Key takeaways

• Companies are increasingly using algorithms to support the development of business strategy and prices, these are designed to predict an answer based on the available data. The more data that is available the more accurate and useful the algorithm becomes.
• A market where all firms unilaterally adopt their own pricing algorithm, accessing their rivals’ real-time pricing and adjusting to each other’s prices within seconds or even in real time can constitute a breeding ground for tacit collusion.
• Price fixing cartels are illegal and questions are starting to be asked by antitrust regulators in the EU and US as to whether the use of algorithms enable a new form of “algorithmic collusion”.
• it is important for antitrust authorities to distinguish between the use of algorithms to detect and punish an existing agreement (explicit collusion), or the use of algorithms that lead to coordinated prices without the parties’ knowledge (tacit collusion).
• In defending cases, companies will need to focus on providing credible explanations regarding transparency of the tools operations and the factors that it uses to make decisions, interpretability in terms of the weight it places on factors and how it responds to changes in these factors, e.g. the prices being set by competitors or providers of complimentary goods and provenance of the input data.
• Before implementing AI into business decision making, it is important to understand how and why decisions will be made, be able to unpick them and ensure that they are compliant with laws and regulations.

The value of data is increasingly being considered by antitrust authorities. This is both in terms of benefits to consumers, e.g. from more tailored adverts or product development but also due to the competitive advantages that it might confer to the data owner and potential for consumer harm that arises from potential explicit or tacit collusion.

Data is often referred to as the “new oil” and although that may convey the power and reach of data, it does not accurately reflect how versatile and reusable data is. Data is only powerful when it is put to use – simply holding data is not of value unless you can derive insights and business benefits from it. How this is done varies from industry to industry and from company to company and often from person to person within a company. Broadly speaking there are five key approaches to deriving value from data:

1. Algorithms are designed to implement a known test, i.e. a human is telling the computer what to do).
2. Algorithms are designed to learn from and mimic human knowledge and actions, referred to as supervised machine learning.
3. Algorithms are designed to predict an answer based on the data available without any human guidance, referred to as unsupervised machine learning).
4. Utilising Artificial Intelligence (AI) to improve processes and generate insights.
5. Visualisations are used to graphically represent any of the above and are designed to aid the interpretation of results rather than generate new results per se.

The more data that is available the more accurate and useful each of the above approaches become. This explains the drive to ‘own’ data, and it is true across all algorithms from speech recognition to automated translations to pricing models. Although it is worth noting that are potentially decreasing returns to more data, as pointed our recently by Varian, who noted that there’s a small statistical point that the accuracy with which you can measure things as they go up is the square root of the sample size. So there’s a kind of natural diminishing returns to scale just because of statistics: you have to have four times as big a sample to get twice as good an estimate.

Pricing algorithms predict the optimal, “profit maximizing”, price given various data inputs and are examples of approaches ii) to iv) above. This data would include market demand and supply conditions, and could include the prices charged by competitors for similar or complimentary goods. A benefit of pricing algorithms over humans is that they can quickly re-optimise prices to reflect changing information. This allows for the implementation of dynamic pricing, including surge pricing or real time prices. For example, Uber adjusts prices in real time to match demand for journeys and supply from drivers. This also enables suppliers to be more flexible and adjusts prices to be more personalised, for example reflecting past consumer behaviour. For example, shopping websites offer discount codes based on past purchasing and browsing history.

Algorithmic pricing is efficient and clearly yields a competitive advantage, which fewer companies will want to or can miss out on. With more and more companies adopting pricing algorithms and more sellers posting their current prices, more market data becomes accessible and market transparency increases. A market where all firms unilaterally adopt their own pricing algorithm, accessing their rivals’ real-time pricing and adjusting to each other’s prices within seconds or even in real time can constitute a breeding ground for tacit collusion. If one firm increases prices, its rivals’ systems will respond immediately. This normally happens without the risk that enough customers will realise and be able to move to other sellers. On the flip side, where a firm decreases its prices, competitors will also adjust theirs straightaway, so that, ultimately, there is no competitive gain and hence no incentive to offer discounts. The risk then arises that market players find a sustainable ‘supra-competitive’ price equilibrium (i.e. an algorithm-determined price which is higher than the price that would exist under competitive market conditions).

Price fixing cartels are illegal and questions are starting to be asked by antitrust regulators in the EU and US as to whether the use of algorithms enable a new form of “algorithmic collusion”. This consists in any form of anti-competitive agreement or coordination between competing firms that is facilitated or implemented through means of automated systems. Are digital algorithms the equivalent of the “golf course” agreement, being used to intentionally implement, monitor and police cartels? Or are they a legitimate method of price competition?

It has been argued that the use of pricing algorithms can change certain structural characteristics of the market. For example, they may provide greater transparency, higher frequency of interactions and faster exchange of information.

By making prices more formulaic, they become more predictable to the competition – and perhaps allow competitors to reach super-competitive equilibria more easily. It is also possible that pricing algorithms might learn to coordinate without any human having programmed them to do so and could, potentially, monitor and facilitate the punishment of deviations from collusion. Could this be considered tactic or explicit collusion? In the former, companies come together to agree prices or a pricing strategy and could knowingly use algorithms as a messenger, to monitor and enforce an anti-competitive agreement. In the latter, companies may find themselves colluding, unknowingly, due to operating practices. Hence, it is important for antitrust authorities to distinguish between the use of algorithms to detect and punish an existing agreement (explicit collusion), or the use of algorithms that lead to coordinated prices without the parties’ knowledge (tacit collusion).

Ezrachi & Stucke (2016)[1] identify three ways through which pricing algorithms may lead to tacit collusion:

1. Hub and Spoke - Where competitors use the same pricing algorithms, and these are used as a central ‘hub’ to coordinate prices. While competitors are not in direct contact, the effect is equivalent to horizontal collusion.
2. Predictable Agent - Where each firm unilaterally creates an algorithm that reacts to market events in a predictable way, allowing competitors to capture signals which could lead to a coordinated outcome.
3. AI or “Digital Eye” - Where pricing algorithms become sophisticated enough to self-learn and can anticipate events in the market even before they happen, potentially leading to a coordinated outcome.

Hub and Spoke occurs when online retailers using third party provider’s algorithms could find themselves facing cartel allegations without, in fact, having intended participation in a cartel. For example, various industry players might use the same third-party provider’s pricing algorithm to determine the market price and their pricing strategy. In this case, the use of the same pricing algorithm by competitors to monitor prices could lead to the (possibly unintentional) fixing of prices. The UK’s Competition and Markets Authority (CMA) considers the Hub and Spoke setting to be the most concerning because it simply requires firms to adopt the use of the same algorithm (CMA, 2018). This could mean that, if the online retailers purchased the third-party pricing algorithm knowing that their competitors had also done so, they could be considered to knowingly be engaging in collusive behaviour?

An example of where algorithms have been shown to be explicitly used to collude is the Poster Cartel case. David Topkins, the founder of Poster Revolution, was the first senior manager from an e-commerce business to be prosecuted under antitrust law by the US Department of Justice. It was found that there was the adoption of specific pricing algorithms that collected competitors’ pricing information, with the goal of coordinating changes to their pricing strategies for the sale of posters on Amazon Marketplace. In this case, it was demonstrated that the company intentionally developed the algorithm with the objective of colluding: the use of algorithms to help execute the cartel’s task had the same effect as a cartel executed by humans, with the computer facilitating the task that otherwise would have been carried out by humans. The European Commissioner for Competition, Margrethe Vestager, summed it up as “companies can’t escape responsibility by hiding behind a computer program.”

As artificial intelligence becomes more prevalent, it is possible that these algorithms could learn to coordinate prices without their developers being aware of this. It may be possible to command an algorithm not to fix prices, but what if through self-learning and experimenting with different solutions to optimise profits, the algorithm learns that pricing coordination gives the most profitable outcome? In this case, there is no anti-competitive intent by any individual and even the individuals who coded the algorithm may not be aware that the collusion is occurring.

Antitrust regulators have been considering this topic for several years. First, whilst pricing algorithms will likely lead to price convergence, this is a prediction of both competition and traditional price fixing collusion. Therefore, authorities will need to consider whether pricing algorithms are driving convergence towards higher or lower prices. Second, under European Union Article 101, the only time the EU looks explicitly at tacit conclusion is during merger control. So, to look at tacit conclusion, in addition to explicit collusion, may require a change in the law. However, legalities aside, the benefits of any potential remedies are not clear cut. For example, one way to limit potential collusion by algorithms is to prevent the creation of an excessively transparent market. However, this may not be an efficient solution since market transparency gives greater information to consumers to make better purchasing decisions. This may instead lead to antitrust authorities pursuing investigations into collusionary behaviour and the definition of explicit versus tactic collusion will become increasingly important, as will questions around agreement and liability. The general principle under EU law is that companies will be held liable for any anti-competitive practices of their employees, even if they can show that they have used their best efforts to prevent such behaviour. Vestager states that “what businesses need to know is that when they decide to use an automated system, they will be held responsible for what it does. So they had better know how that system works”.

It remains unclear as to what action antitrust regulators might take with regards to pricing algorithms and how this might differ between those where there was clear intent to collude behind the development of the algorithm versus those that “machine learnt”. However, as these types of algorithms are increasingly being used within business to improve and make decisions, it will not be long before the decisions made or influenced by AI become subject to regulatory scrutiny or a litigation. In defending these cases, companies will need to focus on providing credible explanations regarding transparency of the tools operations and the factors that it uses to make decisions, interpretability in terms of the weight it places on factors and how it responds to changes in these factors, e.g. the prices being set by competitors or providers of complimentary goods and provenance of the input data. Before implementing AI into business decision making, it is important to understand how and why decisions will be made, be able to unpick them and ensure that they are compliant with laws and regulations.

How can A&M can help?

A&M’s competition economists provide expert economic advice to corporates and their legal advisors before the EC, national competition authorities and national courts in all aspects of competition law. A&M draws upon the sectoral experience of the A&M Economics team but also the wider expertise from across A&M practices. A&M economists can assist clients navigate competition authorities’ investigations into agreements between firms and other conduct; and assess their impact on competition – making use of available data and modelling. This includes:

• Competition compliance: conducting economic tests to advise on compliance with competition law.
• Competition investigations: provide support and expert advice during all phases of a competition authority’s investigation for the parties involved or interested third parties.
• Market studies: support market participants through all stages of such studies from competition authorities, as well as providing analysis and support to the authorities.
• Calculating economic damages arising from competition investigations.

For further information on the areas in which A&M Economics can provide support, please contact one of the authors.

[1] Ezrachi, A, and Stucke, ME (2016), Virtual Competition: The Promise and Perils of the Algorithm-Driven Economy