his plea agreement.

Ms McSweeny predicts more such cases, and Mr Stucke and Mr Ezrachi say existing antitrust rules, developed to control the market power of 19th century titans, may be inadequate in the digital economy.

“What happens if the machines realise it is in their interest to systematically and quickly raise prices in a co-ordinated way without deviating?” Ms McSweeny asks. “What happens then?”

How algorithms can automate collusion

Machines make better merchants. At least that is the view of PROS Holdings, which promotes automated pricing systems, or “robo-sellers”, as a path to more efficient markets.

“Companies are throwing off so much data — digital exhaust, data exhaust,” says Patrick Schneidau, PROS’s chief marketing officer. “Algorithms are coming into their own because there is much more information for them to consume, and learn from, to help the business.”

The Houston company’s software, which executes a customer’s business strategy based on reams of market data, generates prices for corporations such as Virgin Atlantic, Siemens and ABB. PROS says its algorithms set more prices each day than Twitter sends tweets.

Automation makes possible pricing strategies that humans cannot execute. But some antitrust experts fear algorithms might just as easily rig markets as improve them.

If several competitors all adopt the same pricing technology — and react identically to changing market conditions — the result would be the same as if their executives had colluded on prices, according to Maurice Stucke and Ariel Ezrachi, authors of the book Virtual Competition.

Likewise, robo-sellers could eliminate the unpredictable human element. Most cartels collapse because one or more members seek short-term advantage. Imagine Opec, the oil cartel, being managed by artificial intelligence rather than 13 fallible oil ministers. “Big data” and computerised analytics would reduce errors in the market and make it easier to detect cheating, making unilateral price cuts less likely.

Stripped of human emotions like fear and greed, a cartel might persist indefinitely, says Salil Mehra, a Temple University law professor.