Daryl Lim explains that while there is some evidence that pricing algorithms facilitate collusion, there are reasons to be skeptical of their effectiveness. Lim advocates for compliance by design: firms should create algorithms that don’t collude on price, comply with reporting their algorithms transparently, and know that they will be held responsible for the actions the algorithm takes.

Introduction

Property managers across the United States gush about how an algorithm, dubbed YieldStar, boosts profits, even in a downturn. Greystar Real Estate Partners, the nation’s largest property management company with almost 700,000 managed units in 2022, found that its properties using YieldStar “outperformed their markets by 4.8%,” a significant premium above competitors.

YieldStar’s developer, RealPage, touts that its software uses data analytics to suggest daily prices for open units. To arrive at a recommended rent, YieldStar analyzes data from client apartment characteristics, like the number of bedrooms, how many more of a complex’s apartments will likely become available soon, and how full landlords want their buildings to be, and calculates price elasticity too. More controversially, it also crunches private information on nearby rival rents.

Letting units sit empty could be costly and nerve-wracking for leasing agents. Lowering rent to fill a vacancy make sense – until it doesn’t. According to one of the YieldStar’s developers, leasing agents had “too much empathy.” RealPage discourages bargaining with renters and recommends accepting a lower occupancy rate to raise rents. While apartment managers are free to reject YieldStar’s suggestions, they adopt as many as 90% of them. Given that despite replacing more renters, one firm’s revenue grew by 7.4%, it is not hard to see why.

YieldStar’s design and growing reach have raised questions about whether it allows the nation’s largest landlords to illegally coordinate pricing. US Sen. Sherrod Brown, the Ohio Democrat who chairs the Senate Committee on Banking, Housing, and Urban Affairs, recently wrote to FTC chair Lina Khan that YieldStar “raises serious concerns about collusion in the rental market” and “The FTC should review whether rent setting algorithms that analyze rent prices through competitors’ private data, such as YieldStar, violate antitrust laws.”

Pricing algorithms now proliferate across the online marketplace from select industries like airline tickets. Online sales have surpassed $4 trillion worldwide and continue to grow rapidly. As these practices become more widespread, it raises new problems for enforcers trying to ensure prices in the market are competitive. There is concern that algorithms may facilitate new forms of coordination that were unobserved or previously impossible.

A Guy Called ‘Bob’

Antitrust law focuses on coordination, or “meeting of the minds,” that unreasonably restrain competition with the purpose and effect of “raising, depressing, fixing, pegging, or stabilizing the price,” even when they do not set prices directly. The reason for looking for evidence of an agreement rather than collusion is practical: because of its rarity, the law has not regarded tacit collusion as a credible antitrust threat. Firms must stabilize collusion by monitoring adherence and punishing deviations.

Most markets are sufficiently heterogeneous that simultaneously coordinating a cartel is difficult. Policing perceived infractions is also difficult. Then Judge Breyer noted, “It is nearly impossible to devise a judicially enforceable remedy for “interdependent” pricing. How does one order a firm to set its prices without regard to the likely reactions of its competitors?”

Where firms used to collect information about other firms in the marketplace manually, algorithms now enable firms to respond to market conditions, including rivals’ prices rapidly. On one end of the spectrum, rivals might use algorithms to move in lockstep on price, product output, rigging bids, or dividing markets.

Since the core agreement is to restrain competition, existing laws can police these.

In 2015, the Department of Justice indicted David Topkins and other e-commerce art sellers for price-fix posters for sale on Amazon by using pricing algorithms to eliminate online price differences. By agreeing to fix prices for posters, they eliminated competition among themselves and monitored the effectiveness of their algorithms by spot-checking prices.

Algorithms may now enable tacit forms of collusion that were previously unlikely or unsustainable. They curate market information where data is abundant but difficult to process, automate the exchange of information and react to market shocks automatically, accurately predicting rival pricing behavior from sales numbers and demand statistics even if prices are unobservable.

Reducing noise allows firms to differentiate between cartelists defecting and legitimate responses to demand shifts, reducing the propensity for unintended price wars and stabilizing collusion. Firms can thus use them to coordinate parallel behavior directly as a hub or through parallel pricing strategies with separate algorithms operating in tandem.

By minimizing the number of humans needed to implement a collusive strategy, algorithms also minimize the risk that one may become a whistleblower. When they defect, the speed and ease of algorithmic pricing may also reduce firms’ benefit from cutting prices or defecting from collusive pricing as other cartel members can quickly detect and punish such defections.

RealPage claims YieldStar “uses aggregated market data from various sources in a legally compliant manner.” The algorithm works by prioritizing internal supply and demand dynamics above factors like competitors’ rents and, in doing so, minimizing collusion compared with manual pricing, which relies on phone surveys of competitor prices. RealPage feeds clients’ rent data into YieldStar to give landlords an aggregated, anonymous look at what rivals nearby are charging.

Authorities faced with the task of policing algorithmic collusion have two options. First, they could give every firm a free pass. This is a non-starter, as it would allow cartels to act with impunity behind algorithmic veils. Second, they could seek to hold firms accountable.

US Federal Trade Commission’s Former Acting Chair, Maureen Ohlhausen, noted that “Everywhere 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? If it isn’t ok for a guy named Bob to do it, then it probably isn’t ok for an algorithm to do it, either.”

Intervention Concerns

Intervention brings its own set of risks. Artificial Intelligence (AI) relies on reinforcement learning, experimenting, and observing under market conditions with multi-agent interactions and shifting strategies. Computer scientists have warned against overestimating the ability of algorithms to overcome complex coordination hurdles to form and sustain cartels. Some enforcers and academics are skeptical too.

The DeepMind AI developed an algorithm to play video games, but required over 40 years of training data and rapid and highly parallelized simulations that real market conditions will not allow. In a 2019 joint report, the German and French antitrust authorities questioned whether theoretical and experimental modeling results on algorithmic collusion prove their existence in real life. The UK Digital Competition Expert Panel found it hard to predict whether greater use of algorithms will lead to algorithmic collusion. Professor Thibault Schrepel attributed the overrepresentation of academic research on algorithmic collusion to publication bias.

There is also the problem of a lack of human intent to collude. The law prohibits agreements, but algorithms may reach decisions without humans intending to help them reach collusive outcomes. Black box algorithms may independently determine the best way to maximize profit. Neither their designers nor the firms deploying them intend to engage in even tacit collusion, and neither may be able to explain how algorithms arrive at their optimized prices despite knowing the inputs used to write the code and observable outputs. Moreover, personalized pricing and segmented markets make prices less transparent and increase market complexity, with the difficulty of establishing and sustaining collusive agreements using price signals.

Against this less than robust case for enforcement, widespread algorithmic connectivity promises static and dynamic gains. Algorithms can adjust prices based on demand conditions and costs, or those that allocate goods and services based on forecasted conditions allow them to function more efficiently than traditional pricing methods. These process improvements can lower barriers to market entry and exit, making it easier for prospective entrants to recognize the potential for high profits if prices are readily observable and employ “hit and run” tactics if prices fall too quickly. When firms can exit quickly, punishment becomes less of a deterrence. Increased transparency injects data into the market to improve the quality of existing and newly developed products, creating a ripple effect that leads to constant innovation in data-driven industries.

Algorithms also help firms optimize business processes, reduce costs and set prices to maximize profits legally through cost savings or demand inelasticity, both of which are legal. To RealPage, using YieldStar is the difference between working on a typewriter or a computer.

Of course, for tenants of RealPage’s clients, prices have not fallen but risen. But RealPage’s role in rent increases is hard to discern. Higher rents that RealPage charges may be more a function of capturing surpluses that human irrationality or cognitive oversight failed to recognize, and less illegal behavior. Inadequate new construction and the tight market for homebuyers are also plausible contributing factors. The company’s clients may also gravitate toward such markets because those areas will bear more rent hikes and offer an opportunity to make more money.

Mandating that algorithms ignore rival prices that run counter to rational market behavior, and opacity could limit useful information that consumers want. Firms may prefer personalized pricing and segment markets, i.e., price discrimination rather than collusion. Dynamic pricing increases profits by adjusting the price according to changing variables in supply and demand, as anyone using the Uber app during rush hour knows.

When a product or service has limited capacity, a firm can calibrate price swings while still being assured of the sale. As early as the 1980s, American Airlines used it to track route demand number of seats to earn $500 million more annually while offering gains in consumer welfare as leisure travelers who make reservations in advance receive lower prices than business travelers who make last-minute reservations.

Premature regulation chills innovation and forestalls economy-wide efficiency gains from using advanced pricing algorithms. Given antitrust enforcers’ limited information relative to a decentralized market, challenging market outcomes or structure replaces public choice with regulation that may be ill-placed to achieve its intended goals.

The Supreme Court eschews an approach that “requires antitrust courts to act as central planners, identifying the proper price, quantity, and other terms of dealing—a role for which they are ill-suited.” To paraphrase the court, the resulting rewards from the deft employment of algorithms can be “an important element of the free-market system.”

Prudence counsels giving time for theoretical research and empirical evidence to ripen. Professors Ariel Ezrachi and Maurice Stucke called for an “algorithmic collusion incubator” that would test the impact of speed and frequency of price changes and information transparency on market impact. In the meantime, the better approach is compliance by design. 

Compliance by Design

The law can mandate features that minimize the risk of collusion. “What businesses can and must do is to ensure antitrust compliance by design,” Executive Vice-President of A Europe Fit for the Digital Age and Competition, European Commission Margrethe Vestagersaid at the conference in Berlin. “That means pricing algorithms need to be built in a way that doesn’t allow them to collude.”

“What businesses need to know is that when they decide to use an automated system, they will be held responsible for what it does,” she continued. “So they had better know how that system works.”

Like Hart-Scott-Rodino reporting requirements, the law could require firms to declare pricing algorithms and their means of implementation. More data and algorithmic transparency will give policymakers and researchers better insight into how these algorithms function and whether they collude. As Vice-President Vestager stated, firms employing opaque algorithms could face a shift in the burden of proof to them to explain seemingly collusive behavior.

The law could also impose a duty on corporate and outside counsel to ensure a robust compliance program to review information exchange policies and practices. The goalposts have not changed. Former FTC Chair Ohaulason noted, “From my perspective, if the conduct was unlawful before, using an algorithm to effectuate it will not magically transform it into lawful behavior. Likewise, using algorithms that do not offend traditional antitrust norms is unlikely to create novel liability scenarios.”

Algorithmic collusion shares similarities with classic forms of collusion, differing in degree rather than kind, providing the basis to trace and predict the algorithmic collusion’s evolutionary trajectory. Algorithms like those used by Topkins provide enforcers with the full breadth of the collusive strategy in their code. Feigned procompetitive motives would shrivel in the light of that kind of evidence.

For tacit collusion cases, one possibility is to consider algorithmic pricing as a “plus factor,” regarded as evidence that parallel behavior should be treated as satisfying the agreement requirement. The plus factor should also be limited to rivals using similar algorithms even when better ones are available or when firms make it easier for rivals to observe their algorithms or data.

Firms with good reasons to switch to automated pricing, which may be cheaper and better than traditional pricing, can offer procompetitive justifications. In this way, doctrinal evolution has the advantage of putting the problem and solution in terms that a judge can understand and has legal precedent.

The Justice Department has already indicated that it expects firms to flag material anticompetitive risks and ensure the information exchange is reasonably necessary to achieve a legitimate business purpose. Additionally, it expects vertically integrated firms to filter competitively sensitive information that the company is receiving from customers or suppliers in one product or service market who are also rivals in another product or service market to ensure the existence of appropriate safeguards and firewalls.

Finally, antitrust enforcers and courts must employ AI to close the gap with the private sector’s arsenal. The fact that enforcement was the second-largest category of use cases identified in a US government survey is good news. Antitrust authorities have employed screening tools, typically using econometrics, for years. It is time to close the technological gap between enforcers and those they regulate by employing AI themselves. Over time, antitrust agencies can collect data on competitive and collusive episodes. With that data, it can use advanced statistical tools rather than solely relying on markers with benchmarks assessed by human judgment from past investigations. Implementing algorithms for screening could be applied at a large scale, and it costs to find instances where rivals tampered with pricing to stabilize price increases.

Conclusion

Antitrust illegality is based on risk assessment of possibilities, probabilities, and harm. As simpler algorithms continue to evolve to digest big data and deal with the complexities of real-world markets, the tools for collusion will doubtless become more widespread.

Yet there are good reasons to avoid broad brushed condemnation. Micromanaging price risks stymying the individual and aggregate responsiveness of the firm to changing market conditions. We want firms to react rationally and swiftly to changing market conditions. Without compelling evidence that algorithmic pricing does more harm than good, compliance by design is the wiser course forward.

Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.