Three Algorithms in a Room

This article appears in the June 2024 issue of The American Prospect magazine. Subscribe here.

It’s uncanny how most of the modern ills plaguing our economy today can be traced back to airlines. The industry is a petri dish of contaminants, from deregulation to market consolidation to financialization, that metastasized into other sectors in the 20th century.

We can add to that list the rise of algorithmic pricing, an emerging economic configuration where all competitors in a market outsource their price-setting functions to the same third-party software, in a not-that-innocent plot to fix prices.

It all began with the new world of aviation that followed the Airline Deregulation Act, signed into law in 1978 by President Jimmy Carter. By gutting the Civil Aeronautics Board, which had tightly managed airlines, Carter did away with a slew of regulations, including price controls capping airfares.

What followed was a brief window of expanded competition as new airlines entered the market. More airlines led to reduced airfares as competitors tried to gain market share and take over particular routes. To prevent these price wars, the largest airlines got together to come up with a solution.

More from Luke Goldstein

The airlines reorganized an existing quasi-independent service they owned called the Airline Tariff Publishing Company (ATPCO), headquartered near Dulles Airport outside of Washington, D.C. By today’s standards, ATPCO wasn’t especially high-tech, but it essentially functioned as a clearinghouse to share information across the industry, helping airlines to set airfares. Weeks in advance, airlines would send ATPCO scheduled airfares along with detailed route information, seat numbers, and discount loyalty offers. None of this was public information. ATPCO in turn compiled this data and made it available to other airlines, so they could respond accordingly.

In the 1980s and 1990s, this information was digitized, making it available to member airlines on a moment’s notice. Today, ATPCO boasts of processing 18 million fare changes every day in its database, working with 447 member airlines around the world. A spokesperson for ATPCO notes that “ATPCO does not today and has never set any airline’s fares.”

But as fares rose, ATPCO caught the attention of the Department of Justice (DOJ), which filed a lawsuit in the early 1990s. The complaint alleged that by telegraphing competitors’ scheduled airfares, ATPCO allowed airlines to identify when their prices were comparatively low and preemptively raise them to meet a higher benchmark. The DOJ concluded that ATPCO was merely a craftier, more tacit form of traditional collusion outlawed by the Sherman Act. Instead of a handshake agreement in clandestine meetings, airlines just decided to create a third party to do it for them.

That case could have been the end of the road for ATPCO. But inexplicably, it never went to trial. The DOJ opted to settle with ATPCO, a decision that was paradigmatic of the weak, ham-fisted approach that defined antitrust enforcement at the time. In the settlement, ATPCO had to make some light modifications to slow down the pace at which airfare increases could be implemented. The theory was that these measures would provide travel agents, who can also access ATPCO, enough time to make more informed decisions about the best price options for consumers.

Airfares have only risen since that settlement, according to the St. Louis Federal Reserve. (ATPCO cited a study showing inflation-adjusted fares falling, released by the lead trade group for the airlines.)

IN HINDSIGHT, BY NOT ENFORCING MAJOR PENALTIES or banning ATPCO entirely, the DOJ effectively greenlit conduct that its own legal team deemed unlawful. Other actors across the economy took the hint and a proliferation of third-party price-fixing schemes sprung up, now seen in housing, agriculture, hospitality, and even health care.

These new pricing intermediaries are similar to ATPCO, but don’t just act as information exchanges between competitors. They actually set the prices for an entire industry by using machine-learning algorithms and artificial intelligence, which are programmed to maximize profits. To arrive at optimal prices, these software applications aggregate vast amounts of relevant market data, some of which is public and much of which is competitively sensitive information given to them by their clients.

Each algorithmic scheme has its own distinct features, but they all share the same underlying philosophy: Competing on price in an open market is a race to the bottom, so why not instead coordinate together to grow industry’s profits? In other words, it’s another version of the notorious Peter Thiel adage that “competition is for losers.”

These business arrangements are coming under fire from a new crop of antitrust regulators who are far more aggressive than their predecessors. Both the DOJ and the Federal Trade Commission (FTC) are intervening to help in numerous lawsuits making their way through courts that target algorithmic price-fixing. The DOJ has even returned to the arena with its own collusion case against an agricultural information hub called Agri Stats. The outcomes could have major ramifications for this field, and rectify the lax enforcement of ATPCO in the 1990s.

With market power, RealPage could extract higher rents than any one landlord acting unilaterally would be able to pull off.

Both enforcement agencies have issued statements firmly establishing that price-fixing via new technologies instead of human agents must not escape the antitrust laws.

“The idea that we don’t have a video of executives getting together and making their secret deal … does not automatically defeat a price-fixing claim in this new technological environment,” FTC chair Lina Khan told the Prospect. “We just need to be clear-eyed about that fact and update the application of the law to match the new technological and business realities.”

It’s not yet clear whether this view will hold sway in courts, with judges entrenched in more conventional understandings of collusion through a handshake agreement. Some plaintiff cases are already meeting snags for that reason. This intensifies the threat that judges may put an exceptionally high bar on evidence of tacit collusion, which would implicitly legalize price-fixing via algorithms.

THE DUELING SCHOOLS OF ANTITRUST don’t agree on much these days. The neo-Brandeisians at the helm of the Biden administration have overturned the consumer welfare standard in favor of broader anti-competitive considerations. The Chicago school theory of consumer welfare, which reigned supreme for decades, held that prices should be the sole indicator for assessing the harms posed by monopolies.

But even during the Chicago school’s reign, collusion was the one prong of enforcement that transcended these divisions, because cartels both subvert the very underpinnings of market competition and typically result in higher prices for consumers.

Richard Posner, one of the architects of the Chicago school, argued in an influential piece from 1968 that enforcers don’t always need direct transcripts from meetings where participants explicitly agree to collude. Regulators could use “circumstantial evidence” to infer tacit collusion if, for example, all the major players in an industry were raising prices in a coordinated fashion with no clear rationale like increases in input costs.

Posner, who was effectively pro-monopoly in every other respect, believed that going after aboveground oligopolies fixing prices should be the primary function of antitrust. His position proved to be self-defeating, since his fondness for mergers set the conditions for cartels to form among fewer players. Plus, as antitrust was de-emphasized, cuts to enforcement agency funding left meager resources to identify only the most clear-cut, winnable cases against price-fixing schemes.

Regulators might have been overly cautious, but enforcement against collusion generally remained constant, even as antitrust actions dwindled in every other area.

Oligopolies, however, found new ways to innovate their way out of this regulatory burden. They turned to a burgeoning field at the time called “revenue management.” Its leaders described the work as not only designed to benefit individual firms, but to grow profits for the entire industry writ large.

Information exchanges also became easier to facilitate in the 1990s, when regulators instituted “safe harbor” laws. These were essentially carve-outs for legalized information-sharing between competitors, as long as companies could meet certain criteria to justify it. Though specifically targeted for health care, it became common practice for other industries to exploit safe harbors for their own purposes. The DOJ withdrew these safe harbors last year as part of its crackdown.

ATPCO was born out of this new revenue management style of thinking. A former Alaska Airlines executive named Jeffrey Roper was a major adherent; his computer was confiscated during the DOJ’s investigation into ATPCO’s practices, and afterward, Roper left the country. He later returned in the early 2000s as a “principal scientist” for a new startup venture called RealPage.

ROPER’S NEW COMPANY HAD THE OBJECTIVE to revolutionize pricing in real estate by fixing the same problem that he helped the airlines address post-deregulation: to head off price wars by conspiring together. “A rising tide will lift all boats,” as one real estate executive whom RealPage worked with put it.

Roper took a very unsentimental approach to the business of property management, where he saw irrational human behavior driving inefficiencies. According to Roper, landlords had “too much empathy,” which prevented them from raising rent as high as they could. As Roper once put it, “If you have idiots undervaluing, it costs the whole system.”

He went about rectifying that by deposing the human agents who controlled pricing, and instead introducing algorithms that could make less emotional decisions.

In its own words, RealPage promises to “maximize profits” with the ability to “achieve … revenue lift between 3 percent to 7 percent,” even in economic downturns. There should be no doubt that RealPage does so by consistently pushing rent increases, according to testimony from clients in several recent lawsuits. RealPage, according to one lawsuit, told clients that the data they shared would never be used “to undercut RealPage’s higher prices—doing so for too long would mean losing access to RealPage.”

By virtually guaranteeing higher revenue, the service quickly spread across real estate markets and achieved extraordinary market power in several major cities.

In Seattle, for example, one of the most expensive cities for rent in the country, ten property managers control 70 percent of apartment units in one highly sought-after neighborhood. They all use RealPage.

The company became even more powerful after a major merger in 2017 with another software company called Lease Rent Options, which the Trump administration waved through after opening an investigation.

With market power, RealPage could extract higher rents than any one landlord acting unilaterally would be able to pull off in a competitive market. Without RealPage, customers could just look elsewhere for bargains, and landlords might risk not being able to fill vacancies. But when property managers are assured that virtually all surrounding residential buildings use the same service, then rent increases become more possible. Double-digit rent hikes among RealPage clients are common, according to an ongoing lawsuit.

In order to make these rent hikes stick, Roper and RealPage had to upend the conventional wisdom of the industry, called the “heads in beds” strategy. Usually, landlords want to maintain the highest occupancy rates possible, so they’re not wasting space. That means if landlords set their prices too high and can’t get any takers, they’ll usually drop the price to fill the apartment.

RealPage persuaded property managers that this was an arbitrary constraint. Since using their prices, landlords now report that they frequently operate well below 97 percent occupancy, holding out until they get tenants willing to match their higher rents.

Housing prices have soared across every major market RealPage has infiltrated. With government officials looking to bring down rising housing costs, RealPage is now coming under scrutiny, especially after a ProPublica investigation in 2022 into its business practices.

The company is facing a major class action lawsuit for price-fixing, and two others filed last year by attorneys general in Arizona and the District of Columbia. The lawsuits together paint a damning portrait of RealPage as a collusive enterprise, based on rental pricing evidence and scores of interviews with former employees and clients. The DOJ filed a statement of interest in the plaintiff lawsuit, and has its own criminal investigation into the company.

“It just doesn’t really take a leap of faith to see why it might be that rents keep going up in these dense metropolitan areas where RealPage operates,” said a senior official in the D.C. attorney general’s office. In the Washington-Arlington-Alexandria metropolitan area, over 90 percent of units in large buildings are priced using RealPage’s software, and rents have skyrocketed.

The D.C. lawsuit first explains that RealPage demanded its clients hand over highly competitive sensitive information, such as occupancy rates, rents charged for each unit and each floor plan, lease terms, amenities, move-in dates, and move-out dates.

RealPage’s clients give up this information because they know that their competitors will as well. In fact, many of them encouraged competitors to do so, as former employees allege in the lawsuit. That appears to be the “agreement” between competitors, which is a core part of any legal case for collusion.

This vast scale of information obtained by RealPage gives its algorithm a bird’s-eye view into the housing supply in a given market, which it can optimize meticulously and raise prices accordingly at each property. By instituting this regime, RealPage has managed to subvert an ordinarily competitive market for housing and centralize command over pricing.

ONE OF REALPAGE’S SUBSIDIARIES ACQUIRED in the 2017 merger, known as Rainmaker, is also facing legal troubles for facilitating a similar form of price-fixing in the hospitality sector, promising its clients that it can boost revenues by 15 percent.

One lawsuit was brought last year against Rainmaker for fixing prices in the Las Vegas metro area, where the software is used by virtually every major hotel and casino on the Strip.

The lawsuit shows that Rainmaker employs an identical strategy as RealPage to dissuade hotel operators from focusing on occupancy rates at the cost of profits. Casinos in Vegas have historically treated their hotel accommodations as a side business, just to get customers on the gaming floor where the big money is made. They’re willing to offer somewhat cheaper prices just to fill beds and get people in the door.

Rainmaker upended that. According to the lawsuit, their team told clients that “revenue managers must recognize the ultimate goal is not chasing after occupancy growth,” and urged clients to “avoid the infamous ‘race to the bottom’ when competition inevitably becomes fierce within a market.”

This dynamic has held beyond casino hotels. Indeed, overall hotel prices remain at or near their pandemic peak, at odds with other services even within the travel industry. A class action suit filed in February against a separate hotel data-sharing company called CoStar, which helped luxury hotels in 15 cities set prices, quotes the company telling clients that “total revenue grows higher when hotels understand the maximum amount a customer is willing to pay.”

Potentially the biggest example of algorithmic price-fixing was uncovered in the FTC’s lawsuit against Amazon.

But despite the evidence, the Las Vegas lawsuit against Rainmaker has seen the first setback for proponents of more active enforcement against tacit collusion by algorithms. Judge Miranda Du dismissed the case last October, citing a lack of evidence, even though the case had only reached the pleading stage, before discovery and depositions take place. Du, who claimed that she wasn’t an expert in the technology, was not convinced that there was proof of an agreement to collude, since each hotel separately accepted the pricing suggestions from Rainmaker.

RealPage is likely to use this defense as well, arguing that it merely makes “recommendations” for prices that are not binding and purely voluntary. The lawsuits say that this argument is a total ruse. Clients accept the RealPage recommendations over 80 percent of the time, and the company includes provisions in its contracts to ensure rent hikes. It heavily pushes adoption to new clients of an “auto-accept” feature that forces price increases automatically. Landlords who deviate from RealPage’s suggestions at too high a rate are subject to “discipline” and potentially even termination of their contract. “You should be compliant,” one RealPage training document uncovered in a lawsuit stated very clearly.

Rainmaker’s success with an early version of this defense is an ominous circumstance for enforcers and plaintiff lawyers, especially since the company advertises that clients accept its recommendations 90 percent of the time. But Rainmaker isn’t out of the woods. In Atlantic City, where it controls over 70 percent market share, another lawsuit is making its way through the courts. The pleading in that case appears to have a better chance of overcoming dismissal. Plaintiffs are also getting help from the DOJ and FTC, which filed a joint statement of interest.

A SIMILAR ARRANGEMENT HAS ALSO TAKEN HOLD in agriculture markets, with a third-party information exchange called Agri Stats.

Agri Stats has actually been around for a number of years, founded in the 1980s around the same time as ATPCO, when revenue management theory was on the rise.

When the service first emerged, most farmers didn’t think much of it, as Joe Maxwell remembers. He was a farmer in Missouri at the time, who later became the state’s lieutenant governor. The reason it didn’t set off alarm bells for him was in part that monopolization of the industry by dominant middlemen was only just getting under way. Once the Big Four meatpackers gained market share, the anti-competitive impact of Agri Stats would become clearer.

Farmers at the time were accustomed to using their own information service called DTN, which compiled and organized prices in a giant tome the size of a phone book. But DTN just assembled scores of publicly available data. Agriculture is a more coordinated sector than most because it’s prone to booms and busts from annual harvests, which skews production levels and prices for the entire market. For this reason, the USDA requires substantial reporting from every market participant and publishes that information.

Agri Stats is a different beast. It works with each of the Big Four meat processors and requires detailed information about their operations. That includes prices, production schedules, and practices; information about suppliers; and full costs, including how much they pay workers and what benefits are offered. The data that meatpackers hand over entails virtually their entire balance sheet.

Agri Stats then synthesizes that data into granular analysis, which it sends out to its clients in extensive reports. It also consults with individual meatpackers several times a year, giving advice on when to raise prices or cut back supply. As former president Blair Snyder described it, “it’s like Agri Stats is doing the accounting for the whole industry.”

The listings are technically semi-anonymized, but easily identifiable in such a consolidated market. Each of the participants is listed in Agri Stats reports, and industry participation covers at least 80 percent and usually over 90 percent of the various market segments. Several of the meatpackers, notably Tyson, only agreed to participate because they were promised by Agri Stats that their competitor JBS would spill their secrets as well.

Agri Stats, which had to pause its turkey and pork reports due to private antitrust lawsuits, even encourages processors to move in a coordinated direction. According to a Justice Department lawsuit filed last year, one executive at the pork giant Smithfield “summarized Agri Stats’ consulting advice in four words: ‘Just raise your price.’”

“It raises red flags when we see the sharing of competitively sensitive information among rivals, whether directly or through an intermediary,” said Doha Mekki, the principal deputy assistant attorney general at the DOJ’s Antitrust Division. “Corporations ordinarily work hard to safeguard that kind of information precisely because it helps them compete to win business.”

The department believes that Agri Stats’ price-fixing scheme is a quiet driver of high meat prices, which went up after the pandemic and have remained high.

The lawsuit argues that Agri Stats’ extensive reports allow meatpackers to easily identify when their products are priced low compared to their competitors and raise prices accordingly, which has occurred routinely. They can alter production levels to keep supply lower than they otherwise would, because they know they can get away with it. Competition for high rankings in Agri Stats’ pricing books is intense; according to the lawsuit, some processors give bonuses to staff for finishing at or near the top.

As Joe Maxwell explains, if this behavior were exchanged between executives in a secret meeting, there would be no question it’s collusion. “The meatpackers just got a third party to do it under another roof, it’s like they all hired the same economist,” said Maxwell.

Both the Department of Justice and the Federal Trade Commission are also monitoring how algorithmic pricing may be impacting the health care sector, where prices continue to increase. Ending the safe harbor laws for information exchanges in health care was part of those efforts.

One potential target is GoodRx, a common coordinator of reimbursement rates on behalf of pharmacy benefit managers, as an investigation by the investigative website The Capitol Forum revealed. Initially, GoodRx was just a consumer-focused service to help patients search for the cheapest prescription drugs. But on the back end, GoodRx teamed up with each of the largest PBMs to squeeze independent pharmacies. By striking deals with the same intermediary, the PBMs pool their claims through GoodRx, which then calculates the lowest possible reimbursements they have to pay out. Because of the market power PBMs have accrued, most independent pharmacists can’t refuse to work with them.

“If this isn’t algorithmic price fixing, I don’t know what is,” Luke Slindee, a pharmacy consultant at Myers and Stauffer who advocates for anti-monopoly policies, told The Capitol Forum.

Potentially the biggest example of algorithmic price-fixing was uncovered in the FTC’s lawsuit against Amazon. An algorithm called “Project Nessie” would test price increases to see if competitors would follow by hiking prices as well. If they did, the price would stay at its new, higher level. This added an additional $1 billion in revenue for Amazon, to say nothing of its competitors across the internet.

Project Nessie represents an even more arm’s-length version of unilateral price-fixing, entirely done by algorithm and the feedback loops it triggers. Amazon says it ended Project Nessie in 2019, but the technological capability to restart it again is as easy as flicking a switch.

TODAY, THERE’S A BIPARTISAN CONSENSUS emerging that enforcers need to establish a bright-line standard that algorithms can facilitate collusion, especially with the rise of more sophisticated models through artificial intelligence. It’s getting more attention in academia too. A recent paper from Harvard and Penn State researchers found that “[AI] pricing agents autonomously collude in oligopoly settings to the detriment of consumers.”

The current position of the Biden Justice Department on this front in many ways mirrors a speech delivered in 2017 by Maureen Ohlhausen, the former Republican FTC chair under Trump and a Chicago schooler in every respect. She stated very clearly that when it comes to collusion cases, “[e]verywhere the word ‘algorithm’ appears, please just insert the words ‘a guy named Bob.’ Is it OK for a guy named Bob to collect confidential price strategy information from all the participants in a market, and then tell everybody how they should price?” Her answer was affirmatively no.

The only remaining defenders of algorithmic pricing are economists who mostly operate from the theoretical premise of perfect market conditions. In a perfectly competitive market, individual firms using in-house pricing software might be incentivized to lower prices to gain market share. But that’s a figment of the economist’s imaginary models.

“They miss that market power skews the logic of competition,” said University of Tennessee law professor Maurice Stucke, who began investigating new pricing technologies early on.

The only real constraints on the Biden team’s actions are the limited resources they’re allocated to go after practices that appear unlawful. Then, there’s also the task of convincing the judiciary that price-fixing remotely by algorithm is indeed the same as price-fixing in person by humans. But enforcers believe this is an important issue to prioritize before pricing algorithms become completely embedded across the economy.

In previous eras, Doha Mekki explained, “a bad actor might have needed to work harder to facilitate an illegal information-sharing scheme. Today, instantaneous communication, the availability of data at scale, and increasingly widespread use of algorithms, machine learning, and artificial intelligence may have solved those speedbumps.”


POSTSCRIPT: Earlier this week, as MLex first reported, the FBI raided Atlanta-based corporate landlord Cortland Management, as part of an ongoing investigation into RealPages algorithmic price-fixing of real estate markets. Meanwhile, the Justice Department won the right to move forward with its price-fixing conspiracy lawsuit against Agri Stats. And in Las Vegas, Judge Miranda Du dismissed the case against Rainmaker over price-fixing at resort hotels, showing that defining collusion by algorithm as illegal still has a ways to go.

UPDATE: An earlier version of this story stated that Jeffrey Roper was the founder of ATPCO. That was incorrect, and we have updated the story to clarify that. The Prospect regrets the error. We have also added comments from ATPCO.

Read More

Luke Goldstein