Will pricing algorithms be the European Commission’s next antitrust target?

There has been considerable debate over the last year or so about the potential anti-competitive impacts of pricing algorithms. They could lead to discriminatory pricing, for example a company quoting different prices to different people based on an algorithmic analysis of their personal data, or cases of collusion, for example companies using algorithms to automatically fix prices. 

In a recent speech, Commissioner Vestager sounded a clear warning against the latter example: “companies can’t escape responsibility for collusion by hiding behind a computer program”. She also indicated that the use of pricing software forms part of the issues being investigated in the Commission’s new investigation into price-fixing in consumer electronics.

However, as pricing algorithms increase in complexity and sophistication, and their use becomes more prevalent, it will not be easy for competition authorities to establish where the use of such algorithms equates to actionable infringements of competition law.

How might pricing algorithms be used?

A new book by Professors Ariel Ezrachi and Maurice Stucke have identified four scenarios in which pricing algorithms may promote anti-competitive collusion (see here; also developed in more detail in their book Virtual Competition). 

The first is where firms collude as in a traditional cartel, but use computers to manage or implement the cartel more effectively, or to monitor compliance, for example by utilising real-time data analysis. Competition authorities have already investigated this kind of subject-matter – for example, the CMA issued an infringement decision last year against two companies that agreed to use algorithms to fix prices for the sale of posters and frames on Amazon (see here).

The second example is a hub-and-spoke scenario whereby one pricing algorithm may be used to determine prices charged by numerous users. Evaluating this sort of issue is a current challenge for competition authorities. Last year, in Eturas, the CJEU held that travel agents participating in a platform that implemented a discount cap could be liable if they knew about the anti-competitive agreement and failed to distance themselves from it (see here).  An ongoing case in the US (Meyer v Kalanick) is examining Uber’s ‘surge’ pricing algorithm, which increases the price of an Uber journey as demand increases.  The claimants allege that this constitutes an implied horizontal price-fixing agreement.  

The examples seen so far involve relatively straightforward cases of the use of algorithms as an aid or means to fix prices (although the Uber example arguably involves only unilateral conduct, rather than collusion).  However, Ezrachi and Stucke’s final two scenarios move into more uncertain territory – what if there is no express collusion by the companies? 

In the third scenario, each firm independently adopts an algorithm that continually monitors and adjusts prices according to market data. Although this can lead – effectively – to tacit collusion, particularly in oligopolistic markets (those with a small number of sellers), there is no agreement between companies that could form the basis of an investigation.  However, there can evidently be an anti-competitive effect: if an online retailer can track the prices used by another online retailer for common products, and immediately adjust its own prices to match any discounts, it can prevent the second online retailer from gaining a reputation for lower prices. The incentive for either retailer to lower its prices is removed.  On the other hand, examples from the analogue world suggests that this kind of market review can be used to ensure lower prices for consumers, at least for now (think supermarkets’ price match promises…).

In the fourth scenario, machine learning and the increasing sophistication of algorithms expand tacit collusion beyond oligopolistic markets, making it very difficult even to detect when it’s happening.

The latter two examples pose obvious difficulties for competition authorities. If they do consider such actions to be anti-competitive, how would they prove the requisite intention to co-ordinate prices?

How will competition authorities react?

As discussed above, competition authorities have already undertaken investigations against companies using pricing algorithms in collusion. We have previously noted the CMA’s interest in developing digital tools to aid its investigations (here). It seems certain that such tools will be necessary as these algorithms become more sophisticated and harder to detect.

The actions of non-dominant companies in using pricing algorithms whilst acting independently do not fall within the current competition law framework, even if such use ultimately results in higher prices for consumers. Commissioner Vestager has accepted that “what matters is how these algorithms are actually used”. This sensibly suggests that for now the Commission’s focus will remain on the more clear-cut cases of collusion. Anything else is arguably a matter for policy and regulation rather than enforcement by competition authorities.

However, Commissioner Vestager also stated that “pricing algorithms need to be built in a way that doesn’t allow them to collude”, suggesting that they needed to be designed in a way that will oblige them to reject offers of collusion. It is unclear whether this means Commissioner Vestager intends to target the use of pricing algorithms more generally, or simply to drive home that the competition rules apply equally where collusion is achieved algorithmically.

The fourth scenario, where machine learning algorithms tacitly collude to fix prices, does sound speculative. However, recent developments such as Carnegie Mellon’s Liberatus beating four of the world’s best professional poker players (here) and Google Deep Mind’s AlphaGo victory against Lee Sedol (here) indicate that it might not be too far from becoming reality in the near future.

Add comment