AI and Algorithmic Collusion
Pricing and Revenue Management (RM) has always been highly analytical in nature. In a couple of my early papers [1] [2] , I discussed how RM had the potential to serve as a strategic focus and source of competitive advantage for firms. In the two decades since, RM algorithms have become fairly well-formed and standardized, with countless vendors now providing RM solutions. As RM algorithms became more standardized, they themselves became less strategic and, as a result, shifted from being developed in-house to being offered by third-party developers.
An unintended consequence of third-party RM services is the potential for the consolidation of algorithm types — i.e., multiple firms using a common algorithm. Regulators have been paying attention to the anti-competitive impacts of pricing algorithms for quite some time. Meeting notes from the 2017 OECD raised concerns by the U.S. Department of Justice (DOJ) and the Federal Trade Commission (FTC) and explored how technological advances, particularly pricing algorithms, impact competition and collusion risks. They explain that while algorithms can enhance competition by allowing rapid price adjustments, they also introduce new risks for anti-competitive collusion. The agencies distinguish lawful independent pricing from illegal coordinated behavior facilitated by algorithms and warn that agreements to use shared pricing algorithms can constitute unlawful collusion, even without direct communication between competitors. However, mere parallel use of similar algorithms without agreement typically does not violate U.S. antitrust laws.
More recently, following a series of civil cases, the FTC and DOJ released a revised view , specifically stating: “Companies across the economy are increasingly using algorithms to determine their prices. When a small group of algorithm providers can influence a major segment of a market, competitors are better able to use the algorithm provider to facilitate collusion. This risk is even greater as markets have become more concentrated across a wide range of industries. Algorithms that recommend prices to numerous competing hotels make it harder for travelers to comparison-shop for the best rate.”
In their detailed statement, they further elaborate that there does not need to be an explicit agreement between firms, as tacit collusion may still persist, and that even the existence of algorithms providing only recommendations or starting prices (with user override) is insufficient to defend against allegations of collusion.
These statements have profound effects on the practice of RM, as seen in the numerous civil cases currently making their way through the courts, alleging elevated prices by firms as a result of algorithm use.
At present, I am working on a series of projects that dig deeper into the potential for algorithm-driven collusion.
[1] CK Anderson with PC Bell and SP Kaiser. 2003. Strategic operations research and the Edelman Prize finalist applications 1989–1998. Operations Research 51 (1), 17-31
[2]CK Anderson with PC Bell. 2002. In search of strategic operations research/management science Interfaces 32 (2), 28-40
