On 8 October 2018, the UK Competition and Markets Authority (“CMA”) published a Working Paper on the ‘use of pricing algorithms to facilitate collusion and personalized pricing’ (the “Paper”). It follows a number of other initiatives from competition authorities regarding algorithms, including the recent German Monopolies Commission’s proposals regarding pricing algorithms, which was the subject of a Covington Competition Blog post. The CMA’s analysis reflects input from algorithm providers, other competition authorities, and the results of the CMA’s findings from pilot tests. The Paper is economic rather than legal in focus, and assesses the extent to which various algorithm models have the potential to affect competition.

Competition Benefits and Concerns Related to Algorithms

The Paper notes the potential efficiencies generated by pricing algorithms, including reducing transaction costs and improving decision-making of consumers by giving them access to a wealth of information. However, the Paper goes on to discuss the potential for algorithms to harm consumers through explicit or tacit coordination.

Explicit Coordination Concerns

The CMA considers the potential use of algorithms to facilitate explicit agreements. According to the Paper, pricing algorithms could facilitate cartelists increasing the stability of their collusive practice. The CMA explains that, because of the volume of available data and speed of processing, an algorithm would make it easier, quicker and less costly to detect and respond to deviations. High quality data would also reduce the chance of errors. Finally, algorithms can also reduce the incentive of cartelists’ employees to undermine the existing cartel.

Tacit Coordination Concerns

The Paper also considers three possible means by which a pricing algorithm could facilitate tacit coordination. First, it explores tacit collusion in a Hub-and-Spoke setting where several firms employ the same algorithm to determine their pricing behaviour. Second, it discusses the Predictable Agent, where firms unilaterally develop algorithms that are set up to monitor coordination, follow price leadership and punish deviations in order to reach a collusive outcome. Third, it addresses a more complex type of algorithm, namely the Autonomous Machine, which is typically programmed to maximise profit without human intervention, such that it could decide that collusion is the most optimal strategy.

Of these theories of harm, the Paper considers the Hub-and-Spoke scenario to “present the most immediate risk”. It only requires adoption of the same algorithm by several firms. The CMA views the Predictable Agent and Autonomous Machine scenarios as longer-term risks, since they require the wide-spread adoption of technologically sophisticated algorithms. Further, the CMA takes the view that these two models are unlikely to fall within Article 101 TFEU because they would be “generating this coordination themselves”.

The Paper identifies certain risk-factors inherent to algorithms that might facilitate coordination. First, the time-horizon of an algorithm, i.e. whether it places more weight on short-term or long-term profits, can determine how likely it is to facilitate coordination. Second, coordination is more likely where several firms use the same algorithm. Third, the CMA considers the risk of coordination to be higher when algorithms use data from multiple competitors, for example through intermediaries.

Tacit Coordination and Personalized Pricing

Algorithms can also enable companies to offer different prices to different consumers depending on the information they hold about them (“personalized pricing”). Interestingly, the Paper considers that tacit coordination and personalized pricing are very unlikely to occur in the same market. The opaque nature of personalized pricing makes it very difficult to reach a common understanding between companies and more difficult to detect deviations, making collusion more difficult.

Future Areas of Research

Finally, the Paper identifies various future areas of research, including the potential use of auditing algorithms to assess how and why firms are using algorithms. The CMA also recommends the adoption of counter-measures such as joint purchasing, or requesting secret offers (for example through ‘shopbots’) to undermine collusion.

The Paper also suggests identifying certain algorithmic decision making rules that could be anticompetitive. However, it goes on to note that such an approach might be too interventionist, given that an objective to maximize profit through algorithms could be consistent with both pro- and an anti-competitive conduct.

Conclusion

The Paper provides an insightful analysis of algorithms and their economics, including their potential practical impact, contributing to the ongoing debates on algorithms that had otherwise been more theoretical to date. The CMA notes that the Paper could help “provide a means of preliminary prioritisation for considering future complaints or calls for intervention” regarding algorithms.