On 8 October 2018, the UK's Competition & Markets Authority ("CMA") published an economic research paper on the conditions under which algorithmic pricing could cause harm to consumers. The CMA's analysis draws on existing literature, as well as primary research involving discussions with algorithm providers, other competition authorities and the CMA's own pilot tests.
What you need to know - key takeaways |
---|
- The growth of pricing software used by businesses, particularly in online markets, comes at a risk that such tools increase the risks of firms colluding in way which breaches competition law.
- On the other hand, the CMA's report suggests that extensive use of software to generate personalised pricing would make it significantly less likely that algorithms are able to achieve coordinated outcomes.
- As the application of competition law to algorithmic pricing emerges, businesses should ensure that automated and machine-learning pricing technologies are compliant with competition law.
|
The Relevance of algorithms to competition in markets
The focus of the CMA's research paper is on algorithms that use price as an input and/or determine price as an output. This includes price monitoring algorithms, price recommendation algorithms and price setting algorithms.
The CMA acknowledges many of the potential positive benefits of algorithms. For example, algorithms can reduce transaction costs for firms, reduce frictions in markets, and give consumers greater information on which to base their decisions. However, algorithms can also be used to facilitate explicit anti-competitive agreement
The CMA closed its first case involving a price-fixing cartel using automated re-pricing software in 2016, which related to the sale of posters using an online market place (December 2016 newsletter). In this regard, algorithms may make explicit collusion more stable as it:
- is easier to detect and respond to deviations;
- reduces the chance of errors or accidental deviations; and
- reduces "agency slack", i.e. reduces the risks of individuals in a firm undermining the cartel.
A more contentious area concerns the potential of algorithms to engage in tacit coordination (sometimes known as "conscious price parallelism").
Algorithms and tacit coordination
Competition literature identifies three main methods by which algorithms could result in tacit coordination:
- hub-and-spoke – whereby multiple competitors use the same algorithm;
- predictable agent – algorithms designed to give predictable results to external factors which may help sustain tacit coordination; and
- autonomous machines – self-learning algorithms that experiment to achieve the optimal pricing strategies to increase profits, including through tacit coordination.
The CMA identifies hub-and-spoke coordination as the most "immediate risk" presented by algorithms to competition. Tacit coordination in these circumstances may be achieved by competing undertakings using common third party algorithm suppliers, particularly where those third parties have access to competitor data, and, in turn, are incentivised to increase prices above the competitive level to maximise collective profits.
The CMA found that the predictable agent and autonomous machine models of coordination present a less immediate threat to consumers, but may materialise in markets that are already susceptible to coordination if such algorithms become sufficiently advanced and widespread.
The CMA gives a timely reminder of the traditional risk factors associated with tacit coordination, and how the algorithmic pricing and online markets could exacerbate these factors and result in more risk of harm to consumers. In particular, the CMA notes that:
- algorithms can immediately collect information about competitors, meaning that coordination can occur in less concentrated markets, and deviation and punishment strategies can be detected and implemented faster;
- increased availability of data, especially online, results in increased market transparency and enables algorithms to scrape data from websites to detect price deviations and adopt simpler pricing behaviours;
- the capability of implementing price changes automatically increases frequency of interaction and price setting, thereby enabling immediate market feedback and discouraging the short-term benefits of price wars; and
- the availability of online markets creates opportunities for customers with low buyer power to form buying groups, increasing their buying power.
Personalised pricing
A further part of the research paper is concerned with the role of algorithms in personalised pricing. The CMA conducted its own testing on the use of algorithms to personalise pricing using an "internet lab". Although the CMA found evidence of algorithms used to personalise search results, advertising and discounts, in practice limited evidence was found of algorithms used to personalise pricing.
However, the CMA noted that use of such tools can be expected in the future, and are most likely to harm consumers in circumstances where:
- markets have limited competitive constraints;
- consumers can be divided into small groups (based on their willingness to pay);
- the cost to implementing personalisation is significant (and must be recouped through higher retail prices); and
- consumers decide to withdraw demand due to mistrust in the market.
An interesting discussion in the research paper concerns interaction between personalised pricing and tacit coordination. In particular, the CMA's preliminary view is that extensive use of personalised pricing would make it significantly less likely that algorithms are able to achieve coordinated outcomes (because competitors would be unable to observe and detect deviations from the collusive price, which is individualised). Conversely, factors such as transparency, which facilitate tacit coordination, make it harder to engage in personalised pricing (because consumers are able to compare prices and find a better deal). In other words, the two concepts are compatible in theory, but unlikely to co-exist in practice.
Conclusion
Competition authorities are monitoring the use of pricing algorithms and their effect on competition and consumers. The CMA's research paper suggests the need for further research in this area, including whether there are some types of algorithms that should be presumed to be anti-competitive; a likely area of contentious debate in competition policy.
For the time being, existing competition tools provide regulators with sufficient tools to address the potential competitive harm brought by pricing algorithms, as the CMA's Posters case in 2016 demonstrates. Moreover, the use of third party pricing software should also receive careful attention by businesses as regards how competitor data is used to avoid risks of hub-and-spoke price coordination.
With thanks to Emile Abdul-Wahab and Tom Punton of Ashurst for their contribution.