By: Sahana Thamizan.
Suppose you go to rent an apartment and notice that several buildings nearby have almost
identical prices. You suspect that the landlords must have coordinated to raise their prices at the
same time. But what if they simply learned that they could make more money by doing so?
Companies are increasingly using artificial intelligence to determine the prices of products and
services. By studying demand, competitors’ prices, and other factors, these algorithms can adjust
prices faster than any human could. While this practice may make pricing more efficient, it also
creates the possibility of a new form of anti-competitive behavior. Two algorithms may
repeatedly adjust their prices and learn that they can earn more by avoiding competition with one
another. This phenomenon is known as tacit algorithmic collusion, in which algorithms can learn
coordinated behavior without directly communicating. The U.S. Department of Justice’s case
against RealPage demonstrates why this issue has practical significance. As businesses
increasingly rely on artificial intelligence, regulators may need to examine not only whether
companies communicated with one another, but also whether their algorithms learned patterns
that reduce competition.
When humans set prices, they usually consider factors such as costs, demand, and competitors’
prices before making a decision. An algorithm, however, can continuously monitor these factors
and adjust prices automatically. The OECD has noted that pricing algorithms can benefit markets
by allowing companies to respond to changes in demand more quickly, while also creating
potential risks for competition (OECD, 2023). For example, suppose two companies sell
identical products. If one company raises its price and the other notices that customers continue
purchasing its product, the second company may also raise its price. If both companies discover
that maintaining higher prices leads to greater profits, neither has much incentive to lower its
price again.
The important question is how an algorithm can learn this behavior without being explicitly
programmed to do so. Reinforcement-learning algorithms learn by trying different actions and
receiving a numerical reward based on how well those actions achieve their objective. In a
pricing system, for example, earning greater profit could produce a higher reward. Over many
interactions, the algorithm can identify which pricing decisions tend to produce better results. If
it notices that raising prices increases profits and that a competitor responds by also raising its
price, it may learn that maintaining higher prices is beneficial. Eventually, two algorithms can
recognize and respond to each other’s patterns without ever directly communicating.
Calvano et al. demonstrated this possibility through a simulation in which pricing algorithms
learned to behave less competitively (2020). Their algorithms were able to raise prices and
respond to one another’s price changes, including temporarily lowering a price and later
increasing it again. Importantly, the algorithms were not explicitly instructed to collude. Instead,
they learned that certain patterns of behavior produced higher profits. This distinction creates a
difficult regulatory problem: similar prices alone do not prove that companies have colluded, but
algorithms may still learn to maintain higher prices when they recognize that competitors are
likely to respond in the same way. This risk may be particularly significant when companies have
similar costs and sell similar products, such as gas stations located near one another.
The concern is not simply that companies may earn more money; it is that consumers may bear
the cost of reduced competition. When companies face less pressure to compete, they have less
incentive to lower prices. If algorithms repeatedly learn that maintaining higher prices is more
profitable than competing, consumers may have to pay more for the same products or services.
Antitrust law generally focuses on evidence of agreements or coordinated conduct, but
automatically generated pricing decisions can make it difficult to determine whether companies
intentionally reduced competition. The OECD has noted that algorithms can complicate
competition law, including questions about what constitutes an explicit agreement, and that these
situations must be evaluated on a case-by-case basis (OECD, 2017). The challenge becomes
even greater when companies themselves may not have directly instructed their algorithms to
coordinate. A company could simply deploy an algorithm designed to maximize profit, only for
that algorithm to discover that responding to competitors in a certain way produces better results.
While similar prices do not necessarily indicate tacit collusion, the potential becomes more
concerning when pricing algorithms repeatedly respond to one another. This distinction can be
seen by comparing Calvano et al.’s simulation with the U.S. Department of Justice’s case against
RealPage. In Calvano et al.’s simulation, the algorithms learned to behave less competitively
through repeated interactions, without directly communicating. In contrast, the DOJ alleges that
RealPage’s service used information from competing landlords to generate rental-price
recommendations. The case therefore involves a different mechanism: rather than algorithms
independently learning to coordinate, the government alleges that competing landlords shared
information through the pricing system. Nevertheless, the RealPage case demonstrates why
algorithmic pricing has consequences beyond theoretical simulations. Because the technology is
being used to determine housing prices, its effects can directly influence consumers’ living
expenses.
These examples reveal the difficult position of antitrust regulators. They must determine when
algorithmic behavior represents ordinary competition and when it contributes to an unlawful
reduction in competition. Traditional enforcement can identify explicit communication between
companies, but algorithms can create coordinated outcomes without the same obvious evidence.
The OECD has recognized that algorithms can create difficulties in determining whether
companies have explicitly agreed to coordinate or whether algorithms have learned potentially
collusive behavior on their own (2017). Regulators therefore face a difficult balance: they must
prevent anti-competitive conduct without treating every instance of similar pricing as evidence of
wrongdoing. As algorithms become more sophisticated, regulators may need methods for examining not only the final prices they produce, but also the patterns through which those prices
develop.
There is no reason to prevent companies from using algorithms altogether. Automated pricing
can make markets more responsive and allow businesses to react to changing demand efficiently.
Instead, regulators could establish standards that make potentially anti-competitive algorithmic
behavior easier to identify. For example, companies using large-scale pricing algorithms could be
required to document what data their systems use, what objectives they optimize for, and how
they respond to competitors’ prices. Regulators could then examine repeated pricing patterns
rather than isolated price similarities. A single day of similar prices may be coincidental, but
repeated cycles in which algorithms raise prices, respond to competitors, and maintain those
prices could provide stronger evidence that competition is being reduced. Such monitoring
would allow regulators to preserve the benefits of algorithmic pricing while addressing situations
in which automated systems contribute to anti-competitive outcomes.
Artificial intelligence is redefining one of the concepts that once seemed most basic: the price of
a good. Computers can now use thousands of factors to determine how much a product should
cost and can learn from the behavior of competing algorithms. Research has shown that
reinforcement-learning algorithms can learn to maintain prices above competitive levels without
directly communicating, while the RealPage case demonstrates how algorithmic pricing can also
affect real-world markets. The challenge for regulators is therefore not to prevent businesses
from using artificial intelligence, but to distinguish between algorithms that improve market
efficiency and those that contribute to reduced competition. As these systems become more
complex, antitrust enforcement will need to consider not only what companies explicitly agree to
do, but also what their algorithms are capable of learning on their own.
References
Calvano, E., Calzolari, G., Denicolò, V., & Pastorello, S. (2020). Artificial
intelligence, algorithmic pricing, and collusion. American Economic Review, 110(10),
3267–3297. https://doi.org/10.1257/aer.20190623
Organisation for Economic Co-operation and Development. (2017). Algorithms and
collusion: Competition policy in the digital age. OECD Publishing.
https://doi.org/10.1787/258dcb14-en
Organisation for Economic Co-operation and Development. (2023). Algorithmic
competition. OECD Publishing. https://doi.org/10.1787/cb3b2075-en
Organisation for Economic Co-operation and Development. (2025). Algorithmic pricing
and competition in G7 jurisdictions: Emerging trends and responses. OECD Publishing.
https://doi.org/10.1787/f36dacf8-en
U.S. Department of Justice. (2024, August 23). Justice Department sues RealPage for
algorithmic pricing scheme that harms millions of American renters. Office of Public Affairs.
https://www.justice.gov/archives/opa/pr/justice-department-sues-realpage-algorithmic-pricing-s
c heme-harms-millions-american-renters
U.S. Department of Justice. (2025, January 7). Justice Department sues six large landlords for
algorithmic pricing scheme that harms millions of American renters. Office of Public Affairs.
https://www.justice.gov/archives/opa/pr/justice-department-sues-six-large-landlords-algorithmic
pricing-scheme-harms-millions
Leave a Reply