One of the implications of the arrival of Big Data and artificial intelligence (AI) is that it is now possible for a firm to outsource its pricing decision to a third party, such as a software or data analytics company. A third party is likely to develop better pricing algorithms because it has more expertise and experience, access to more data, and stronger incentives to invest in their development (as the pricing algorithm can be licensed to many firms).
At the same time, concerns have been expressed that third party delegation could facilitate coordinated pricing between competitors. For example, the UK's Competition and Markets Authority (CMA) has warned: “If a sufficiently large proportion of an industry uses a single algorithm to set prices, this could result in a ... structure that may have the ability and incentive to increase prices” (CMA 2018). Similarly, the German Monopolies Commission (2018) has noted that it is possible a third party, when selling a pricing algorithm, “knows or accepts [it] could contribute to a collusive market outcome [and] it is even conceivable that [they] see such a contribution as an advantage, as it makes the algorithm more attractive for users...”.
There is some evidence, and claims of evidence, that these anticompetitive effects have occurred. For example, data analytics companies a2i Systems and Kalibrate developed pricing algorithms to assist retail gasoline companies in their pricing. After the wide adoption of their pricing software in Germany, a recent study finds evidence of anticompetitive effect (Assad et al. 2024).
In several legal cases in the US in the markets for apartments and hotels, plaintiffs claim that a third party and subscribers had an unlawful agreement to restrain competition (Harrington 2024a). In the case of apartments, a recent study offers evidence for both procompetitive and anticompetitive effects (Calder-Wang and Kim 2024).
On the policy front, the US Senate has introduced legislation to restrict the conduct of data analytics companies on the hope of preventing anticompetitive harm (US Senate 2024). It is clear that the challenge of third-party pricing algorithms for competition policy is here and now. An additional challenge for competition policy is algorithms learning to collude without human intent (Pastorello et al. 2019, Calvano et al. 2020).
Critical to the proper design of policy is understanding the incentives of a third party. For that purpose, I have engaged in research that addresses the following question: When does a third party have the incentive to design its pricing algorithm to produce supracompetitive prices? A cursory examination of the problem has suggested to some commentators (such as those quoted above) that a third party may want to design the pricing algorithm to maximise the collective profits of its subscribers – in effect, to act as a cartel manager. But the analysis in Harrington (2022) shows that this is not necessarily the case.
The setting in Harrington (2022) has a third-party developer supplying a pricing algorithm that allows firms to engage in more price discrimination – tailoring prices to narrow market segments – or more dynamic pricing – adjusting price to high-frequency demand shocks. An equilibrium is characterised for which the third party designs the pricing algorithm and sets a licensing fee to maximise its profit from selling it and firms optimally decide whether to adopt the pricing algorithm and pay the fee. Each firm independently makes an adoption decision.
The analysis shows that the third party does not programme a supracompetitive markup into its pricing algorithm. More specifically, average price is the same as when there is no third party and each firm is able to design its pricing algorithm. The third party optimally designs the pricing algorithm to maximise a firm’s willingness-to-pay (WTP) which is the profit it earns from adopting the algorithm minus the profit from not adopting it. While a higher markup raises the profit from adopting the algorithm (as all adopting firms are pricing higher), it also raises the profit from not adopting because a firm could profitably undercut the high prices set by rival adopting firms.
It is shown that raising the average markup above the competitive level raises the profit from not adopting more than from adopting, so it would lower a firm’s WTP and thus lower the profit that the third party can extract from adopting firms. Even when all firms are adopting the third party’s pricing algorithm, the average markup is not anticompetitive. (While the average price is the same, price is more sensitive to changes in demand when the pricing algorithm is designed by a third party. Thus, compared with when firms design their own pricing algorithms, a third-party’s pricing algorithm results in a higher price when demand is strong but a lower price when demand is weak.)
Key to the preceding analysis is that firms make independent adoption decisions. Suppose instead that there is an unlawful agreement between the third party who develops the pricing algorithm and competitors in a market who adopt it. This case is examined in Harrington (2024b). Now it is appropriate for the third party to design the pricing algorithm to maximise firms’ joint profits. The concern about a supracompetitive pricing algorithm making it attractive for a firm not to adopt it – so it can exploit adopting rival firms’ high prices – is no longer relevant when firms have an agreement. Consequently, the third party is indeed acting as a cartel manager.
By comparing the pricing algorithms when firms make coordinated and independent adoption decisions, some testable predictions emerge that can allow us to determine when there is an agreement based on price and adoption data. If adoptions are coordinated, then adopters’ prices will be increasing in the adoption rate (i.e. the fraction of firms that adopt) and, on average, adopters will price higher than non-adopters.
The reason is that a higher adoption rate is analogous to a more inclusive cartel and the optimal collusive price is higher when there are fewer firms outside the cartel (here, they are the non-adopting firms). In contrast, if firms’ adoption decisions are independent, then adopters’ prices do not change with the adoption rate and, on average, adopters and non-adopters price the same. Recall the average price is the same as under competition without a third party and thus does not depend on how widely adopted the pricing algorithm is.
This study shows how the properties of a third party's pricing algorithm depend on the manner in which adoption decisions are made by competitors. If firms independently decide whether to adopt the pricing algorithm, the average price of adopting firms is predicted to be independent of the number of firms that adopt, while if firms coordinate their adoption decisions then the average price of adopting firms is predicted to be increasing in the number of firms that adopt.
This finding delivers a test for determining whether there is an unlawful agreement between a third party and adopting firms. The evidence it provides could be used to screen possible cases for investigation or used in conjunction with other evidence to prove a violation of Article 101 of the Treaty on the Functioning of the European Union.
References
Assad, S, R Clark, D Ershov and L Xu (2024), “Algorithmic Pricing and Competition: Empirical Evidence from the German Retail Gasoline Market”, Journal of Political Economy 132: 723-71.
Calder-Wang, S, and G H Kim (2024), “Algorithmic Pricing in Multifamily Rentals: Efficiency Gains or Price Coordination?”, working paper.
Calvano, E, G Calzolari, V Denicolò, J E Harrington, Jr. and S Pastorello (2020), “Protecting Consumers from Collusive Prices Due to AI”, Science 370: 1040-42.
CMA – Competition and Markets Authority (2018), Pricing Algorithms: Economic working paper on the use of algorithms to facilitate collusion and personalised pricing, CMA94.
German Monopolies Commission (2018), “Algorithms and Collusion”, in Biennial Report.
Harrington, J E (2022), “The Effect of Outsourcing Pricing Algorithms on Market Competition”, Management Science 68: 6889-6906.
Harrington, J E (2024a), “The Challenges of Third-Party Pricing Algorithms for Competition Law”, working paper.
Harrington, J E (2024b), “An Economic Test for an Unlawful Agreement to Adopt a Third-Party’s Pricing Algorithm”, working paper.
Pastorello, S, G Calzolari, V Denicolò and E Calvano (2019), “Artificial Intelligence, Algorithmic Pricing, and Collusion”, VoxEU.org, 3 February
US Senate (2024), Preventing Algorithmic Collusion Act of 2024, 30 January.
