Information can be used by competitors to collude or to compete, and the challenge for competition law is to spot the differences. Signalling and any other type of informational exchange outside the scope of cartels are an emanation of tacit collusion. Tacit collusion, however, is generally considered unobjectionable, because firms are deemed to have the right to adapt intelligently to their rivals’ conduct. The law puts different labels on what is ultimately the same economic phenomenon, that is, conduct that leads to supra-competitive outcomes.
The traditional legal approach for distinguishing between illicit collusion and legitimate oligopoly conduct is to rely on criteria that relate to the means and form of how rivals interact, such as elements of “practical cooperation” or findings of anticompetitive intent. This article, available here, contends that, outside the scope of classic cartel agreements, it is not possible to properly distinguish between illicit collusion and legitimate independent conduct by relying on proxies such as elements of practical cooperation, other so-called “plus factors”, or anticompetitive intent. Instead, the author suggests scrutinising singular elements of communication, that is, informational signals, within an existing oligopolistic setting for their propensity to create consumer harm.
Section II includes an analysis of the potential risks to unimpeded competition associated with the evolution of algorithms.
Where a supra-competitive equilibrium is sustained without a cartel agreement, this is often referred to as tacit collusion or conscious parallelism. Despite tacit collusion not involving any explicit agreement by definition, this type of collusion can be furthered by communication, which can take many forms.
The extent to which firms can communicate is limited by their ability to receive, process, and release information in terms of quantity, speed, and precision. That is where algorithms and machine learning come into play. These developments can pave the way for collusive equilibria as they reduce the need for time-consuming decisions or constant monitoring by natural persons. Being able to process larger amounts of information more quickly than human beings, computer algorithms might well result in more efficient exchange of information, and thus create the framework for collusion mechanisms that otherwise would not have existed. Several authors have expressed concerns over the capability of the existing legal framework in Europe and the United States to remedy the risks flowing from this effectively.
Section III evaluates the existing criteria for distinguishing illicit collusion from legitimate communication.
While tacit collusion is often deemed a “normal” outcome of an oligopolistic market structure, a concerted practice is said to amount to an “artificial” reduction of competition. How to distinguish concerted practices from the reduction of competition produced by ‘normal’ tacit collusion is a challenging issue, for which EU jurisprudence and legal scholarship have suggested a variety of qualitative criteria. These criteria include the existence of ‘practical cooperation’ (Dyestuffs, Irish Sugar), ‘knowingly’ substituting competition for cooperation, artificially reducing competition, exchanging information without a business justification and the absence of alternative explanations for a collusive outcome. Ultimately, these criteria conceive of a concerted practice as a type of interaction that can be captured as an ontological category in its own right, and which can be distinguished from mere tacit collusion in qualitative terms.
The paper also provides an overview of parallel discussions under Section 1 of the Sherman Act. From the 1950s, it was asserted in the US that tacit collusion should be sufficient to establish the requisite agreement under Section 1 of the Sherman Act without more so long as the firms involved were aware of their interdependencies within the oligopoly. Others vigorously opposed this view, arguing that mere conscious parallelism was individually rational and unavoidable, and did not amount to an agreement. The U.S. Supreme Court eventually decided that consciously parallel behaviour was beyond the reach of Section 1.
Since then, debate in the United States has focused largely on the evidentiary requirements for proof of an agreement under Section 1 in cases where no explicit communication between firms can be shown. One strand of case law and commentary focused on so-called plus factors from which collusion could be inferred – e.g. prolonged parallel conduct; the existence of a rational motive to act in concert and perhaps against the individual best interest of each participant; the defendants’ participation in past collusion cases; evidence that the firms had the opportunity to communicate or actually did so; and so forth. However, if each “plus factor” is as consistent with unilateral conduct as with conspiracy or agreement, then the conduct, without more, cannot be said to constitute an agreement. Another stand of case law focuses on so-called “facilitating practices” or “facilitating devices”, whereby oligopolists n engage in avoidable conduct that reduces uncertainty about its and its rival’s future conduct.
Section IV then undertakes a critical assessment of EU criteria.
The European formula for identifying “practical cooperation” are so broad that they provide limited exegetical value. Further, they have no economic meaning – there is no categorical difference in the economic effects of explicit collusion and tacit collusion.
Given these difficulties, some authors have suggested reliance on subjective criteria. However, in addition to evidentiary difficulties, it is unclear what degree of knowledge is supposed to be necessary to qualify the conduct as a concerted practice. To those who promote an ‘artificiality’ test, the author answers by noting that this is not a category that can be explained by the economics of collusion. When an agreement is absent, the release of informational signals is as natural a phenomenon as tacit collusion, precluding the distinction between normal and artificial parallelism. The business justification paradigm also raises problems, in particular the concept is not sufficiently precise to serve as a legal criterion to distinguish between lawful and unlawful conduct.
Section V argues for putting greater weight on effects analysis.
The qualitative criteria discussed above, which relate either to the process of communication or to the firms’ inner spheres, are insufficient to distinguish between illicit collusion and lawful interdependent conduct. Given this, two solutions are possible.
First, one can prohibit tacit collusion in its entirety, as was defended by some in the US. This view, however, faces objections. It is widely agreed that parallel conduct is not per se anticompetitive, and for good reason. Where a mechanism of detection and retaliation exists, sticking to supra-competitively high prices can be individually rational for any of the rivals involved. It would be unconvincing to interpret the antitrust laws in a way that would oblige firms to act irrationally. Even if it could be said that it would be (more) rational for firms to avoid sanctions, one would still need to identify conduct which can avoid liability. The problem with prohibiting tacit collusion as such would be that firms in oligopolistic markets “cannot avoid knowing their prices are interdependent when they set their prices, so that it would be hard to define any prohibition in a way that tells firms how to behave.” Prohibiting oligopoly conduct as such would involve the impossible task of defining a ‘sufficiently non-oligopolistic conduct’ to comply with the law.
Two, one can distinguish lawful from unlawful conduct without referring to tacit collusion. Instead, one could simply prohibit certain conduct so long as these are clearly defined by reference to an external criterion. This article suggests that this criterion should be whether the conduct is conducive to consumer harm. A concerted practice should therefore be any informational signals of two or more firms which have, as their object or effect, a restriction of competition, in that they harm consumers through a collusive equilibrium without creating offsetting efficiencies. Instead of comparing the equilibrium price with a hypothetical competitive outcome, each element of communication of the firms should be checked for its propensity to cause a consumer harm. If a certain element of communication is found to create harm, then the firms are obliged to refrain from this type of informational signal completely.
Section VI analyses the burden of proof.
The approach suggested in this piece comes at the expense of an economic effects analysis that can be challenging. This should not, however, be considered a reason to reject it. The intricacies of an effects analysis can be accounted for by defining the evidentiary burdens in an appropriate way. The default rule for dealing with this problem is to distribute the burden of proof in an effective but fair way that calibrates for the likelihood of consumer harm. If measured against these standards, the release of informational signals, by computers as well as by human beings, can amount restrict competition by object or by effect depending on the likelihood of harm to consumers.
Economic theory suggests that there is an “ambiguity of welfare effects” in signalling. A private exchange of information between rivals is oftentimes perceived as more clearly posing a risk to competition, because this type of communication is deemed to be: “significantly less costly and more effective than public announcements” if the purpose of the firms involved is to provide a new focal point for pricing. Private communication between competitors is also deemed unlikely to produce efficiency justifications. Public price announcements, on the other hand, are often associated with the creation of “significant benefits to consumers”. Between a strictly private exchange of information and public signalling, however, are varying grades of exposure or exclusivity, which can make anticompetitive effects more or less likely.
This is an impressive and thoughtful paper. It provides a complete overview of the issues raised by oligopoly pricing and tacit collusion for competition law and policy, and a thorough criticism of the mechanisms adopted under EU law to address them. The proposed solutions are principled and well considered. In short, I really enjoyed this paper and found its arguments appealing. My comments below should be read in this light.
First, I was not fully convinced by the criticism of the use of qualitative criteria. The author’s main criticisms seem to swing between the criteria’s lack of practicality and their lack of reference to consumer welfare. The latter criticism seems to overlook the goal of these criteria, which is to provide a threshold that must be met before one is able to determine the competitive effects of business conduct. As with the concept of ‘agreement’, the goal is precisely to set a threshold that must be met before one looks at the impact of business conduct on consumer welfare. For this criticism to make sense, one would also have to challenge the need to find an agreement in order to establish collusion because this concept does not look into the effects of the parties’ arrangement – which the author does not do, and with good reason. Like the concept of ‘agreement’, qualitative criteria do not look at consumer welfare. Instead, they merely seek to approximate the concept of ‘agreement’, thereby making enforcement against collusive practices coherent.
On the other hand, the criticism against the practicality of the qualitative criteria may be justified in each specific case, but one must accept that proxies are by definition imperfect and reflect difficulties in coming up with better (more practical) criteria. Developing improved proxies is not easy feat in this area, as is apparent from similar difficulties in the US in distinguishing between lawful and unlawful tacit coordination.
Of course, the author’s ultimate argument is that we should just remove the distinction between lawful and unlawful tacit collusion, and sanction all types of tacit coordination that are welfare decreasing. This approach poses a challenge that goes to the heart of competition law. Typically, only once we find that there is an agreement or concerted practice would we look at their impact on consumer welfare. Without a dominant position, unilateral behaviour is not the subject of competition law, and tacit collusion is, absent evidence of some ‘common understanding’, merely a sum of unilateral actions that happen to lead (rationally) to supracompetitive prices. It would have been good for the author to explain why tacit collusion would merit an effects’ treatment across the board when otherwise we require some thresholds to be met before we pursue such an exercise. On the other hand, it may be argued that this conflation of ‘coordination’ and ‘anticompetitive effects’ is not unheard of – it is how information exchanges are dealt with in practice, and one could also think of some hub-and-spoke investigations. But there is a clear rationale for this conflation in each of these cases, so one would still need that rationale to be developed for ‘tacit collusion’.
More importantly, I think that, even under the terms of the author’s argument, the proposed solution is bound to be very onerous, and likely less practical than the qualitative criteria that the author decries. The fact that the author focuses on prohibiting specific types of ‘information signal’ entirely is unlikely to simplify enforcement as much as the author believes. The question would turn into how to identify each type of signal in practice, or how to evaluate conduct that does not fall squarely within the category of ‘anticompetitive information signal’, as is bound to happen as firms adapt to this new approach. To address these questions, either we fall back on qualitative criteria or we pursue an effects’ analysis in each case. The author dismisses qualitative criteria and says that we can do effects’ analysis often, and calibrate it by reference to burdens of proof. However, this overlooks the fact that the categorisation of business conduct as a concerted practice acts not only acts theoretically as a thresholds to effects analysis (i.e. it simplifies the analysis), but is, in practice, often a prelude to a finding of infringement by object. In other words, the goal of efforts to distinguish between concerted practices and lawful tacit collusion is to assign liability while avoiding an effects analysis.
Lastly, while I think I understand the author’s concerns underpinning this paper – the fact that, under machine learning and algorithms, existing rules on collusive conduct are likely to be ineffective – I am not sure that expanding the concept of collusion is the way to address them. As noted in most papers on algorithmic collusion, the problem is that, absent human intervention, there is no ‘collusion’ and existing competition instruments are unlikely to work. As with the discussions on killer mergers and common ownership, I am sympathetic to the concerns and agree that one most deal with issues that arise with the tools available to us – but this should not blind us to the fact that such issues may require action outside the narrow confines of existing antitrust law.
As a final comment, I note that this paper provides a useful bridge to discussions on algorithms, which I will engage with next week.