CYIL vol. 11 (2020)

CYIL 11 (2020) ARTIFICIAL INTELLIGENCE AND COMPETITION LAW… such a change does not seem likely and in particular, we do not find it necessary or even desirable. 71 Thus, an anticompetitive agreement among undertakings not communicating with one another and individually using price algorithms may only be found where the algorithms themselves are able to communicate. This may not be possible currently, but it is well imaginable in the future. Even today, the concept of price signalling should nonetheless be explored. Under the theory of competition law, it is well established that anticompetitive agreements do not need to be realised by an explicit communication, but by price signalling, e.g. by public announcements on future behaviour addressed not primarily to customers, but to competitors. As the European Commission puts it: Where a company makes a unilateral announcement that is also genuinely public, for example through a newspaper, this generally does not constitute a concerted practice within the meaning of Article 101 (1). However […] the possibility of finding a concerted practice cannot be excluded, for example in a situation where such an announcement was followed by public announcements by other competitors, not least because strategic responses of competitors to each other’s public announcements (which, to take one instance, might involve readjustments of their own earlier announcements to announcements made by competitors) could prove to be a strategy for reaching a common understanding about the terms of coordination. For example in the Container Shipping case, the Commission expressed the view that the regular early public announcements of intended future price increases by container liner shipping undertakings constituted a concerted practice and a by-object restriction; the announcements may have had the objective of communicating pricing intentions to competitors rather than informing customers about price developments and thus, the signalling might enable competitors to “test” a potential implementation of price increases without a risk of losing market share or triggering a price war. 72 Under normal conditions, price signalling might be costly. Whenever a firm increases the price to indicate an intention to collude, the signalling firm loses sales and profits if most competitors do not receive the signal or intentionally decide not to react, 73 Most of the real- world cases therefore presupposed a direct communication among the undertakings, who agreed on the signalling rules. These risks of price signalling might however be significantly reduced by the use of algorithms. As the OECD puts forward: Algorithms might reduce or even entirely eliminate the cost of signalling, by enabling companies to automatically set very fast iterative actions that cannot be exploited by consumers, but which can still be read by rivals possessing good analytical algorithms. There may be several ways to achieve this. For instance, firms may program snapshot price changes during the middle of the night, which won’t have any impact on sales but may be identified as a signal by rivals’ algorithms. 74 71 For similar conclusions, see EZRACHI, STUCKE ( op. cit. sub 68), p. 37.

72 Commission, Decision of 07 July 2016 AT.39850 ( Container Shipping ), par. 45 et seq . 73 HARRINGTON, J. E., ZHAO, W. Mathematical Social Sciences , 2012 (3), p. 277. 74 OECD Report on Algorithms , p. 30.

251